ArthurZ HF Staff commited on
Commit
a58b9e7
·
verified ·
1 Parent(s): 6f77350

Per-PR dots with hover cards, model-additions row, release markers, owner/churn columns, torch.export track

Browse files
Files changed (7) hide show
  1. README.md +13 -1
  2. build.py +54 -8
  3. data.js +0 -0
  4. index.html +263 -169
  5. models.json +1 -0
  6. pr_spans.json +1280 -266
  7. releases.json +1 -0
README.md CHANGED
@@ -7,7 +7,7 @@ sdk: static
7
  app_file: index.html
8
  pinned: false
9
  license: apache-2.0
10
- short_description: One roadmap timeline, measured from real PR commit dates
11
  ---
12
 
13
  # Transformers roadmap — timeline
@@ -23,6 +23,16 @@ months; the framework removal ran a year and a half; the modular detector was a
23
  Every item carries a **big impact tag** — what the block actually bought you (`vLLM imports it`,
24
  `enables native FSDP2 + TP`, `57 rules · v0.1.4`, `3 frameworks → 1`) rather than just a task name.
25
 
 
 
 
 
 
 
 
 
 
 
26
  ## Where the data comes from
27
 
28
  | | |
@@ -31,6 +41,8 @@ Every item carries a **big impact tag** — what the block actually bought you (
31
  | PR numbers, titles, states, commit dates | GitHub GraphQL API + `git log` on `main` (every merged commit carries its `(#NNNNN)`) |
32
  | `v5.0.0` date | Git tags — `v5.0.0rc0` 2025-12-01, `v5.0.0` 2026-01-26 |
33
  | Atlas / CI dashboard dates | HF Hub Space `created_at` / `last_modified` |
 
 
34
  | `transformers-to-mlx` dates | GitHub repo commit history |
35
 
36
  ## Board items shown as what actually happened
 
7
  app_file: index.html
8
  pinned: false
9
  license: apache-2.0
10
+ short_description: One roadmap timeline; every dot is a PR you can hover
11
  ---
12
 
13
  # Transformers roadmap — timeline
 
23
  Every item carries a **big impact tag** — what the block actually bought you (`vLLM imports it`,
24
  `enables native FSDP2 + TP`, `57 rules · v0.1.4`, `3 frameworks → 1`) rather than just a task name.
25
 
26
+ ## What's on it
27
+
28
+ - **Every dot is a PR.** Hover for a mini pull-request card — title, author, state, description, commits, files, ± lines.
29
+ - **Bars that fade out are still running**; solid bars have shipped.
30
+ - **A `Model additions` row** with one dot per new model, detected from the first commit that creates each
31
+ `src/transformers/models/<name>/` directory — 267 models from 126 authors since June 2024.
32
+ - **Releases along the top**: every `v4.x`/`v5.x` tag as a tick, with `v5.0.0rc0` and `v5.0.0` called out.
33
+ - **Owners on the left** (top contributors per item, from PR authorship), **churn on the right**
34
+ (PR count, ± lines, files, commits).
35
+
36
  ## Where the data comes from
37
 
38
  | | |
 
41
  | PR numbers, titles, states, commit dates | GitHub GraphQL API + `git log` on `main` (every merged commit carries its `(#NNNNN)`) |
42
  | `v5.0.0` date | Git tags — `v5.0.0rc0` 2025-12-01, `v5.0.0` 2026-01-26 |
43
  | Atlas / CI dashboard dates | HF Hub Space `created_at` / `last_modified` |
44
+ | Model additions | First commit creating each `src/transformers/models/<name>/` directory |
45
+ | Releases | `git tag` (`v4.42.0` → `v5.16.1`) |
46
  | `transformers-to-mlx` dates | GitHub repo commit history |
47
 
48
  ## Board items shown as what actually happened
build.py CHANGED
@@ -17,11 +17,27 @@ def span(prs):
17
  return [min(f), max(l)]
18
 
19
  def prmeta(prs):
 
20
  out = []
21
  for n in prs:
22
  v = SP.get(str(n))
23
- if v: out.append([n, v["merged"] or v["last"], v["t"], v["state"]])
24
- return sorted(out, key=lambda r: r[1])
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
25
 
26
  # ---------------------------------------------------------------- tracks
27
  T = []
@@ -135,13 +151,31 @@ track("Training & the 5 parallelisms", 48,
135
  note="No PRs found in transformers as of Sep 2026."),
136
  ])
137
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
138
  track("Quantization", 350, "Simplify, expand to other libraries, quantization by default.", [
139
- it("Quantization on the dynamic loader", "quant via WeightConverter", "done", [42254, 42634],
140
- d="bnb integration cleaned up to go through the weight converter — the “simplified quantization” item."),
 
141
  it("Kernel optimization for quant methods", "FP8 / MXFP4 via `kernels`", "done", [45634, 41897, 47863]),
142
  it("`transformers serve` + quantization", "quantized serving", "done", [41253]),
143
- it("MacOS / GGUF loading", "GGUF checkpoints load", "done", [38645, 39638, 41111],
144
- note="GGUF side shipped; the MLX-kernels half of “MacOS first-class” is covered by transformers-to-mlx, not by in-repo kernels."),
 
 
145
  it("Quantization-aware training", "partial", "wip", [41897],
146
  note="MXFP8/MXFP4 backwards landed; QAT / QAD as scoped on the board did not."),
147
  ])
@@ -177,6 +211,7 @@ for tr in T:
177
  if i.get("until"):
178
  sp = [sp[0], max(sp[1], i["until"])] if sp else [i["until"], i["until"]]
179
  i["span"] = sp
 
180
  i.pop("until", None); i.pop("at", None)
181
  i["prs"] = prmeta(i["prs"])
182
 
@@ -184,12 +219,23 @@ allspans = [i["span"] for tr in T for i in tr["items"] if i["span"]]
184
  lo = min(s[0] for s in allspans); hi = max(s[1] for s in allspans)
185
  print("span range:", lo, "→", hi, file=sys.stderr)
186
  n_items = sum(len(t["items"]) for t in T)
187
- n_prs = len({p[0] for tr in T for i in tr["items"] for p in i["prs"]})
188
  print(f"tracks {len(T)} items {n_items} prs {n_prs}", file=sys.stderr)
189
 
190
  hdr = ("// Generated by build.py — do not hand-edit.\n"
191
  "// Item spans are REAL: first commit of the earliest PR -> last commit / merge of the latest PR.\n"
192
  "// Source board: https://miro.com/app/board/uXjVL_vDn1k=/\n\n")
193
- out = hdr + "const AXIS = " + json.dumps([lo, hi]) + ";\n\nconst TRACKS = " + json.dumps(T, indent=1, ensure_ascii=False) + ";\n"
 
 
 
 
 
 
 
 
 
 
 
194
  open("/Users/arthurzucker/Work/transformers-roadmap-space/data.js", "w").write(out)
195
  print("wrote data.js", len(out), "bytes", file=sys.stderr)
 
17
  return [min(f), max(l)]
18
 
19
  def prmeta(prs):
20
+ """full record per PR: number, dot date, title, state, author, stats, body"""
21
  out = []
22
  for n in prs:
23
  v = SP.get(str(n))
24
+ if not v: continue
25
+ out.append({"n": n, "d": v["merged"] or v["last"], "t": v["t"], "st": v["state"],
26
+ "by": v.get("by",""), "a": v.get("add",0), "r": v.get("del",0),
27
+ "f": v.get("files",0), "c": v.get("commits",0), "b": v.get("body","")})
28
+ return sorted(out, key=lambda r: r["d"])
29
+
30
+ def rollup(prs):
31
+ """owners + totals for the flanking stat columns"""
32
+ from collections import Counter
33
+ who = Counter(); a=r=f=c=0
34
+ for n in prs:
35
+ v = SP.get(str(n))
36
+ if not v: continue
37
+ if v.get("by"): who[v["by"]] += 1
38
+ a+=v.get("add",0); r+=v.get("del",0); f+=v.get("files",0); c+=v.get("commits",0)
39
+ return {"owners":[[k,v] for k,v in who.most_common(4)], "n":len(prs),
40
+ "add":a, "del":r, "files":f, "commits":c}
41
 
42
  # ---------------------------------------------------------------- tracks
43
  T = []
 
151
  note="No PRs found in transformers as of Sep 2026."),
152
  ])
153
 
154
+ track("Export & on-device", 190,
155
+ "Get a transformers model out of Python and onto a device.", [
156
+ it("torch.export-compatible modeling", "models export cleanly", "done",
157
+ [32168, 33707, 35124, 37124, 37728, 39836, 47871, 47096],
158
+ d="Static cache made export-safe, generate from an exported model, then a common `torch.export` test so every new model stays exportable — plus a long tail of per-model fixes."),
159
+ it("`transformers.exporters`", "one export entry point", "done",
160
+ [41992, 47711, 47243, 47900],
161
+ d="The HF exporters module: a single supported path out of transformers instead of per-model export scripts."),
162
+ it("ExecuTorch integration", "on-device runtime", "done",
163
+ [34101, 34102, 36486, 38646, 41627, 47201],
164
+ note="ExecuTorch was pulled from CI in #47242 until the latest version is supported — the integration ships, the CI coverage is currently paused.",
165
+ d="Llama, Qwen2.5, T5 encoder-decoder and the encoder models exported to ExecuTorch; xnnpack backend in the exporter tests.",
166
+ links=[["ExecuTorch docs", "https://huggingface.co/docs/transformers/community_integrations/executorch"]]),
167
+ ])
168
+
169
  track("Quantization", 350, "Simplify, expand to other libraries, quantization by default.", [
170
+ it("Quantization on the dynamic loader", "quant via WeightConverter", "done",
171
+ [41138, 42043, 42254, 42426, 42634, 44576, 46325],
172
+ d="Quantizers made “good citizens loading-wise”, bnb re-plumbed through the weight converter, custom conversion mappings, async loading disabled when quantizing on the fly, TorchAo safetensors params."),
173
  it("Kernel optimization for quant methods", "FP8 / MXFP4 via `kernels`", "done", [45634, 41897, 47863]),
174
  it("`transformers serve` + quantization", "quantized serving", "done", [41253]),
175
+ it("MacOS first-class: GGUF + Metal kernels", "Metal dequant via kernels-community", "open",
176
+ [38645, 39638, 41111, 44348, 45975, 46357, 46792, 47779],
177
+ d="GGUF checkpoint loading shipped; the Metal/MPS kernel half is in review — optional Metal dequant fast path, MetalConfig for pre-quantized MLX checkpoints from the Hub, MPS paged-decode fast path for continuous batching, and a GGUF inference-speed refactor.",
178
+ note="Merged loading PRs plus four open kernel PRs — the bar runs to the latest commit on the open ones."),
179
  it("Quantization-aware training", "partial", "wip", [41897],
180
  note="MXFP8/MXFP4 backwards landed; QAT / QAD as scoped on the board did not."),
181
  ])
 
211
  if i.get("until"):
212
  sp = [sp[0], max(sp[1], i["until"])] if sp else [i["until"], i["until"]]
213
  i["span"] = sp
214
+ i["roll"] = rollup(i["prs"])
215
  i.pop("until", None); i.pop("at", None)
216
  i["prs"] = prmeta(i["prs"])
217
 
 
219
  lo = min(s[0] for s in allspans); hi = max(s[1] for s in allspans)
220
  print("span range:", lo, "→", hi, file=sys.stderr)
221
  n_items = sum(len(t["items"]) for t in T)
222
+ n_prs = len({p["n"] for tr in T for i in tr["items"] for p in i["prs"]})
223
  print(f"tracks {len(T)} items {n_items} prs {n_prs}", file=sys.stderr)
224
 
225
  hdr = ("// Generated by build.py — do not hand-edit.\n"
226
  "// Item spans are REAL: first commit of the earliest PR -> last commit / merge of the latest PR.\n"
227
  "// Source board: https://miro.com/app/board/uXjVL_vDn1k=/\n\n")
228
+ MODELS = json.load(open("models.json"))
229
+ RELEASES = json.load(open("releases.json"))
230
+ lo = min(lo, MODELS[0]["d"]); hi = max(hi, MODELS[-1]["d"])
231
+ from collections import Counter
232
+ mw = Counter(m["by"] for m in MODELS)
233
+ MODEL_ROW = {"n": len(MODELS), "span": [MODELS[0]["d"], MODELS[-1]["d"]],
234
+ "owners": [[k,v] for k,v in mw.most_common(4)], "authors": len(mw)}
235
+ out = (hdr + "const AXIS = " + json.dumps([lo, hi]) + ";\n"
236
+ + "const RELEASES = " + json.dumps(RELEASES) + ";\n"
237
+ + "const MODELS = " + json.dumps(MODELS, ensure_ascii=False) + ";\n"
238
+ + "const MODEL_ROW = " + json.dumps(MODEL_ROW, ensure_ascii=False) + ";\n\n"
239
+ + "const TRACKS = " + json.dumps(T, indent=1, ensure_ascii=False) + ";\n")
240
  open("/Users/arthurzucker/Work/transformers-roadmap-space/data.js", "w").write(out)
241
  print("wrote data.js", len(out), "bytes", file=sys.stderr)
data.js CHANGED
The diff for this file is too large to render. See raw diff
 
index.html CHANGED
@@ -7,14 +7,14 @@
7
  <style>
8
  :root{
9
  --bg:#fbfbfa; --panel:#fff; --ink:#1b1b1a; --dim:#6f6f69; --faint:#9a9a93; --line:#e7e6e2;
10
- --today:#e8590c; --done:#1a9c4b; --wip:#c07800; --open:#6a5cd6; --plan:#9a9a93;
11
  --shadow:0 1px 2px rgba(0,0,0,.05), 0 10px 30px -22px rgba(0,0,0,.4);
12
- --lbl:390px;
13
  }
14
  @media (prefers-color-scheme: dark){
15
  :root:not([data-theme=light]){
16
  --bg:#121213; --panel:#191a1b; --ink:#eceae5; --dim:#a09e95; --faint:#75736c; --line:#2c2d2f;
17
- --today:#ff8f4d; --done:#3ad07d; --wip:#e0a341; --open:#a48bf0; --plan:#75736c;
18
  --shadow:0 1px 2px rgba(0,0,0,.4), 0 10px 30px -22px #000;
19
  }
20
  }
@@ -24,271 +24,365 @@
24
  -webkit-font-smoothing:antialiased}
25
  code{font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:.92em}
26
  a{color:inherit}
27
-
28
- .page{max-width:1640px;margin:0 auto;padding:34px 24px 90px}
29
  h1{margin:0;font-size:28px;letter-spacing:-.022em;font-weight:680}
30
- .sub{color:var(--dim);margin:8px 0 0;max-width:80ch}
31
  .src{margin-top:13px;display:flex;flex-wrap:wrap;gap:7px;align-items:center}
32
  .pill{display:inline-flex;align-items:center;gap:6px;border:1px solid var(--line);background:var(--panel);
33
  border-radius:999px;padding:4px 11px;font-size:12px;color:var(--dim);text-decoration:none}
34
  a.pill:hover{color:var(--ink);border-color:var(--dim)}
35
 
36
- .stats{display:flex;gap:9px;flex-wrap:wrap;margin:20px 0 18px}
37
  .stat{border:1px solid var(--line);background:var(--panel);border-radius:11px;padding:9px 14px;box-shadow:var(--shadow)}
38
  .stat b{display:block;font-size:19px;letter-spacing:-.01em}
39
  .stat span{font-size:11px;color:var(--dim);text-transform:uppercase;letter-spacing:.055em}
40
-
41
- .filters{display:flex;gap:6px;flex-wrap:wrap;margin:0 0 14px;align-items:center}
42
  .f{border:1px solid var(--line);background:var(--panel);color:var(--dim);border-radius:8px;
43
  padding:5px 11px;font-size:12px;font-weight:560;cursor:pointer}
44
  .f[aria-pressed=true]{color:var(--ink);border-color:var(--ink)}
45
  .f:hover{color:var(--ink)}
46
 
47
  .scroll{overflow-x:auto;border:1px solid var(--line);border-radius:14px;background:var(--panel);box-shadow:var(--shadow)}
48
- .grid{min-width:1180px;position:relative}
49
-
50
- .axis{position:sticky;top:0;z-index:6;display:grid;grid-template-columns:var(--lbl) 1fr;
51
  background:var(--panel);border-bottom:1px solid var(--line)}
52
- .axis .corner{padding:9px 14px;font-size:11px;color:var(--faint);align-self:end;letter-spacing:.04em;text-transform:uppercase}
53
- .axl{position:relative;height:46px}
54
- .qt{position:absolute;top:20px;font-size:10.5px;color:var(--faint);padding-left:5px;
55
- border-left:1px solid var(--line);height:26px;white-space:nowrap}
 
56
  .qt.y{color:var(--ink);font-weight:680;border-left:2px solid var(--dim);font-size:11.5px}
57
- .todaylab{position:absolute;top:2px;z-index:8;transform:translateX(-50%);
 
 
 
 
58
  background:var(--today);color:#fff;font-size:9.5px;font-weight:800;letter-spacing:.07em;
59
  padding:2px 7px;border-radius:999px;white-space:nowrap}
60
 
61
  .track{border-top:1px solid var(--line)}
62
- .thead{display:flex;gap:11px;align-items:baseline;flex-wrap:wrap;padding:11px 14px 10px;
63
  background:hsl(var(--h) 60% 96.5%);border-left:4px solid hsl(var(--h) 50% 56%)}
64
  @media (prefers-color-scheme: dark){:root:not([data-theme=light]) .thead{background:hsl(var(--h) 26% 13%)}}
65
  .thead h2{margin:0;font-size:14.5px;font-weight:700;letter-spacing:-.012em}
66
  .thead .blurb{font-size:12.5px;color:var(--dim);font-style:italic}
67
  .thead .cnt{font-size:11px;color:var(--faint);margin-left:auto}
68
 
69
- .row{display:grid;grid-template-columns:var(--lbl) 1fr;border-top:1px solid var(--line);align-items:stretch}
70
  .row:hover{background:hsl(var(--h) 60% 97.6%)}
71
  @media (prefers-color-scheme: dark){:root:not([data-theme=light]) .row:hover{background:#1f2022}}
72
  .row.hide{display:none}
73
 
74
- .lbl{padding:9px 14px 10px;min-width:0}
75
  .name{display:flex;gap:7px;align-items:flex-start;line-height:1.35}
76
  .dot{flex:none;width:8px;height:8px;border-radius:50%;margin-top:5px;background:var(--plan)}
77
  .s-done .dot{background:var(--done)} .s-wip .dot{background:var(--wip)} .s-open .dot{background:var(--open)}
78
  .ttl{font-size:13px;font-weight:640;letter-spacing:-.005em}
79
-
80
- /* the big impact tag */
81
- .tag{display:inline-block;margin:7px 0 0 15px;font-size:12px;font-weight:700;letter-spacing:-.005em;
82
- padding:3px 10px;border-radius:7px;line-height:1.3;
83
- background:hsl(var(--h) 55% 90%);color:hsl(var(--h) 65% 24%);border:1px solid hsl(var(--h) 45% 78%)}
84
  @media (prefers-color-scheme: dark){:root:not([data-theme=light]) .tag{
85
  background:hsl(var(--h) 40% 20%);color:hsl(var(--h) 70% 82%);border-color:hsl(var(--h) 35% 32%)}}
86
  .s-planned .tag{background:transparent;color:var(--faint);border-style:dashed;border-color:var(--line);font-weight:600}
87
-
88
- .desc{font-size:11.7px;color:var(--dim);margin:6px 0 0 15px;line-height:1.5}
89
- .note{font-size:11.4px;color:var(--faint);margin:5px 0 0 15px;line-height:1.5;font-style:italic}
90
- .chips{display:flex;flex-wrap:wrap;gap:4px;margin:7px 0 0 15px}
91
- .pr{font-family:ui-monospace,SFMono-Regular,Menlo,monospace;font-size:10.5px;text-decoration:none;
92
- border:1px solid var(--line);border-radius:6px;padding:1px 6px;color:var(--dim);background:var(--bg);white-space:nowrap}
93
- .pr:hover{color:var(--ink);border-color:var(--done)}
94
- .pr.open{border-color:var(--open);color:var(--open)}
95
  .ext{font-size:10.5px;text-decoration:none;border:1px solid var(--line);border-radius:6px;
96
- padding:1px 7px;color:var(--dim);background:var(--bg);white-space:nowrap}
97
  .ext:hover{color:var(--ink);border-color:var(--dim)}
98
  .ext::before{content:"↗ "}
99
 
100
- .lane{position:relative;min-height:40px}
101
  .gl{position:absolute;top:0;bottom:0;width:0;border-left:1px solid var(--line)}
102
  .gl.y{border-left:2px solid var(--line)}
103
- .bar{position:absolute;top:50%;transform:translateY(-50%);height:15px;border-radius:8px;min-width:5px;
104
- background:hsl(var(--h) 52% 58% / .40);border:1px solid hsl(var(--h) 46% 46% / .6)}
105
- .s-done .bar{background:hsl(var(--h) 52% 55% / .55)}
106
- .s-wip .bar{background:repeating-linear-gradient(115deg,hsl(var(--h) 52% 58% / .55) 0 6px,transparent 6px 12px);
107
- border-color:var(--wip)}
108
- .s-open .bar{background:repeating-linear-gradient(115deg,hsl(var(--h) 52% 58% / .5) 0 5px,transparent 5px 10px);
109
- border-color:var(--open)}
 
 
 
 
 
 
 
110
  .dur{position:absolute;top:50%;transform:translateY(-50%);font-size:10px;color:var(--faint);
111
- white-space:nowrap;padding-left:7px;font-variant-numeric:tabular-nums}
112
  .none{position:absolute;top:50%;transform:translateY(-50%);font-size:10.5px;color:var(--faint);
113
  font-style:italic;padding-left:8px}
114
  .todayline{position:absolute;top:0;bottom:0;width:0;border-left:2px solid var(--today);z-index:4;pointer-events:none}
115
 
116
- .legend{display:flex;gap:18px;flex-wrap:wrap;margin:15px 3px 0;font-size:11.5px;color:var(--dim)}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
117
  .legend i{font-style:normal;display:inline-flex;gap:6px;align-items:center}
118
  .key{width:9px;height:9px;border-radius:50%;display:inline-block}
119
- footer{margin-top:36px;color:var(--dim);font-size:12px;max-width:96ch;line-height:1.65}
120
- footer h3{font-size:13px;margin:20px 0 6px;color:var(--ink)}
121
  footer p{margin:6px 0}
122
  </style>
123
  </head>
124
  <body>
125
  <div class="page">
126
-
127
  <h1>Transformers roadmap — timeline</h1>
128
  <p class="sub">The <b>huggingface/transformers</b> roadmap board, fused into one timeline and grounded in git.
129
- Every bar is measured from real commits: it starts at the <b>first commit of the earliest PR</b> and ends at the
130
- <b>last commit of the latest PR</b> — so the length is how long the work actually took, not where a sticky note sat.</p>
 
131
  <div class="src">
132
  <a class="pill" href="https://miro.com/app/board/uXjVL_vDn1k=/" target="_blank" rel="noopener">◎ Source board</a>
133
  <a class="pill" href="https://github.com/huggingface/transformers" target="_blank" rel="noopener">⌥ huggingface/transformers</a>
134
  <span class="pill" id="meta"></span>
135
  </div>
136
-
137
  <div class="stats" id="stats"></div>
138
  <div class="filters" id="filters"></div>
139
  <div class="scroll"><div class="grid" id="grid"></div></div>
140
  <div class="legend" id="legend"></div>
141
-
142
  <footer>
143
  <h3>How to read it</h3>
144
- <p>A bar's <b>left edge</b> is the first commit anyone pushed toward that item; its <b>right edge</b> is the last
145
- commit (or merge) of the last PR. A long bar means a long grind — the DTensor TP work ran two months, the
146
- framework removal ran over a year. Solid bars shipped; hatched bars are still moving; dashed outlines never started.</p>
 
147
  <h3>Where the data comes from</h3>
148
- <p>Item text and grouping come from the Miro board. Everything else is measured: PR numbers, titles, states and
149
- commit dates come from the GitHub API and <code>git log</code> on <code>main</code>, where every merged commit
150
- carries its <code>(#NNNNN)</code>.</p>
151
- <p>Board items that turned out to be superseded are shown as what actually happened, not as drawn — “coding rules to
152
- ruff linter” became <b>mlinter</b>, its own AST-based package with 57 rules; “visual modular” became the
153
- <b>Atlas</b> architecture gallery plus the open modular-v2 PR; “monkey patching API” shipped as the
154
- <b>model patching API</b>.</p>
155
- <p>Items marked <i>not started</i> are ones where no PRs could be found in this repo as of September 2026 — they may
156
- be tracked elsewhere. Two items shipped in other repos by design: the <b>TGI backend</b> and the
157
- <b>transformers→MLX</b> conversion skill.</p>
158
  </footer>
159
  </div>
 
160
 
161
  <script src="data.js"></script>
162
  <script>
163
- const MON = ["Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"];
164
- const D = s => new Date(s + "T00:00:00Z");
165
- const NOW = new Date();
 
 
 
 
 
 
166
 
167
- // axis: cover every span plus today, with a little breathing room
168
- const A0 = D(AXIS[0]);
169
- const A1raw = Math.max(D(AXIS[1]).getTime(), NOW.getTime());
170
- const A1 = new Date(A1raw + 45 * 864e5);
171
- const SPAN = A1 - A0;
172
- const pct = d => ((d instanceof Date ? d : D(d)) - A0) / SPAN * 100;
173
-
174
- function el(t, c, x){ const n = document.createElement(t); if(c) n.className = c; if(x!=null) n.textContent = x; return n; }
175
- function dur(a, b){
176
- const days = Math.round((D(b) - D(a)) / 864e5);
177
- if (days < 14) return days + "d";
178
- if (days < 70) return Math.round(days/7) + "w";
179
- const m = days / 30.44;
180
- return m < 24 ? Math.round(m) + "mo" : (m/12).toFixed(1) + "y";
 
 
 
 
 
 
 
 
 
 
181
  }
 
 
 
 
 
 
 
 
 
182
 
183
  // ---------- stats ----------
184
- const ITEMS = TRACKS.flatMap(t => t.items);
185
- const PRS = new Set(ITEMS.flatMap(i => i.prs.map(p => p[0])));
186
- const shipped = ITEMS.filter(i => i.s === "done").length;
187
- const S = document.getElementById("stats");
188
- [[ITEMS.length,"roadmap items"], [shipped,"shipped"], [ITEMS.filter(i=>i.s==="wip"||i.s==="open").length,"in flight"],
189
- [PRS.size,"linked PRs"], [dur(AXIS[0], AXIS[1]),"of elapsed work"]]
190
- .forEach(([b,s]) => { const d = el("div","stat"); d.append(el("b",null,b), el("span",null,s)); S.append(d); });
191
- document.getElementById("meta").textContent = PRS.size + " PRs · " + AXIS[0] + " → " + AXIS[1];
 
 
192
 
193
  // ---------- filters ----------
194
- const FILTERS = [["all","Everything"],["done","Shipped"],["wip","In progress"],["open","In review"],["planned","Not started"]];
195
- let active = "all";
196
- const F = document.getElementById("filters");
197
- FILTERS.forEach(([k,label]) => {
198
- const b = el("button","f",label); b.setAttribute("aria-pressed", k === "all");
199
- b.onclick = () => { active = k; [...F.children].forEach((c,i) => c.setAttribute("aria-pressed", FILTERS[i][0] === k)); apply(); };
200
- F.append(b);
201
- });
202
  function apply(){
203
- document.querySelectorAll(".row").forEach(r =>
204
- r.classList.toggle("hide", active !== "all" && r.dataset.s !== active));
205
- document.querySelectorAll(".track").forEach(t => {
206
- const vis = [...t.querySelectorAll(".row")].filter(r => !r.classList.contains("hide")).length;
207
- t.style.display = vis ? "" : "none";
208
- const c = t.querySelector(".cnt"); if (c) c.textContent = vis + (vis === 1 ? " item" : " items");
209
- });
210
  }
211
 
212
- // ---------- gridlines / axis ----------
213
- function ticks(){
214
- const out = [];
215
- const d = new Date(Date.UTC(A0.getUTCFullYear(), Math.floor(A0.getUTCMonth()/3)*3, 1));
216
- while (d < A1){
217
- if (d >= A0) out.push({ p: pct(d), y: d.getUTCMonth() === 0, lab:
218
- d.getUTCMonth() === 0 ? String(d.getUTCFullYear()) : MON[d.getUTCMonth()] });
219
- d.setUTCMonth(d.getUTCMonth() + 3);
220
- }
221
- return out;
222
- }
223
- const TICKS = ticks(), TP = pct(NOW);
 
224
 
225
- const axis = el("div","axis");
226
- axis.append(el("div","corner","Item / impact"));
227
- const axl = el("div","axl");
228
- TICKS.forEach(t => { const q = el("div","qt" + (t.y ? " y" : ""), t.lab); q.style.left = t.p + "%"; axl.append(q); });
229
- const tl = el("div","todaylab","TODAY"); tl.style.left = TP + "%"; axl.append(tl);
230
- axis.append(axl);
231
- const G = document.getElementById("grid");
232
- G.append(axis);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
233
 
234
  // ---------- rows ----------
235
- for (const tr of TRACKS){
236
- const sec = el("section","track"); sec.style.setProperty("--h", tr.hue);
237
- const th = el("div","thead");
238
- th.append(el("h2",null,tr.name), el("span","blurb",tr.blurb), el("span","cnt", tr.items.length + " items"));
239
  sec.append(th);
240
-
241
- for (const i of tr.items){
242
- const row = el("div","row s-" + i.s); row.dataset.s = i.s;
243
-
244
- const L = el("div","lbl");
245
- const nm = el("div","name"); nm.append(el("span","dot"), el("span","ttl", i.t)); L.append(nm);
246
- if (i.tag) L.append(el("div", null, "").appendChild(el("span","tag", i.tag)).parentNode);
247
- if (i.d) L.append(el("div","desc", i.d));
248
- if (i.note) L.append(el("div","note", i.note));
249
- if (i.prs.length || i.links.length){
250
- const c = el("div","chips");
251
- i.prs.forEach(([n,d,t,st]) => {
252
- const a = el("a","pr" + (st !== "MERGED" ? " open" : ""), "#" + n);
253
- a.href = "https://github.com/huggingface/transformers/pull/" + n;
254
- a.target = "_blank"; a.rel = "noopener";
255
- a.title = (st === "MERGED" ? "merged " : st.toLowerCase() + " · ") + d + " — " + t;
256
- c.append(a);
257
- });
258
- i.links.forEach(([label,url]) => {
259
- const a = el("a","ext",label); a.href = url; a.target = "_blank"; a.rel = "noopener"; c.append(a);
260
- });
261
- L.append(c);
262
  }
263
  row.append(L);
264
 
265
- const lane = el("div","lane");
266
- TICKS.forEach(t => { const g = el("div","gl" + (t.y ? " y" : "")); g.style.left = t.p + "%"; lane.append(g); });
267
- const tline = el("div","todayline"); tline.style.left = TP + "%"; lane.append(tline);
268
-
269
- if (i.span){
270
- const [a,b] = i.span, l = pct(a), w = Math.max(pct(b) - l, 0.25);
271
- const bar = el("div","bar"); bar.style.left = l + "%"; bar.style.width = w + "%";
272
- bar.title = a + " → " + b + " (" + dur(a,b) + ")";
273
- lane.append(bar);
274
- const dd = el("div","dur", dur(a,b)); dd.style.left = (l + w) + "%"; lane.append(dd);
275
  } else {
276
- const txt = i.s === "planned" ? "not started" : "shipped outside this repo";
277
- const n = el("div","none", txt); n.style.left = "0%"; lane.append(n);
278
  }
279
- row.append(lane);
 
 
 
 
 
 
 
 
 
 
280
  sec.append(row);
281
  }
282
  G.append(sec);
283
  }
284
 
285
- document.getElementById("legend").innerHTML =
286
- '<i><span class="key" style="background:var(--done)"></span> shipped</i>' +
287
- '<i><span class="key" style="background:var(--wip)"></span> in progress (hatched bar)</i>' +
288
- '<i><span class="key" style="background:var(--open)"></span> in review</i>' +
289
- '<i><span class="key" style="background:var(--plan)"></span> not started</i>' +
290
- '<i>bar length = first commit → last commit</i>' +
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
291
  '<i><span style="color:var(--today);font-weight:800">▌</span> today</i>';
 
292
  </script>
293
  </body>
294
  </html>
 
7
  <style>
8
  :root{
9
  --bg:#fbfbfa; --panel:#fff; --ink:#1b1b1a; --dim:#6f6f69; --faint:#9a9a93; --line:#e7e6e2;
10
+ --today:#e8590c; --done:#1a9c4b; --wip:#c07800; --open:#6a5cd6; --plan:#9a9a93; --rel:#3f6fd8;
11
  --shadow:0 1px 2px rgba(0,0,0,.05), 0 10px 30px -22px rgba(0,0,0,.4);
12
+ --lbl:340px; --rgt:132px;
13
  }
14
  @media (prefers-color-scheme: dark){
15
  :root:not([data-theme=light]){
16
  --bg:#121213; --panel:#191a1b; --ink:#eceae5; --dim:#a09e95; --faint:#75736c; --line:#2c2d2f;
17
+ --today:#ff8f4d; --done:#3ad07d; --wip:#e0a341; --open:#a48bf0; --plan:#75736c; --rel:#7fa6f5;
18
  --shadow:0 1px 2px rgba(0,0,0,.4), 0 10px 30px -22px #000;
19
  }
20
  }
 
24
  -webkit-font-smoothing:antialiased}
25
  code{font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace;font-size:.92em}
26
  a{color:inherit}
27
+ .page{max-width:1760px;margin:0 auto;padding:34px 22px 90px}
 
28
  h1{margin:0;font-size:28px;letter-spacing:-.022em;font-weight:680}
29
+ .sub{color:var(--dim);margin:8px 0 0;max-width:82ch}
30
  .src{margin-top:13px;display:flex;flex-wrap:wrap;gap:7px;align-items:center}
31
  .pill{display:inline-flex;align-items:center;gap:6px;border:1px solid var(--line);background:var(--panel);
32
  border-radius:999px;padding:4px 11px;font-size:12px;color:var(--dim);text-decoration:none}
33
  a.pill:hover{color:var(--ink);border-color:var(--dim)}
34
 
35
+ .stats{display:flex;gap:9px;flex-wrap:wrap;margin:20px 0 16px}
36
  .stat{border:1px solid var(--line);background:var(--panel);border-radius:11px;padding:9px 14px;box-shadow:var(--shadow)}
37
  .stat b{display:block;font-size:19px;letter-spacing:-.01em}
38
  .stat span{font-size:11px;color:var(--dim);text-transform:uppercase;letter-spacing:.055em}
39
+ .filters{display:flex;gap:6px;flex-wrap:wrap;margin:0 0 14px}
 
40
  .f{border:1px solid var(--line);background:var(--panel);color:var(--dim);border-radius:8px;
41
  padding:5px 11px;font-size:12px;font-weight:560;cursor:pointer}
42
  .f[aria-pressed=true]{color:var(--ink);border-color:var(--ink)}
43
  .f:hover{color:var(--ink)}
44
 
45
  .scroll{overflow-x:auto;border:1px solid var(--line);border-radius:14px;background:var(--panel);box-shadow:var(--shadow)}
46
+ .grid{min-width:1320px;position:relative}
47
+ .axis{position:sticky;top:0;z-index:6;display:grid;grid-template-columns:var(--lbl) 1fr var(--rgt);
 
48
  background:var(--panel);border-bottom:1px solid var(--line)}
49
+ .corner{padding:9px 13px;font-size:10px;color:var(--faint);align-self:end;letter-spacing:.05em;text-transform:uppercase}
50
+ .corner.r{text-align:right}
51
+ .axl{position:relative;height:66px}
52
+ .qt{position:absolute;top:19px;font-size:10.5px;color:var(--faint);padding-left:5px;
53
+ border-left:1px solid var(--line);height:22px;white-space:nowrap}
54
  .qt.y{color:var(--ink);font-weight:680;border-left:2px solid var(--dim);font-size:11.5px}
55
+ .rtick{position:absolute;bottom:5px;width:1px;height:9px;background:var(--rel);opacity:.55;cursor:default}
56
+ .rtick:hover{opacity:1;height:14px}
57
+ .rbig{position:absolute;bottom:4px;transform:translateX(-50%);font-size:9.5px;font-weight:800;
58
+ color:#fff;background:var(--rel);padding:1px 6px;border-radius:999px;white-space:nowrap;letter-spacing:.03em}
59
+ .todaylab{position:absolute;top:1px;z-index:8;transform:translateX(-50%);
60
  background:var(--today);color:#fff;font-size:9.5px;font-weight:800;letter-spacing:.07em;
61
  padding:2px 7px;border-radius:999px;white-space:nowrap}
62
 
63
  .track{border-top:1px solid var(--line)}
64
+ .thead{display:flex;gap:11px;align-items:baseline;flex-wrap:wrap;padding:11px 13px 10px;
65
  background:hsl(var(--h) 60% 96.5%);border-left:4px solid hsl(var(--h) 50% 56%)}
66
  @media (prefers-color-scheme: dark){:root:not([data-theme=light]) .thead{background:hsl(var(--h) 26% 13%)}}
67
  .thead h2{margin:0;font-size:14.5px;font-weight:700;letter-spacing:-.012em}
68
  .thead .blurb{font-size:12.5px;color:var(--dim);font-style:italic}
69
  .thead .cnt{font-size:11px;color:var(--faint);margin-left:auto}
70
 
71
+ .row{display:grid;grid-template-columns:var(--lbl) 1fr var(--rgt);border-top:1px solid var(--line)}
72
  .row:hover{background:hsl(var(--h) 60% 97.6%)}
73
  @media (prefers-color-scheme: dark){:root:not([data-theme=light]) .row:hover{background:#1f2022}}
74
  .row.hide{display:none}
75
 
76
+ .lbl{padding:9px 13px 10px;min-width:0}
77
  .name{display:flex;gap:7px;align-items:flex-start;line-height:1.35}
78
  .dot{flex:none;width:8px;height:8px;border-radius:50%;margin-top:5px;background:var(--plan)}
79
  .s-done .dot{background:var(--done)} .s-wip .dot{background:var(--wip)} .s-open .dot{background:var(--open)}
80
  .ttl{font-size:13px;font-weight:640;letter-spacing:-.005em}
81
+ .tag{display:inline-block;margin:7px 0 0 15px;font-size:12px;font-weight:700;padding:3px 10px;border-radius:7px;
82
+ line-height:1.3;background:hsl(var(--h) 55% 90%);color:hsl(var(--h) 65% 24%);border:1px solid hsl(var(--h) 45% 78%)}
 
 
 
83
  @media (prefers-color-scheme: dark){:root:not([data-theme=light]) .tag{
84
  background:hsl(var(--h) 40% 20%);color:hsl(var(--h) 70% 82%);border-color:hsl(var(--h) 35% 32%)}}
85
  .s-planned .tag{background:transparent;color:var(--faint);border-style:dashed;border-color:var(--line);font-weight:600}
86
+ .desc{font-size:11.6px;color:var(--dim);margin:6px 0 0 15px;line-height:1.5}
87
+ .note{font-size:11.3px;color:var(--faint);margin:5px 0 0 15px;line-height:1.5;font-style:italic}
88
+ .owners{display:flex;align-items:center;gap:4px;margin:8px 0 0 15px;flex-wrap:wrap}
89
+ .av{width:19px;height:19px;border-radius:50%;border:1px solid var(--line);background:var(--bg);display:block}
90
+ .owners .more{font-size:10.5px;color:var(--faint)}
 
 
 
91
  .ext{font-size:10.5px;text-decoration:none;border:1px solid var(--line);border-radius:6px;
92
+ padding:1px 7px;color:var(--dim);background:var(--bg);white-space:nowrap;margin-left:4px}
93
  .ext:hover{color:var(--ink);border-color:var(--dim)}
94
  .ext::before{content:"↗ "}
95
 
96
+ .lane{position:relative;min-height:46px}
97
  .gl{position:absolute;top:0;bottom:0;width:0;border-left:1px solid var(--line)}
98
  .gl.y{border-left:2px solid var(--line)}
99
+ .rl{position:absolute;top:0;bottom:0;width:0;border-left:1px dotted var(--rel);opacity:.4}
100
+ .rl.big{border-left:1.5px solid var(--rel);opacity:.75}
101
+ .bar{position:absolute;top:50%;transform:translateY(-50%);height:13px;border-radius:7px;min-width:5px;
102
+ background:hsl(var(--h) 52% 55% / .38);border:1px solid hsl(var(--h) 46% 46% / .5)}
103
+ .s-wip .bar,.s-open .bar{border-style:dashed;
104
+ -webkit-mask-image:linear-gradient(90deg,#000 0,#000 82%,transparent 100%);
105
+ mask-image:linear-gradient(90deg,#000 0,#000 82%,transparent 100%)}
106
+ .pdot{position:absolute;top:50%;width:9px;height:9px;border-radius:50%;
107
+ transform:translate(-50%,-50%);background:hsl(var(--h) 60% 42%);border:1.5px solid var(--panel);
108
+ cursor:pointer;z-index:3;transition:transform .08s}
109
+ .pdot:hover{transform:translate(-50%,-50%) scale(1.65);z-index:9}
110
+ .pdot.open{background:var(--open)}
111
+ .pdot.model{width:6px;height:6px;background:hsl(var(--h) 62% 46%);border-width:1px;opacity:.82}
112
+ .pdot.model:hover{opacity:1}
113
  .dur{position:absolute;top:50%;transform:translateY(-50%);font-size:10px;color:var(--faint);
114
+ white-space:nowrap;padding-left:9px;font-variant-numeric:tabular-nums}
115
  .none{position:absolute;top:50%;transform:translateY(-50%);font-size:10.5px;color:var(--faint);
116
  font-style:italic;padding-left:8px}
117
  .todayline{position:absolute;top:0;bottom:0;width:0;border-left:2px solid var(--today);z-index:4;pointer-events:none}
118
 
119
+ .rstat{padding:11px 13px 10px;text-align:right;border-left:1px solid var(--line);font-variant-numeric:tabular-nums}
120
+ .rstat .big{font-size:15px;font-weight:700;letter-spacing:-.01em}
121
+ .rstat .sm{font-size:10.5px;color:var(--faint);margin-top:2px;line-height:1.5}
122
+ .plus{color:var(--done)} .minus{color:#d1453b}
123
+ @media (prefers-color-scheme: dark){:root:not([data-theme=light]) .minus{color:#ff7a70}}
124
+
125
+ /* mini PR card */
126
+ #card{position:fixed;z-index:99;max-width:390px;background:var(--panel);border:1px solid var(--line);
127
+ border-radius:11px;box-shadow:0 10px 40px -10px rgba(0,0,0,.35);padding:11px 13px;display:none;pointer-events:none}
128
+ #card .h{display:flex;gap:7px;align-items:baseline;flex-wrap:wrap}
129
+ #card .num{font-family:ui-monospace,Menlo,monospace;font-size:11.5px;color:var(--dim)}
130
+ #card .st{font-size:9.5px;font-weight:800;letter-spacing:.05em;padding:1px 6px;border-radius:999px;
131
+ background:var(--done);color:#fff;text-transform:uppercase}
132
+ #card .st.open{background:var(--open)} #card .st.closed{background:var(--faint)}
133
+ #card .t{font-size:12.8px;font-weight:660;margin:5px 0 0;line-height:1.35}
134
+ #card .b{font-size:11.4px;color:var(--dim);margin:6px 0 0;line-height:1.5;
135
+ display:-webkit-box;-webkit-line-clamp:4;-webkit-box-orient:vertical;overflow:hidden}
136
+ #card .m{font-size:10.5px;color:var(--faint);margin-top:8px;display:flex;gap:9px;flex-wrap:wrap;
137
+ border-top:1px solid var(--line);padding-top:7px;font-variant-numeric:tabular-nums}
138
+
139
+ .legend{display:flex;gap:17px;flex-wrap:wrap;margin:15px 3px 0;font-size:11.5px;color:var(--dim)}
140
  .legend i{font-style:normal;display:inline-flex;gap:6px;align-items:center}
141
  .key{width:9px;height:9px;border-radius:50%;display:inline-block}
142
+ footer{margin-top:34px;color:var(--dim);font-size:12px;max-width:98ch;line-height:1.65}
143
+ footer h3{font-size:13px;margin:19px 0 6px;color:var(--ink)}
144
  footer p{margin:6px 0}
145
  </style>
146
  </head>
147
  <body>
148
  <div class="page">
 
149
  <h1>Transformers roadmap — timeline</h1>
150
  <p class="sub">The <b>huggingface/transformers</b> roadmap board, fused into one timeline and grounded in git.
151
+ Bars run from the <b>first commit of the earliest PR</b> to the <b>last commit of the latest</b>, so length is how
152
+ long the work really took. <b>Every dot is a PR</b> — hover it for the pull request. Owners sit on the left,
153
+ churn on the right, releases along the top.</p>
154
  <div class="src">
155
  <a class="pill" href="https://miro.com/app/board/uXjVL_vDn1k=/" target="_blank" rel="noopener">◎ Source board</a>
156
  <a class="pill" href="https://github.com/huggingface/transformers" target="_blank" rel="noopener">⌥ huggingface/transformers</a>
157
  <span class="pill" id="meta"></span>
158
  </div>
 
159
  <div class="stats" id="stats"></div>
160
  <div class="filters" id="filters"></div>
161
  <div class="scroll"><div class="grid" id="grid"></div></div>
162
  <div class="legend" id="legend"></div>
 
163
  <footer>
164
  <h3>How to read it</h3>
165
+ <p>A bar's left edge is the first commit anyone pushed toward that item; its right edge is the last commit or merge.
166
+ Solid bars shipped. <b>Bars that fade out on the right are still running</b> — the work has no end date yet.
167
+ Dots are the individual PRs; purple dots are still open. Blue ticks along the top are releases,
168
+ with <code>v5.0.0rc0</code> and <code>v5.0.0</code> called out.</p>
169
  <h3>Where the data comes from</h3>
170
+ <p>Item text and grouping come from the Miro board. Everything measured — PR numbers, titles, descriptions, authors,
171
+ states, commit dates, additions/deletions — comes from the GitHub GraphQL API and <code>git log</code> on
172
+ <code>main</code>. Model additions are detected from the first commit that creates each
173
+ <code>src/transformers/models/&lt;name&gt;/</code> directory, so the row is exact rather than label-dependent.</p>
174
+ <h3>Board items shown as what actually happened</h3>
175
+ <p>“Coding rules to ruff linter” became <b>mlinter</b> (TRF001–TRF057, its own package). “Visual modular” became the
176
+ <b>Atlas</b> gallery plus the open modular-v2 PR. “Monkey patching API” shipped as the <b>model patching API</b>.
177
+ “Explore transformers → mlx” shipped as <code>huggingface/transformers-to-mlx</code>. CircleCI is gone — the
178
+ config is a deliberate no-op. Items marked <i>not started</i> have no PRs found in this repo as of September 2026.</p>
 
179
  </footer>
180
  </div>
181
+ <div id="card"></div>
182
 
183
  <script src="data.js"></script>
184
  <script>
185
+ const MON=["Jan","Feb","Mar","Apr","May","Jun","Jul","Aug","Sep","Oct","Nov","Dec"];
186
+ const D=s=>new Date(s+"T00:00:00Z"), NOW=new Date();
187
+ const A0=D(AXIS[0]), A1=new Date(Math.max(D(AXIS[1]).getTime(),NOW.getTime())+45*864e5), SPAN=A1-A0;
188
+ const pct=d=>((d instanceof Date?d:D(d))-A0)/SPAN*100;
189
+ const el=(t,c,x)=>{const n=document.createElement(t);if(c)n.className=c;if(x!=null)n.textContent=x;return n;};
190
+ const nf=n=>n>=1000?(n/1000).toFixed(n>=10000?0:1)+"k":String(n);
191
+ function dur(a,b){const d=Math.round((D(b)-D(a))/864e5);
192
+ if(d<14)return d+"d"; if(d<70)return Math.round(d/7)+"w";
193
+ const m=d/30.44; return m<24?Math.round(m)+"mo":(m/12).toFixed(1)+"y";}
194
 
195
+ // ---------- mini PR card ----------
196
+ const card=document.getElementById("card");
197
+ function showCard(e,o){
198
+ card.innerHTML="";
199
+ const h=el("div","h");
200
+ h.append(el("span","num","#"+o.n));
201
+ const st=el("span","st"+(o.st==="OPEN"?" open":o.st==="CLOSED"?" closed":""),
202
+ o.st==="MERGED"?"merged":o.st.toLowerCase());
203
+ h.append(st);
204
+ if(o.by) h.append(el("span","num","@"+o.by));
205
+ card.append(h, el("div","t",o.t));
206
+ if(o.b) card.append(el("div","b",o.b));
207
+ const m=el("div","m");
208
+ m.append(el("span",null,o.d));
209
+ if(o.c) m.append(el("span",null,o.c+" commit"+(o.c>1?"s":"")));
210
+ if(o.f) m.append(el("span",null,o.f+" file"+(o.f>1?"s":"")));
211
+ if(o.a||o.r){ const s=el("span"); s.innerHTML='<span class="plus">+'+nf(o.a)+'</span> <span class="minus">−'+nf(o.r)+'</span>'; m.append(s); }
212
+ card.append(m);
213
+ card.style.display="block";
214
+ const r=e.target.getBoundingClientRect(), w=card.offsetWidth, hh=card.offsetHeight;
215
+ let x=r.left+r.width/2-w/2, y=r.top-hh-11;
216
+ x=Math.max(8,Math.min(x,innerWidth-w-8));
217
+ if(y<8) y=r.bottom+11;
218
+ card.style.left=x+"px"; card.style.top=y+"px";
219
  }
220
+ const hideCard=()=>card.style.display="none";
221
+
222
+ // ---------- ticks ----------
223
+ function ticks(){const out=[];const d=new Date(Date.UTC(A0.getUTCFullYear(),Math.floor(A0.getUTCMonth()/3)*3,1));
224
+ while(d<A1){ if(d>=A0) out.push({p:pct(d),y:d.getUTCMonth()===0,
225
+ lab:d.getUTCMonth()===0?String(d.getUTCFullYear()):MON[d.getUTCMonth()]});
226
+ d.setUTCMonth(d.getUTCMonth()+3);} return out;}
227
+ const TICKS=ticks(), TP=pct(NOW);
228
+ const RELS=RELEASES.filter(r=>D(r.d)>=A0);
229
 
230
  // ---------- stats ----------
231
+ const ITEMS=TRACKS.flatMap(t=>t.items);
232
+ const PRS=new Set(ITEMS.flatMap(i=>i.prs.map(p=>p.n)));
233
+ const totAdd=ITEMS.reduce((s,i)=>s+i.roll.add,0), totDel=ITEMS.reduce((s,i)=>s+i.roll.del,0);
234
+ const S=document.getElementById("stats");
235
+ [[ITEMS.length,"roadmap items"],[ITEMS.filter(i=>i.s==="done").length,"shipped"],
236
+ [ITEMS.filter(i=>i.s==="wip"||i.s==="open").length,"in flight"],
237
+ [PRS.size,"roadmap PRs"],[MODEL_ROW.n,"models added"],[MODEL_ROW.authors,"model authors"],
238
+ ["+"+nf(totAdd)+" / −"+nf(totDel),"lines changed"],[dur(AXIS[0],AXIS[1]),"elapsed"]]
239
+ .forEach(([b,s])=>{const d=el("div","stat");d.append(el("b",null,b),el("span",null,s));S.append(d);});
240
+ document.getElementById("meta").textContent=PRS.size+" PRs · "+MODEL_ROW.n+" models · "+AXIS[0]+" → "+AXIS[1];
241
 
242
  // ---------- filters ----------
243
+ const FL=[["all","Everything"],["done","Shipped"],["wip","In progress"],["open","In review"],["planned","Not started"]];
244
+ let active="all"; const F=document.getElementById("filters");
245
+ FL.forEach(([k,l])=>{const b=el("button","f",l);b.setAttribute("aria-pressed",k==="all");
246
+ b.onclick=()=>{active=k;[...F.children].forEach((c,i)=>c.setAttribute("aria-pressed",FL[i][0]===k));apply();};F.append(b);});
 
 
 
 
247
  function apply(){
248
+ document.querySelectorAll(".row").forEach(r=>r.classList.toggle("hide",active!=="all"&&r.dataset.s!==active));
249
+ document.querySelectorAll(".track").forEach(t=>{
250
+ const v=[...t.querySelectorAll(".row")].filter(r=>!r.classList.contains("hide")).length;
251
+ t.style.display=v?"":"none"; const c=t.querySelector(".cnt"); if(c)c.textContent=v+(v===1?" item":" items");});
 
 
 
252
  }
253
 
254
+ // ---------- axis ----------
255
+ const axis=el("div","axis");
256
+ axis.append(el("div","corner","Item · impact · owners"));
257
+ const axl=el("div","axl");
258
+ TICKS.forEach(t=>{const q=el("div","qt"+(t.y?" y":""),t.lab);q.style.left=t.p+"%";axl.append(q);});
259
+ RELS.forEach(r=>{
260
+ if(r.big){const b=el("div","rbig",r.n);b.style.left=pct(r.d)+"%";
261
+ b.style.bottom=(r.n.includes("rc")?18:4)+"px";b.title=r.n+" — "+r.d;axl.append(b);}
262
+ else{const k=el("div","rtick");k.style.left=pct(r.d)+"%";k.title=r.n+" — "+r.d;axl.append(k);}
263
+ });
264
+ const tl=el("div","todaylab","TODAY");tl.style.left=TP+"%";axl.append(tl);
265
+ axis.append(axl, el("div","corner r","PRs · churn"));
266
+ const G=document.getElementById("grid"); G.append(axis);
267
 
268
+ function lane(){
269
+ const L=el("div","lane");
270
+ TICKS.forEach(t=>{const g=el("div","gl"+(t.y?" y":""));g.style.left=t.p+"%";L.append(g);});
271
+ RELS.filter(r=>r.big).forEach(r=>{const g=el("div","rl big");g.style.left=pct(r.d)+"%";L.append(g);});
272
+ const t=el("div","todayline");t.style.left=TP+"%";L.append(t);
273
+ return L;
274
+ }
275
+ function dot(o,cls){
276
+ const d=el("div","pdot"+(cls||"")+(o.st==="OPEN"?" open":""));
277
+ d.style.left=pct(o.d)+"%";
278
+ d.addEventListener("mouseenter",e=>showCard(e,o));
279
+ d.addEventListener("mouseleave",hideCard);
280
+ d.addEventListener("click",()=>window.open("https://github.com/huggingface/transformers/pull/"+o.n,"_blank"));
281
+ return d;
282
+ }
283
+ function owners(list,total){
284
+ if(!list.length) return null;
285
+ const o=el("div","owners");
286
+ list.forEach(([login,n])=>{
287
+ const a=document.createElement("a");
288
+ a.href="https://github.com/"+encodeURIComponent(login); a.target="_blank"; a.rel="noopener";
289
+ a.title=login+" — "+n+" PR"+(n>1?"s":"");
290
+ const img=el("img","av"); img.src="https://github.com/"+encodeURIComponent(login)+".png?size=48";
291
+ img.alt=login; img.loading="lazy"; a.append(img); o.append(a);
292
+ });
293
+ if(total>list.length) o.append(el("span","more","+"+(total-list.length)));
294
+ return o;
295
+ }
296
 
297
  // ---------- rows ----------
298
+ for(const tr of TRACKS){
299
+ const sec=el("section","track"); sec.style.setProperty("--h",tr.hue);
300
+ const th=el("div","thead");
301
+ th.append(el("h2",null,tr.name),el("span","blurb",tr.blurb),el("span","cnt",tr.items.length+" items"));
302
  sec.append(th);
303
+ for(const i of tr.items){
304
+ const row=el("div","row s-"+i.s); row.dataset.s=i.s;
305
+ const L=el("div","lbl");
306
+ const nm=el("div","name"); nm.append(el("span","dot"),el("span","ttl",i.t)); L.append(nm);
307
+ if(i.tag){const w=el("div");w.append(el("span","tag",i.tag));L.append(w);}
308
+ if(i.d) L.append(el("div","desc",i.d));
309
+ if(i.note) L.append(el("div","note",i.note));
310
+ const uniq=new Set(i.prs.map(p=>p.by).filter(Boolean)).size;
311
+ const ow=owners(i.roll.owners,uniq);
312
+ if(ow) L.append(ow);
313
+ if(i.links.length){
314
+ const lw=ow||el("div","owners"); if(!ow) L.append(lw);
315
+ i.links.forEach(([lab,url])=>{const a=el("a","ext",lab);a.href=url;a.target="_blank";a.rel="noopener";lw.append(a);});
 
 
 
 
 
 
 
 
 
316
  }
317
  row.append(L);
318
 
319
+ const ln=lane();
320
+ if(i.span){
321
+ const [a,b]=i.span, l=pct(a);
322
+ const running=(i.s==="wip"||i.s==="open");
323
+ const end=running?Math.max(pct(b),TP):pct(b);
324
+ const bar=el("div","bar"); bar.style.left=l+"%"; bar.style.width=Math.max(end-l,.25)+"%";
325
+ bar.title=a+" → "+(running?"ongoing":b)+" ("+dur(a,b)+(running?", still running":"")+")";
326
+ ln.append(bar);
327
+ i.prs.forEach(p=>ln.append(dot(p)));
328
+ const dd=el("div","dur",dur(a,b)+(running?" →":"")); dd.style.left=end+"%"; ln.append(dd);
329
  } else {
330
+ const n=el("div","none",i.s==="planned"?"not started":"shipped outside this repo");
331
+ n.style.left="0%"; ln.append(n);
332
  }
333
+ row.append(ln);
334
+
335
+ const R=el("div","rstat");
336
+ if(i.prs.length){
337
+ R.append(el("div","big",i.roll.n+" PR"+(i.roll.n>1?"s":"")));
338
+ const s=el("div","sm");
339
+ s.innerHTML='<span class="plus">+'+nf(i.roll.add)+'</span> <span class="minus">−'+nf(i.roll.del)+'</span><br>'
340
+ +i.roll.files+' files · '+i.roll.commits+' commits';
341
+ R.append(s);
342
+ } else R.append(el("div","sm","—"));
343
+ row.append(R);
344
  sec.append(row);
345
  }
346
  G.append(sec);
347
  }
348
 
349
+ // ---------- model additions ----------
350
+ const msec=el("section","track"); msec.style.setProperty("--h",300);
351
+ const mth=el("div","thead");
352
+ mth.append(el("h2",null,"Model additions"),
353
+ el("span","blurb","Every new model directory that landed on main — one dot per PR."),
354
+ el("span","cnt","1 item"));
355
+ msec.append(mth);
356
+ const mrow=el("div","row s-done"); mrow.dataset.s="done";
357
+ const ML=el("div","lbl");
358
+ const mnm=el("div","name"); mnm.append(el("span","dot"),el("span","ttl","New model architectures")); ML.append(mnm);
359
+ const mw=el("div"); mw.append(el("span","tag",MODEL_ROW.n+" models · "+MODEL_ROW.authors+" authors")); ML.append(mw);
360
+ ML.append(el("div","desc","Detected from the first commit creating each src/transformers/models/<name>/ directory. "
361
+ +"Hover any dot for the model and its PR."));
362
+ const mo=owners(MODEL_ROW.owners,MODEL_ROW.authors); if(mo) ML.append(mo);
363
+ mrow.append(ML);
364
+ const mln=lane();
365
+ const mbar=el("div","bar"); mbar.style.left=pct(MODEL_ROW.span[0])+"%";
366
+ mbar.style.width=(Math.max(pct(MODEL_ROW.span[1]),TP)-pct(MODEL_ROW.span[0]))+"%";
367
+ mbar.title=MODEL_ROW.span[0]+" → ongoing"; mln.append(mbar);
368
+ MODELS.forEach(m=>mln.append(dot({n:m.pr,d:m.d,t:m.m+" — "+m.t,st:"MERGED",by:m.by,b:"",a:0,r:0,f:0,c:0}," model")));
369
+ const mdd=el("div","dur",MODEL_ROW.n+" →"); mdd.style.left=Math.max(pct(MODEL_ROW.span[1]),TP)+"%"; mln.append(mdd);
370
+ mrow.append(mln);
371
+ const MR=el("div","rstat");
372
+ MR.append(el("div","big",MODEL_ROW.n));
373
+ const ms=el("div","sm"); ms.innerHTML="new models<br>"+MODEL_ROW.authors+" authors"; MR.append(ms);
374
+ mrow.append(MR);
375
+ msec.append(mrow); G.append(msec);
376
+
377
+ document.getElementById("legend").innerHTML=
378
+ '<i><span class="key" style="background:var(--done)"></span> shipped</i>'+
379
+ '<i><span class="key" style="background:var(--wip)"></span> in progress — bar fades out</i>'+
380
+ '<i><span class="key" style="background:var(--open)"></span> in review</i>'+
381
+ '<i><span class="key" style="background:var(--plan)"></span> not started</i>'+
382
+ '<i>● each dot = one PR (hover)</i>'+
383
+ '<i><span style="color:var(--rel);font-weight:800">┊</span> release</i>'+
384
  '<i><span style="color:var(--today);font-weight:800">▌</span> today</i>';
385
+ addEventListener("scroll",hideCard,{passive:true});
386
  </script>
387
  </body>
388
  </html>
models.json ADDED
@@ -0,0 +1 @@
 
 
1
+ [{"m": "rt_detr", "d": "2024-06-22", "pr": 29077, "by": "Sangbum Daniel Choi", "t": "New model support RTDETR"}, {"m": "instructblipvideo", "d": "2024-06-25", "pr": 30182, "by": "Raushan Turganbay", "t": "Add video modality for InstrucBLIP"}, {"m": "llava_next_video", "d": "2024-06-26", "pr": 31252, "by": "Raushan Turganbay", "t": "Add LLaVa NeXT Video"}, {"m": "gemma2", "d": "2024-06-27", "pr": 31659, "by": "Arthur", "t": "Add gemma 2"}, {"m": "zoedepth", "d": "2024-07-08", "pr": 30136, "by": "NielsRogge", "t": "Add ZoeDepth"}, {"m": "hiera", "d": "2024-07-12", "pr": 30356, "by": "Naman Garg", "t": "Adding hiera"}, {"m": "chameleon", "d": "2024-07-17", "pr": 31534, "by": "Raushan Turganbay", "t": "Chameleon: add model"}, {"m": "nemotron", "d": "2024-08-06", "pr": 31699, "by": "Ao Tang", "t": "Add Nemotron HF Support"}, {"m": "mamba2", "d": "2024-08-06", "pr": 32080, "by": "Pablo Montalvo", "t": "Add codestral mamba2"}, {"m": "qwen2_audio", "d": "2024-08-08", "pr": 32137, "by": "Yunfei Chu", "t": "Add Qwen2-Audio"}, {"m": "falcon_mamba", "d": "2024-08-12", "pr": 32615, "by": "Younes Belkada", "t": "Add new model"}, {"m": "dac", "d": "2024-08-19", "pr": 31494, "by": "Kamil Akesbi", "t": "Add Descript-Audio-Codec model"}, {"m": "qwen2_vl", "d": "2024-08-26", "pr": 32318, "by": "Shijie", "t": "support qwen2-vl"}, {"m": "granite", "d": "2024-08-27", "pr": 31502, "by": "Mayank Mishra", "t": "Granite language models"}, {"m": "olmoe", "d": "2024-09-03", "pr": 32406, "by": "Niklas Muennighoff", "t": "Add OLMoE"}, {"m": "llava_onevision", "d": "2024-09-05", "pr": 32673, "by": "Raushan Turganbay", "t": "Llava Onevision: add model"}, {"m": "pixtral", "d": "2024-09-14", "pr": 33449, "by": "Arthur", "t": "Add support for Pixtral"}, {"m": "mimi", "d": "2024-09-18", "pr": 33565, "by": "Yoach Lacombe", "t": "Codec integration"}, {"m": "granitemoe", "d": "2024-09-20", "pr": 33207, "by": "Mayank Mishra", "t": "Granitemoe"}, {"m": "omdet_turbo", "d": "2024-09-25", "pr": 31843, "by": "Yoni Gozlan", "t": "Add OmDet-Turbo"}, {"m": "mllama", "d": "2024-09-25", "pr": 33703, "by": "Arthur", "t": "Add MLLama"}, {"m": "idefics3", "d": "2024-09-25", "pr": 32473, "by": "Andr\u00e9s Marafioti", "t": "Add Idefics 3!"}, {"m": "phimoe", "d": "2024-10-04", "pr": 33363, "by": "Amit Garg", "t": "PhiMoE"}, {"m": "zamba", "d": "2024-10-04", "pr": 30950, "by": "pglorio", "t": "Add Zamba"}, {"m": "myt5", "d": "2024-10-06", "pr": 31286, "by": "TomLim", "t": "[WIP] Add Tokenizer for MyT5 Model"}, {"m": "moshi", "d": "2024-10-16", "pr": 33624, "by": "Yoach Lacombe", "t": "Moshi integration"}, {"m": "glm", "d": "2024-10-18", "pr": 33823, "by": "Cyril Vallez", "t": "add Glm"}, {"m": "olmo_1124", "d": "2024-11-18", "pr": 34551, "by": "Shane A", "t": "Add OLMo November 2024"}, {"m": "ijepa", "d": "2024-12-05", "pr": 33125, "by": "Jo\u00e3o Marcelo", "t": "Add I-JEPA"}, {"m": "aria", "d": "2024-12-06", "pr": 34157, "by": "Aymeric Roucher", "t": "Add Aria"}, {"m": "timm_wrapper", "d": "2024-12-11", "pr": 34564, "by": "Pavel Iakubovskii", "t": "Add TimmWrapper"}, {"m": "cohere2", "d": "2024-12-13", "pr": 35224, "by": "alexrs-cohere", "t": "Add Cohere2 model"}, {"m": "colpali", "d": "2024-12-17", "pr": 33736, "by": "Tony Wu", "t": "Add ColPali to \ud83e\udd17 transformers"}, {"m": "bamba", "d": "2024-12-19", "pr": 34982, "by": "Yu Chin Fabian Lim", "t": "Add the Bamba Model"}, {"m": "modernbert", "d": "2024-12-19", "pr": 35158, "by": "Benjamin Warner", "t": "Add ModernBERT to Transformers"}, {"m": "dinov2_with_registers", "d": "2024-12-24", "pr": 35348, "by": "NielsRogge", "t": "Add DINOv2 with registers"}, {"m": "diffllama", "d": "2025-01-07", "pr": 34083, "by": "\u677e\u672c\u548c\u771f", "t": "Add diffllama"}, {"m": "textnet", "d": "2025-01-08", "pr": 34979, "by": "Jade Choghari", "t": "Add TextNet"}, {"m": "vitpose", "d": "2025-01-08", "pr": 30530, "by": "NielsRogge", "t": "Add ViTPose"}, {"m": "vitpose_backbone", "d": "2025-01-08", "pr": 30530, "by": "NielsRogge", "t": "Add ViTPose"}, {"m": "moonshine", "d": "2025-01-10", "pr": 34784, "by": "eustlb", "t": "Add Moonshine"}, {"m": "emu3", "d": "2025-01-10", "pr": 33770, "by": "Raushan Turganbay", "t": "[WIP] Emu3: add model"}, {"m": "helium", "d": "2025-01-13", "pr": 35669, "by": "Arthur", "t": "Add-helium"}, {"m": "superglue", "d": "2025-01-20", "pr": 29886, "by": "StevenBucaille", "t": "Add SuperGlue model"}, {"m": "qwen2_5_vl", "d": "2025-01-23", "pr": 35569, "by": "ShuaiBai623", "t": "add qwen2.5vl"}, {"m": "zamba2", "d": "2025-01-27", "pr": 34517, "by": "pglorio", "t": "Add Zamba2"}, {"m": "got_ocr2", "d": "2025-01-31", "pr": 34721, "by": "Yoni Gozlan", "t": "Add GOT-OCR 2.0 to Transformers"}, {"m": "dab_detr", "d": "2025-02-04", "pr": 30803, "by": "David", "t": "Add DAB-DETR for object detection"}, {"m": "rt_detr_v2", "d": "2025-02-06", "pr": 34773, "by": "Jade Choghari", "t": "Adding RT-DETRv2 for object detection"}, {"m": "depth_pro", "d": "2025-02-10", "pr": 34583, "by": "Armaghan Shakir", "t": "Add Apple's Depth-Pro for depth estimation"}, {"m": "granitemoeshared", "d": "2025-02-14", "pr": 35894, "by": "Mayank Mishra", "t": "add shared experts for upcoming Granite 4.0 language models"}, {"m": "smolvlm", "d": "2025-02-20", "pr": 36126, "by": "Orr Zohar", "t": "SmolVLM2"}, {"m": "siglip2", "d": "2025-02-21", "pr": 36323, "by": "Pavel Iakubovskii", "t": "Add SigLIP 2"}, {"m": "aya_vision", "d": "2025-03-04", "pr": 36521, "by": "Arthur", "t": "Add aya"}, {"m": "gemma3", "d": "2025-03-12", "pr": 36658, "by": "Ryan Mullins", "t": "Gemma3"}, {"m": "mistral3", "d": "2025-03-18", "pr": 36790, "by": "Cyril Vallez", "t": "Add Mistral3"}, {"m": "shieldgemma2", "d": "2025-03-20", "pr": 36678, "by": "Ryan Mullins", "t": "Shieldgemma2"}, {"m": "prompt_depth_anything", "d": "2025-03-21", "pr": 35401, "by": "Haotong LIN", "t": "Add Prompt Depth Anything Model"}, {"m": "phi4_multimodal", "d": "2025-03-25", "pr": 36939, "by": "Cyril Vallez", "t": "Add Phi4 multimodal"}, {"m": "deepseek_v3", "d": "2025-03-28", "pr": 35926, "by": "Minho Ryu", "t": "[WIP] add deepseek-v3"}, {"m": "qwen3", "d": "2025-03-31", "pr": 36878, "by": "Bo Zheng", "t": "Adding Qwen3 and Qwen3MoE"}, {"m": "qwen3_moe", "d": "2025-03-31", "pr": 36878, "by": "Bo Zheng", "t": "Adding Qwen3 and Qwen3MoE"}, {"m": "llama4", "d": "2025-04-05", "pr": 37307, "by": "Arthur", "t": "Add llama4"}, {"m": "glm4", "d": "2025-04-09", "pr": 37388, "by": "Arthur", "t": "Add glm4"}, {"m": "granite_speech", "d": "2025-04-11", "pr": 36801, "by": "Alex Brooks", "t": "Add Granite Speech Support"}, {"m": "qwen2_5_omni", "d": "2025-04-14", "pr": 36752, "by": "BakerBunker", "t": "Add Qwen2.5-Omni"}, {"m": "mlcd", "d": "2025-04-15", "pr": 36182, "by": "Huajie Tan", "t": "Add MLCD model"}, {"m": "timesfm", "d": "2025-04-16", "pr": 34082, "by": "Jinan Zhou", "t": "Add TimesFM Time Series Forecasting Model"}, {"m": "janus", "d": "2025-04-17", "pr": 36053, "by": "Yaswanth Gali", "t": "Add Janus model"}, {"m": "internvl", "d": "2025-04-18", "pr": 35968, "by": "Yoni Gozlan", "t": "Add InternVL (2.5 MPO)"}, {"m": "bitnet", "d": "2025-04-28", "pr": 37742, "by": "Mohamed Mekkouri", "t": "Add Bitnet model"}, {"m": "sam_hq", "d": "2025-04-28", "pr": 35147, "by": "sushmanth reddy", "t": "Samhq model addition"}, {"m": "d_fine", "d": "2025-04-29", "pr": 36261, "by": "Vladislav Bronzov", "t": "Add D-FINE Model into Transformers"}, {"m": "hgnet_v2", "d": "2025-04-29", "pr": 36261, "by": "Vladislav Bronzov", "t": "Add D-FINE Model into Transformers"}, {"m": "granitemoehybrid", "d": "2025-05-05", "pr": 37658, "by": "Sukriti Sharma", "t": "Add GraniteMoeHybrid support for 4.0"}, {"m": "csm", "d": "2025-05-07", "pr": 36719, "by": "eustlb", "t": "Add CSM model"}, {"m": "falcon_h1", "d": "2025-05-21", "pr": 38249, "by": "Younes Belkada", "t": "[MODEL] Add Falcon H1"}, {"m": "colqwen2", "d": "2025-06-02", "pr": 35778, "by": "Tony Wu", "t": "Add ColQwen2 to \ud83e\udd17 transformers"}, {"m": "minimax", "d": "2025-06-04", "pr": 35831, "by": "Armaghan Shakir", "t": "Add support for MiniMax's MiniMax-Text-01"}, {"m": "vjepa2", "d": "2025-06-11", "pr": 38746, "by": "Pavel Iakubovskii", "t": "Add V-JEPA 2"}, {"m": "lightglue", "d": "2025-06-17", "pr": 31718, "by": "StevenBucaille", "t": "Add LightGlue model"}, {"m": "arcee", "d": "2025-06-24", "pr": 38621, "by": "Crystalcareai", "t": "Add Arcee model support"}, {"m": "stt", "d": "2025-06-24", "pr": 38909, "by": "eustlb", "t": "Add kyutai stt"}, {"m": "glm4v", "d": "2025-06-25", "pr": 38431, "by": "Yuxuan Zhang", "t": "GLM-4.1V Model support"}, {"m": "t5gemma", "d": "2025-06-25", "pr": 38332, "by": "Biao Zhang", "t": "Encoder-Decoder Gemma"}, {"m": "dots1", "d": "2025-06-25", "pr": 38143, "by": "redmoe-moutain", "t": "[Model] add dots1"}, {"m": "smollm3", "d": "2025-06-25", "pr": 38755, "by": "Anton Lozhkov", "t": "Add SmolLM3"}, {"m": "dia", "d": "2025-06-26", "pr": 38405, "by": "Jaeyong Sung", "t": "Add Dia model"}, {"m": "gemma3n", "d": "2025-06-26", "pr": 39059, "by": "Ryan Mullins", "t": "Gemma 3n"}, {"m": "eomt", "d": "2025-06-27", "pr": 37610, "by": "Yaswanth Gali", "t": "\u2728 Add EoMT Model || \ud83d\udea8 Fix Mask2Former loss calculation"}, {"m": "doge", "d": "2025-07-08", "pr": 35891, "by": "Jingze Shi", "t": "Add Doge model"}, {"m": "aimv2", "d": "2025-07-08", "pr": 36625, "by": "Yaswanth Gali", "t": "Add Aimv2 model"}, {"m": "deepseek_v2", "d": "2025-07-09", "pr": 36400, "by": "Vladislav Bronzov", "t": "Add DeepSeek V2 Model into Transformers"}, {"m": "lfm2", "d": "2025-07-10", "pr": 39340, "by": "Paul Pak", "t": "LFM2"}, {"m": "perception_lm", "d": "2025-07-11", "pr": 37878, "by": "Shuming Hu", "t": "PerceptionLM"}, {"m": "modernbert_decoder", "d": "2025-07-15", "pr": 38967, "by": "Orion Weller", "t": "Add ModernBERT Decoder Models - ModernBERT, but trained with CLM!"}, {"m": "voxtral", "d": "2025-07-18", "pr": 39429, "by": "eustlb", "t": "Add voxtral"}, {"m": "glm4_moe", "d": "2025-07-21", "pr": 39393, "by": "Yuxuan Zhang", "t": "GLM-4 Update"}, {"m": "ernie4_5", "d": "2025-07-21", "pr": 39228, "by": "Anton Vlasjuk", "t": "[`Ernie 4.5`] Add ernie text models"}, {"m": "ernie4_5_moe", "d": "2025-07-21", "pr": 39228, "by": "Anton Vlasjuk", "t": "[`Ernie 4.5`] Add ernie text models"}, {"m": "efficientloftr", "d": "2025-07-22", "pr": 36355, "by": "StevenBucaille", "t": "Add EfficientLoFTR model"}, {"m": "deepseek_vl", "d": "2025-07-25", "pr": 36248, "by": "Armaghan Shakir", "t": "Add support for DeepseekAI's DeepseekVL"}, {"m": "deepseek_vl_hybrid", "d": "2025-07-25", "pr": 36248, "by": "Armaghan Shakir", "t": "Add support for DeepseekAI's DeepseekVL"}, {"m": "xlstm", "d": "2025-07-25", "pr": 39665, "by": "Cyril Vallez", "t": "Add xlstm model"}, {"m": "evolla", "d": "2025-07-26", "pr": 36232, "by": "Xibin Bayes Zhou", "t": "Add evolla rebase main"}, {"m": "exaone4", "d": "2025-07-26", "pr": 39129, "by": "lgai-exaone", "t": "Add EXAONE 4.0 model"}, {"m": "cohere2_vision", "d": "2025-07-31", "pr": 39810, "by": "Raushan Turganbay", "t": "[Model] Cohere2 Vision"}, {"m": "mm_grounding_dino", "d": "2025-08-01", "pr": 37925, "by": "rziga", "t": "Add MM Grounding DINO"}, {"m": "gpt_oss", "d": "2025-08-05", "pr": 39923, "by": "Arthur", "t": "Add GPT OSS model from OpenAI"}, {"m": "glm4v_moe", "d": "2025-08-08", "pr": 39805, "by": "Yuxuan Zhang", "t": "GLM-4.5V Model Support"}, {"m": "sam2", "d": "2025-08-14", "pr": 32317, "by": "Sangbum Daniel Choi", "t": "Add Segment Anything 2 (SAM2)"}, {"m": "sam2_video", "d": "2025-08-14", "pr": 32317, "by": "Sangbum Daniel Choi", "t": "Add Segment Anything 2 (SAM2)"}, {"m": "dinov3_convnext", "d": "2025-08-14", "pr": 40167, "by": "Pavel Iakubovskii", "t": "Fix docs typo"}, {"m": "dinov3_vit", "d": "2025-08-14", "pr": 40167, "by": "Pavel Iakubovskii", "t": "Fix docs typo"}, {"m": "xcodec", "d": "2025-08-15", "pr": 38248, "by": "Manal ML", "t": "Add X-Codec model"}, {"m": "ovis2", "d": "2025-08-18", "pr": 37088, "by": "Eon Kim", "t": "Add Ovis2 model and processor implementation"}, {"m": "kosmos2_5", "d": "2025-08-19", "pr": 31711, "by": "tic-top", "t": "Add Kosmos-2.5"}, {"m": "metaclip_2", "d": "2025-08-20", "pr": 39826, "by": "NielsRogge", "t": "Add MetaCLIP 2"}, {"m": "florence2", "d": "2025-08-20", "pr": 38188, "by": "Duc-Viet Hoang", "t": "Add support for Florence-2"}, {"m": "hunyuan_v1_dense", "d": "2025-08-22", "pr": 39606, "by": "yjc9696", "t": "HunYuan opensource"}, {"m": "hunyuan_v1_moe", "d": "2025-08-22", "pr": 39606, "by": "yjc9696", "t": "HunYuan opensource"}, {"m": "seed_oss", "d": "2025-08-22", "pr": 40272, "by": "Fazzie", "t": "Addiing ByteDance Seed Seed-OSS"}, {"m": "apertus", "d": "2025-08-28", "pr": 39381, "by": "EduardDurech", "t": "Add Apertus"}, {"m": "qwen3_next", "d": "2025-09-10", "pr": 40771, "by": "Bo Zheng", "t": "Adding Support for Qwen3-Next"}, {"m": "ministral", "d": "2025-09-11", "pr": 40247, "by": "Manuel de Prada Corral", "t": "\u26a0\ufe0f \ud83d\udd34 Add ministral model"}, {"m": "vaultgemma", "d": "2025-09-12", "pr": 40851, "by": "Ryan Mullins", "t": "add: differential privacy research model"}, {"m": "qwen3_vl", "d": "2025-09-15", "pr": 40795, "by": "JJJYmmm", "t": "Adding Support for Qwen3-VL Series"}, {"m": "qwen3_vl_moe", "d": "2025-09-15", "pr": 40795, "by": "JJJYmmm", "t": "Adding Support for Qwen3-VL Series"}, {"m": "olmo3", "d": "2025-09-16", "pr": 40778, "by": "Shane A", "t": "Add Olmo3 model"}, {"m": "longcat_flash", "d": "2025-09-17", "pr": 40730, "by": "Pablo Montalvo", "t": "Add LongCat-Flash"}, {"m": "flex_olmo", "d": "2025-09-18", "pr": 40921, "by": "Shane A", "t": "Add FlexOlmo model"}, {"m": "lfm2_vl", "d": "2025-09-18", "pr": 40624, "by": "Raushan Turganbay", "t": "Add new model LFM2-VL"}, {"m": "blt", "d": "2025-09-19", "pr": 38579, "by": "Ita Zaporozhets", "t": "blt wip"}, {"m": "qwen3_omni_moe", "d": "2025-09-21", "pr": 41025, "by": "BakerBunker", "t": "Adding support for Qwen3Omni"}, {"m": "parakeet", "d": "2025-09-25", "pr": 39062, "by": "Nithin Rao", "t": "Add Parakeet"}, {"m": "edgetam", "d": "2025-09-29", "pr": 39800, "by": "Yoni Gozlan", "t": "Add EdgeTAM"}, {"m": "edgetam_video", "d": "2025-09-29", "pr": 39800, "by": "Yoni Gozlan", "t": "Add EdgeTAM"}, {"m": "lfm2_moe", "d": "2025-10-07", "pr": 41401, "by": "Paul Pak", "t": "[Model] Lfm2Moe"}, {"m": "cwm", "d": "2025-10-09", "pr": 41199, "by": "Jacob Kahn", "t": "Add Code World Model (CWM)"}, {"m": "video_llama_3", "d": "2025-10-13", "pr": 40499, "by": "Kehan Li", "t": "[model] Add VideoLLaMA3 implementation"}, {"m": "audioflamingo3", "d": "2025-11-12", "pr": 40290, "by": "Lasha Koroshinadze", "t": "[models] Add AudioFlamingo3 integration"}, {"m": "glm46v", "d": "2025-11-15", "pr": 42122, "by": "Yuxuan Zhang", "t": "GLM-V update with new processor"}, {"m": "sam3", "d": "2025-11-19", "pr": 42285, "by": "Yoni Gozlan", "t": "Add SAM3 to \ud83e\udd17 Transformers"}, {"m": "sam3_tracker", "d": "2025-11-19", "pr": 42285, "by": "Yoni Gozlan", "t": "Add SAM3 to \ud83e\udd17 Transformers"}, {"m": "sam3_tracker_video", "d": "2025-11-19", "pr": 42285, "by": "Yoni Gozlan", "t": "Add SAM3 to \ud83e\udd17 Transformers"}, {"m": "sam3_video", "d": "2025-11-19", "pr": 42285, "by": "Yoni Gozlan", "t": "Add SAM3 to \ud83e\udd17 Transformers"}, {"m": "nanochat", "d": "2025-11-27", "pr": 41634, "by": "burtenshaw", "t": "[MODEL] Nanochat implementation"}, {"m": "afmoe", "d": "2025-11-29", "pr": 42168, "by": "Raghav Ravishankar", "t": "Add afmoe model"}, {"m": "t5gemma2", "d": "2025-12-01", "pr": 41834, "by": "Biao Zhang", "t": "T5gemma2"}, {"m": "ministral3", "d": "2025-12-01", "pr": 42498, "by": "Patrick von Platen", "t": "[Ministral 3] Add ministral 3"}, {"m": "fast_vlm", "d": "2025-12-02", "pr": 41112, "by": "Kamila Luczaj (Woo-chai)", "t": "Add FastVLM"}, {"m": "lasr", "d": "2025-12-05", "pr": 42648, "by": "eustlb", "t": "Lasr model"}, {"m": "paddleocr_vl", "d": "2025-12-11", "pr": 42178, "by": "zhang-prog", "t": "[Model] Add PaddleOCR-VL Model Support"}, {"m": "pe_audio", "d": "2025-12-16", "pr": 42905, "by": "eustlb", "t": " Sam: Perception Encoder Audiovisual"}, {"m": "pe_audio_video", "d": "2025-12-16", "pr": 42905, "by": "eustlb", "t": " Sam: Perception Encoder Audiovisual"}, {"m": "pe_video", "d": "2025-12-16", "pr": 42905, "by": "eustlb", "t": " Sam: Perception Encoder Audiovisual"}, {"m": "jais2", "d": "2025-12-17", "pr": 42684, "by": "sarathc-cerebras", "t": "adds jais2 model support"}, {"m": "pixio", "d": "2025-12-17", "pr": 42795, "by": "Lihe Yang", "t": "Add Pixio pre-trained models"}, {"m": "ernie4_5_vl_moe", "d": "2025-12-19", "pr": 39585, "by": "Anton Vlasjuk", "t": "[`Ernie 4.5`] Ernie VL models"}, {"m": "glmasr", "d": "2025-12-24", "pr": 42875, "by": "Yuxuan Zhang", "t": "GLM-ASR Support"}, {"m": "minimax_m2", "d": "2026-01-09", "pr": 42028, "by": "Roger Young", "t": "Add support for MiniMax-M2"}, {"m": "lw_detr", "d": "2026-01-12", "pr": 40991, "by": "StevenBucaille", "t": "Add LWDetr model"}, {"m": "glm4_moe_lite", "d": "2026-01-13", "pr": 43031, "by": "Yuxuan Zhang", "t": "[GLM-4.7] GLM-Lite Supoort"}, {"m": "glm_image", "d": "2026-01-13", "pr": 43100, "by": "Yuxuan Zhang", "t": "[GLM-Image] AR Model Support for GLM-Image"}, {"m": "lighton_ocr", "d": "2026-01-14", "pr": 41621, "by": "Aubertin Baptiste", "t": "Add LightOnOCR model implementation"}, {"m": "solar_open", "d": "2026-01-22", "pr": 43244, "by": "Inseo Song", "t": "Add Solar-Open Model"}, {"m": "glm_ocr", "d": "2026-01-27", "pr": 43391, "by": "Yuxuan Zhang", "t": "[GLM-OCR] GLM-OCR Support"}, {"m": "youtu", "d": "2026-01-28", "pr": 43166, "by": "Lu junru", "t": "Add Youtu-LLM model"}, {"m": "pp_doclayout_v3", "d": "2026-01-29", "pr": 43098, "by": "zhang-prog", "t": "[Model] Add PP-DocLayoutV3 Model Support"}, {"m": "eomt_dinov3", "d": "2026-02-02", "pr": 41212, "by": "NielsRogge", "t": "Add EoMT with DINOv3 backbone"}, {"m": "exaone_moe", "d": "2026-02-04", "pr": 43080, "by": "Junwon Hwang", "t": "Add EXAONE-MoE implementations"}, {"m": "moonshine_streaming", "d": "2026-02-04", "pr": 43702, "by": "eustlb", "t": "Add moonshine streaming"}, {"m": "vibevoice_acoustic_tokenizer", "d": "2026-02-06", "pr": 43400, "by": "Eric Bezzam", "t": "Add VibeVoice Acoustic Tokenizer"}, {"m": "qwen3_5", "d": "2026-02-09", "pr": 43830, "by": "Bo Zheng", "t": "Adding Support for Qwen3.5"}, {"m": "qwen3_5_moe", "d": "2026-02-09", "pr": 43830, "by": "Bo Zheng", "t": "Adding Support for Qwen3.5"}, {"m": "glm_moe_dsa", "d": "2026-02-09", "pr": 43858, "by": "Cyril Vallez", "t": "Add GlmMoeDsa"}, {"m": "voxtral_realtime", "d": "2026-02-16", "pr": 43769, "by": "eustlb", "t": "Add Voxtral Realtime"}, {"m": "higgs_audio_v2", "d": "2026-02-19", "pr": 40294, "by": "Shuai Zheng", "t": "Add Higgs Audio V2 Model"}, {"m": "higgs_audio_v2_tokenizer", "d": "2026-02-19", "pr": 40294, "by": "Shuai Zheng", "t": "Add Higgs Audio V2 Model"}, {"m": "colmodernvbert", "d": "2026-02-23", "pr": 42504, "by": "Paul Teiletche", "t": "Add ModernVBERT models"}, {"m": "modernvbert", "d": "2026-02-23", "pr": 42504, "by": "Paul Teiletche", "t": "Add ModernVBERT models"}, {"m": "olmo_hybrid", "d": "2026-02-26", "pr": 43358, "by": "Yanhong Li", "t": "Add OLMo Hybrid model"}, {"m": "pp_doclayout_v2", "d": "2026-02-27", "pr": 43018, "by": "zhang-prog", "t": "[Model] Add PP-DocLayoutV2 Model Support"}, {"m": "timesfm2_5", "d": "2026-02-27", "pr": 41763, "by": "Kashif Rasul", "t": "Timesfm 2.5"}, {"m": "vibevoice_asr", "d": "2026-03-02", "pr": 43625, "by": "Eric Bezzam", "t": "Add VibeVoice ASR"}, {"m": "nemotron_h", "d": "2026-03-03", "pr": 44390, "by": "Li Ding", "t": "add support for nemotron_3"}, {"m": "eurobert", "d": "2026-03-04", "pr": 39455, "by": "Arthur", "t": "Add eurobert"}, {"m": "chmv2", "d": "2026-03-11", "pr": 44595, "by": "Yoni Gozlan", "t": "Add CHMv2"}, {"m": "pp_ocrv5_server_det", "d": "2026-03-13", "pr": 43274, "by": "XingweiDeng", "t": "[Model] Add PP-OCRV5_server_det Model Support"}, {"m": "pp_lcnet", "d": "2026-03-13", "pr": 43247, "by": "XingweiDeng", "t": "[Model] Add PP-OCRV5_mobile_det Model Support"}, {"m": "pp_lcnet_v3", "d": "2026-03-13", "pr": 43247, "by": "XingweiDeng", "t": "[Model] Add PP-OCRV5_mobile_det Model Support"}, {"m": "pp_ocrv5_mobile_det", "d": "2026-03-13", "pr": 43247, "by": "XingweiDeng", "t": "[Model] Add PP-OCRV5_mobile_det Model Support"}, {"m": "pi0", "d": "2026-03-16", "pr": 44160, "by": "Pablo Montalvo", "t": "Add model lerobot PI0 to transformers"}, {"m": "mistral4", "d": "2026-03-16", "pr": 44760, "by": "Julien Denize", "t": "Add Mistral 4"}, {"m": "jina_embeddings_v3", "d": "2026-03-19", "pr": 44251, "by": "Sai-Suraj-27", "t": "Add `Jina-Embeddings-V3` Model"}, {"m": "pp_ocrv5_mobile_rec", "d": "2026-03-19", "pr": 44808, "by": "zhang-prog", "t": "[Model] Add PP-OCRv5_server_rec and PP-OCRv5_mobile_rec models Support"}, {"m": "pp_ocrv5_server_rec", "d": "2026-03-19", "pr": 44808, "by": "zhang-prog", "t": "[Model] Add PP-OCRv5_server_rec and PP-OCRv5_mobile_rec models Support"}, {"m": "pp_chart2table", "d": "2026-03-20", "pr": 43767, "by": "XingweiDeng", "t": "[Model] Add PP-Chart2Table Model Support"}, {"m": "slanext", "d": "2026-03-21", "pr": 43707, "by": "Liu Jiaxuan", "t": "[Model] Add SLANeXt Model Support"}, {"m": "uvdoc", "d": "2026-03-21", "pr": 43385, "by": "XingweiDeng", "t": "[Model] Add UVDoc Model Support"}, {"m": "videomt", "d": "2026-03-25", "pr": 44285, "by": "NielsRogge", "t": "Add VidEoMT"}, {"m": "cohere_asr", "d": "2026-03-26", "pr": 45023, "by": "eustlb", "t": "Add cohere asr"}, {"m": "musicflamingo", "d": "2026-03-30", "pr": 43538, "by": "Lasha Koroshinadze", "t": "Add Music Flamingo"}, {"m": "nomic_bert", "d": "2026-04-02", "pr": 43067, "by": "Sonny Cooper", "t": "Internalise the NomicBERT model"}, {"m": "gemma4", "d": "2026-04-02", "pr": 45192, "by": "Ryan Mullins", "t": "casually dropping the most capable open weights on the planet"}, {"m": "sam3_lite_text", "d": "2026-04-13", "pr": 44320, "by": "NielsRogge", "t": "Add SAM3-LiteText"}, {"m": "qianfan_ocr", "d": "2026-04-17", "pr": 45280, "by": "bairongz", "t": "add Qianfan-OCR model definition"}, {"m": "slanet", "d": "2026-04-22", "pr": 45532, "by": "zhang-prog", "t": "[Model] Add SLANet Model Support"}, {"m": "hy_v3", "d": "2026-04-22", "pr": 45574, "by": "Anton Vlasjuk", "t": "Fix typos"}, {"m": "openai_privacy_filter", "d": "2026-04-22", "pr": 45580, "by": "Anton Vlasjuk", "t": "[`Privacy Filter`] Add model"}, {"m": "deimv2", "d": "2026-04-27", "pr": 44339, "by": "Harshal Janjani", "t": "model: Add DEIMv2 to Transformers"}, {"m": "laguna", "d": "2026-04-28", "pr": 45673, "by": "Joe Rowell", "t": "Laguna XS.2 implementation"}, {"m": "minicpmv4_6", "d": "2026-04-28", "pr": 45689, "by": "Anton Vlasjuk", "t": "fixing more typos"}, {"m": "granite_speech_plus", "d": "2026-04-29", "pr": 45695, "by": "Zvi Kons", "t": "Support for a new Granite-Speech-Plus model"}, {"m": "pp_formulanet", "d": "2026-04-30", "pr": 45626, "by": "zhang-prog", "t": "[Model] Add PP-FormulaNet Model Support"}, {"m": "deepseek_v4", "d": "2026-05-02", "pr": 45643, "by": "Arthur", "t": "Add DeepSeek V4"}, {"m": "exaone4_5", "d": "2026-05-04", "pr": 45471, "by": "Junwon Hwang", "t": "Add EXAONE 4.5 implementations"}, {"m": "gemma4_assistant", "d": "2026-05-05", "pr": 45788, "by": "Sindhu Raghuram Panyam", "t": "First model"}, {"m": "granite4_vision", "d": "2026-05-05", "pr": 45597, "by": "artem-spector", "t": "Add Granite 4.1 Vision (granite4_vision)"}, {"m": "rf_detr", "d": "2026-05-07", "pr": 36895, "by": "StevenBucaille", "t": "Add RF-DETR"}, {"m": "hyperclovax", "d": "2026-05-08", "pr": 44956, "by": "bigshanedogg", "t": "Add HyperCLOVAX SEED Think 14B"}, {"m": "hrm_text", "d": "2026-05-18", "pr": 46025, "by": "yifei wu", "t": "Add hrm text"}, {"m": "pp_lcnet_v4", "d": "2026-05-19", "pr": 45838, "by": "zhang-prog", "t": "[Model] Add PP-OCRv6 Models Support"}, {"m": "pp_ocrv6_medium_det", "d": "2026-05-19", "pr": 45838, "by": "zhang-prog", "t": "[Model] Add PP-OCRv6 Models Support"}, {"m": "pp_ocrv6_small_det", "d": "2026-05-19", "pr": 45838, "by": "zhang-prog", "t": "[Model] Add PP-OCRv6 Models Support"}, {"m": "pp_ocrv6_small_rec", "d": "2026-05-19", "pr": 45838, "by": "zhang-prog", "t": "[Model] Add PP-OCRv6 Models Support"}, {"m": "pp_ocrv6_tiny_rec", "d": "2026-05-19", "pr": 45838, "by": "zhang-prog", "t": "[Model] Add PP-OCRv6 Models Support"}, {"m": "cohere2_moe", "d": "2026-05-20", "pr": 46115, "by": "Cyril Vallez", "t": "Add new cohere2_moe model"}, {"m": "glmga", "d": "2026-05-26", "pr": 46184, "by": "Yuxuan Zhang", "t": "[GLM-4.6V] Update with GLM-GA Processor"}, {"m": "mellum", "d": "2026-05-28", "pr": 46112, "by": "Madeesh Kannan", "t": "feat: Add support for JetBrains' `Mellum` v2 code generation model"}, {"m": "deepseek_ocr2", "d": "2026-06-02", "pr": 45075, "by": "Eon Kim", "t": "Add Deepseek-OCR-2 model"}, {"m": "sapiens2", "d": "2026-06-03", "pr": 45919, "by": "guarin", "t": "Add Sapiens2 Model"}, {"m": "gemma4_unified", "d": "2026-06-03", "pr": 46385, "by": "Douglas Reid", "t": "who needs encoders?"}, {"m": "gemma4_unified_assistant", "d": "2026-06-03", "pr": 46385, "by": "Douglas Reid", "t": "who needs encoders?"}, {"m": "cosmos3_omni", "d": "2026-06-05", "pr": 46146, "by": "MaciejBalaNV", "t": "Added cosmos3 model"}, {"m": "diffusion_gemma", "d": "2026-06-10", "pr": 46540, "by": "Joao Gante", "t": "GPU go brr"}, {"m": "deepseek_v32", "d": "2026-06-11", "pr": 41251, "by": "Arthur", "t": "Add deepseek 3.2 exp"}, {"m": "minimax_m3_vl", "d": "2026-06-12", "pr": 46600, "by": "Arthur", "t": "Add minimax m3vl"}, {"m": "nemotron_asr_streaming", "d": "2026-06-19", "pr": 46332, "by": "eustlb", "t": "Add Nemotron ASR Streaming"}, {"m": "videoprism", "d": "2026-06-20", "pr": 39895, "by": "MAHIR DAIYAN", "t": "Add Videoprism"}, {"m": "minicpm3", "d": "2026-06-23", "pr": 41116, "by": "Minho Ryu", "t": "Add MiniCPM3"}, {"m": "xcodec2", "d": "2026-06-25", "pr": 44178, "by": "Eric Bezzam", "t": "Add xcodec2 model"}, {"m": "qwen3_asr", "d": "2026-06-26", "pr": 43838, "by": "Muhammed", "t": "Qwen3 ASR and Forced Aligner"}, {"m": "nemotron3_5_asr", "d": "2026-06-27", "pr": 46565, "by": "eustlb", "t": "Add Nemotron 3.5 ASR Streaming"}, {"m": "radio", "d": "2026-07-01", "pr": 46425, "by": "meatybobby", "t": "Add support for RADIO models"}, {"m": "zaya", "d": "2026-07-01", "pr": 45862, "by": "JJJYmmm", "t": "[new model] Add Zyphra/ZAYA1-8B"}, {"m": "mimo_v2_flash", "d": "2026-07-01", "pr": 45144, "by": "casinca", "t": "Add Xiaomi MiMo-V2"}, {"m": "hunyuan_vl", "d": "2026-07-03", "pr": 46417, "by": "Mi Jiazhi", "t": "Add HunYuan VL model"}, {"m": "kimi_k25", "d": "2026-07-03", "pr": 45630, "by": "Raushan Turganbay", "t": "Add new model: Kimi2-6"}, {"m": "tipsv2", "d": "2026-07-07", "pr": 46347, "by": "Zixuan Zhu", "t": "Add TIPSv2"}, {"m": "tipsv2_dpt", "d": "2026-07-07", "pr": 46347, "by": "Zixuan Zhu", "t": "Add TIPSv2"}, {"m": "inkling", "d": "2026-07-15", "pr": 47347, "by": "Pablo Montalvo", "t": "Add Inkling model"}, {"m": "cosmos3_edge", "d": "2026-07-16", "pr": 47181, "by": "Atharva Joshi", "t": "Add Cosmos3 Edge model support"}, {"m": "axk1", "d": "2026-07-24", "pr": 46867, "by": "Minsang Kim", "t": "add_axk1"}, {"m": "axk2", "d": "2026-07-25", "pr": 47528, "by": "Anton Vlasjuk", "t": "Add AXK2 from SKT"}, {"m": "granite_swa", "d": "2026-07-29", "pr": 47179, "by": "Davis Wertheimer", "t": "Add Granite-swa and Granitemoe-swa model support"}, {"m": "granitemoe_swa", "d": "2026-07-29", "pr": 47179, "by": "Davis Wertheimer", "t": "Add Granite-swa and Granitemoe-swa model support"}, {"m": "muse_glimmer", "d": "2026-08-10", "pr": 47867, "by": "Cyril Vallez", "t": "Add MuseGlimmer"}, {"m": "muse_glimmer_assistant", "d": "2026-08-10", "pr": 47867, "by": "Cyril Vallez", "t": "Add MuseGlimmer"}, {"m": "cohere_compass", "d": "2026-08-10", "pr": 47878, "by": "calpt", "t": "Add CohereCompass modeling"}, {"m": "step3p7", "d": "2026-08-16", "pr": 46658, "by": "Ita Zaporozhets", "t": "[new model] step 3.7"}, {"m": "esmc", "d": "2026-08-20", "pr": 46419, "by": "Matt", "t": "Port ESMC and ESMFold2 to Transformers"}, {"m": "esmfold2", "d": "2026-08-20", "pr": 46419, "by": "Matt", "t": "Port ESMC and ESMFold2 to Transformers"}, {"m": "granite_speech5", "d": "2026-08-25", "pr": 48288, "by": "eustlb", "t": "gs"}, {"m": "qwen4_exp", "d": "2026-08-26", "pr": 48337, "by": "Cyril Vallez", "t": "Add Qwen4Exp model"}, {"m": "glm5_next", "d": "2026-08-26", "pr": 48342, "by": "Shijin Zhang", "t": "[Glm 5.3 Flash] GLM 5.3 Flash Support"}, {"m": "vibevoice", "d": "2026-08-27", "pr": 40546, "by": "Zhiliang Peng", "t": "Implement VibeVoice"}]
pr_spans.json CHANGED
@@ -2,1041 +2,2055 @@
2
  "28847": {
3
  "t": "Fast image processor",
4
  "state": "MERGED",
5
- "created": "2024-02-02",
6
  "merged": "2024-06-11",
7
  "first": "2024-06-10",
8
- "last": "2024-06-11"
 
 
 
 
 
 
9
  },
10
  "31247": {
11
  "t": "Add torch.compile Support For Mamba",
12
  "state": "MERGED",
13
- "created": "2024-06-04",
14
  "merged": "2024-07-18",
15
  "first": "2024-06-04",
16
- "last": "2024-07-16"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
17
  },
18
  "32227": {
19
  "t": ">3-5x faster torch.compile forward compilation for autoregressive decoder models",
20
  "state": "MERGED",
21
- "created": "2024-07-25",
22
  "merged": "2024-07-31",
23
  "first": "2024-07-25",
24
- "last": "2024-07-31"
 
 
 
 
 
 
25
  },
26
  "33248": {
27
  "t": "Modular `transformers`: modularity and inheritance for new model additions",
28
  "state": "MERGED",
29
- "created": "2024-09-02",
30
  "merged": "2024-09-24",
31
  "first": "2024-09-02",
32
- "last": "2024-09-24"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33
  },
34
  "33730": {
35
  "t": "Make audio classification pipeline spec-compliant and add test",
36
  "state": "MERGED",
37
- "created": "2024-09-26",
38
  "merged": "2024-09-27",
39
  "first": "2024-09-27",
40
- "last": "2024-09-27"
 
 
 
 
 
 
41
  },
42
  "33769": {
43
  "t": "Make ASR pipeline compliant with Hub spec + add tests",
44
  "state": "MERGED",
45
- "created": "2024-09-27",
46
  "merged": "2024-10-01",
47
  "first": "2024-09-27",
48
- "last": "2024-10-01"
 
 
 
 
 
 
49
  },
50
  "33771": {
51
  "t": "[`AutoDocstring`] Based on inspect parsing of the signature",
52
  "state": "MERGED",
53
- "created": "2024-09-27",
54
  "merged": "2025-05-08",
55
  "first": "2024-10-10",
56
- "last": "2025-05-08"
 
 
 
 
 
 
57
  },
58
  "33899": {
59
  "t": "Image pipelines spec compliance",
60
  "state": "MERGED",
61
- "created": "2024-10-02",
62
  "merged": "2024-10-08",
63
  "first": "2024-10-02",
64
- "last": "2024-10-08"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
65
  },
66
  "34184": {
67
  "t": "Simplify Tensor Parallel implementation with PyTorch TP",
68
  "state": "MERGED",
69
- "created": "2024-10-15",
70
  "merged": "2024-11-18",
71
  "first": "2024-10-15",
72
- "last": "2024-11-18"
 
 
 
 
 
 
73
  },
74
  "34194": {
75
  "t": "feat: add support for tensor parallel training workflow with accelerate",
76
  "state": "MERGED",
77
- "created": "2024-10-16",
78
  "merged": "2025-02-18",
79
  "first": "2025-02-18",
80
- "last": "2025-02-18"
 
 
 
 
 
 
81
  },
82
  "34288": {
83
  "t": "Sync video classification pipeline with huggingface_hub spec",
84
  "state": "MERGED",
85
- "created": "2024-10-21",
86
  "merged": "2024-10-22",
87
  "first": "2024-10-21",
88
- "last": "2024-10-22"
 
 
 
 
 
 
89
  },
90
  "34487": {
91
  "t": "Large modular logic refactoring",
92
  "state": "MERGED",
93
- "created": "2024-10-29",
94
  "merged": "2024-11-01",
95
  "first": "2024-10-31",
96
- "last": "2024-11-01"
 
 
 
 
 
 
97
  },
98
  "34544": {
99
  "t": "Test: generate with `torch.compile(model.forward)` as a fast test",
100
  "state": "MERGED",
101
- "created": "2024-10-31",
102
  "merged": "2025-01-28",
103
  "first": "2024-10-31",
104
- "last": "2025-01-28"
 
 
 
 
 
 
105
  },
106
  "34564": {
107
  "t": "Add TimmWrapper",
108
  "state": "MERGED",
109
- "created": "2024-11-01",
110
  "merged": "2024-12-11",
111
  "first": "2024-11-01",
112
- "last": "2024-12-11"
 
 
 
 
 
 
113
  },
114
  "34729": {
115
  "t": "feat: allow to use hf-models",
116
  "state": "MERGED",
117
- "created": "2024-11-14",
118
  "merged": "2024-11-19",
119
  "first": "2024-11-14",
120
- "last": "2024-11-14"
 
 
 
 
 
 
121
  },
122
  "34800": {
123
  "t": "protect tensor parallel usage",
124
  "state": "MERGED",
125
- "created": "2024-11-19",
126
  "merged": "2024-11-19",
127
  "first": "2024-11-19",
128
- "last": "2024-11-19"
 
 
 
 
 
 
129
  },
130
  "34896": {
131
  "t": "[`GPTNeoX`] Flex Attention + Refactor",
132
  "state": "MERGED",
133
- "created": "2024-11-23",
134
  "merged": "2024-12-04",
135
  "first": "2024-11-23",
136
- "last": "2024-12-02"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
137
  },
138
  "35184": {
139
  "t": "Add compile test for fast image processor",
140
  "state": "MERGED",
141
- "created": "2024-12-10",
142
  "merged": "2024-12-23",
143
  "first": "2024-12-11",
144
- "last": "2024-12-18"
 
 
 
 
 
 
145
  },
146
  "35187": {
147
  "t": "make LlamaModel._update_causal_mask torch compilable",
148
  "state": "MERGED",
149
- "created": "2024-12-10",
150
  "merged": "2024-12-23",
151
  "first": "2024-12-23",
152
- "last": "2024-12-23"
 
 
 
 
 
 
153
  },
154
  "35235": {
155
  "t": "\ud83d\udea8All attention refactor\ud83d\udea8",
156
  "state": "MERGED",
157
- "created": "2024-12-12",
158
  "merged": "2024-12-18",
159
  "first": "2024-12-16",
160
- "last": "2024-12-18"
 
 
 
 
 
 
161
  },
162
  "35342": {
163
  "t": "Cleaner attention interfaces",
164
  "state": "MERGED",
165
- "created": "2024-12-19",
166
  "merged": "2024-12-20",
167
  "first": "2024-12-19",
168
- "last": "2024-12-20"
 
 
 
 
 
 
169
  },
170
  "35567": {
171
  "t": "Trainer Refactor: Part 1",
172
  "state": "MERGED",
173
- "created": "2025-01-08",
174
  "merged": "2025-01-29",
175
  "first": "2025-01-07",
176
- "last": "2025-01-23"
 
 
 
 
 
 
177
  },
178
  "35658": {
179
  "t": "[`Compile`] Only test compiling model forward pass",
180
  "state": "MERGED",
181
- "created": "2025-01-13",
182
  "merged": "2025-01-13",
183
  "first": "2025-01-13",
184
- "last": "2025-01-13"
 
 
 
 
 
 
185
  },
186
  "35794": {
187
  "t": "Auto-add `timm` tag to timm-wrapper models.",
188
  "state": "MERGED",
189
- "created": "2025-01-20",
190
  "merged": "2025-01-21",
191
  "first": "2025-01-20",
192
- "last": "2025-01-20"
 
 
 
 
 
 
193
  },
194
  "36091": {
195
  "t": "Add pipeline parallel plan to `PretrainedConfig` and `PreTrainedModel`",
196
  "state": "MERGED",
197
- "created": "2025-02-07",
198
  "merged": "2025-02-12",
199
  "first": "2025-02-07",
200
- "last": "2025-02-10"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
201
  },
202
  "36835": {
203
  "t": "Enhance Model Loading By Providing Parallelism, Uses Optional Env Flag",
204
  "state": "MERGED",
205
- "created": "2025-03-19",
206
  "merged": "2025-05-23",
207
  "first": "2025-03-19",
208
- "last": "2025-05-23"
 
 
 
 
 
 
209
  },
210
  "36853": {
211
  "t": "Use `deformable_detr` kernel from the Hub",
212
  "state": "MERGED",
213
- "created": "2025-03-20",
214
  "merged": "2025-03-21",
215
  "first": "2025-03-21",
216
- "last": "2025-03-21"
 
 
 
 
 
 
217
  },
218
  "36934": {
219
  "t": "Aligning modling code for GPT2 to work with vLLM (fallback)",
220
  "state": "MERGED",
221
- "created": "2025-03-24",
222
  "merged": "2025-05-02",
223
  "first": "2025-03-24",
224
- "last": "2025-05-01"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
225
  },
226
  "37376": {
227
  "t": "Add moe kernels",
228
  "state": "MERGED",
229
- "created": "2025-04-08",
230
  "merged": "2025-04-11",
231
  "first": "2025-04-08",
232
- "last": "2025-04-11"
 
 
 
 
 
 
233
  },
234
  "37381": {
235
  "t": "Fix mask handling for flex attention in llama/gemma2/mistral/qwen2",
236
  "state": "MERGED",
237
- "created": "2025-04-09",
238
  "merged": "2025-04-14",
239
  "first": "2025-04-08",
240
- "last": "2025-04-12"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
241
  },
242
  "37877": {
243
  "t": "parallelism goes brrr",
244
  "state": "MERGED",
245
- "created": "2025-04-29",
246
  "merged": "2025-05-20",
247
  "first": "2025-04-29",
248
- "last": "2025-05-20"
 
 
 
 
 
 
249
  },
250
  "37915": {
251
  "t": "[transformers x vLLM] standardize processors",
252
  "state": "MERGED",
253
- "created": "2025-05-01",
254
  "merged": "2025-05-27",
255
  "first": "2025-05-01",
256
- "last": "2025-05-26"
 
 
 
 
 
 
257
  },
258
  "38090": {
259
  "t": "Refactor `get_XXX_dataloader` from Trainer",
260
  "state": "MERGED",
261
- "created": "2025-05-12",
262
  "merged": "2025-05-19",
263
  "first": "2025-05-12",
264
- "last": "2025-05-13"
 
 
 
 
 
 
265
  },
266
  "38108": {
267
  "t": "\ud83d\udd34\ud83d\udd34\ud83d\udd34 [`Attention`] Refactor Attention Interface for Bart-based Models",
268
  "state": "MERGED",
269
- "created": "2025-05-13",
270
  "merged": "2025-05-22",
271
  "first": "2025-05-12",
272
- "last": "2025-05-22"
 
 
 
 
 
 
273
  },
274
  "38205": {
275
  "t": "Add kernelize to transformers",
276
  "state": "MERGED",
277
- "created": "2025-05-19",
278
  "merged": "2025-06-24",
279
  "first": "2025-06-24",
280
- "last": "2025-06-24"
 
 
 
 
 
 
281
  },
282
  "38242": {
283
  "t": "Add support for auto_docstring with model outputs",
284
  "state": "MERGED",
285
- "created": "2025-05-20",
286
  "merged": "2025-06-23",
287
  "first": "2025-05-20",
288
- "last": "2025-06-23"
 
 
 
 
 
 
289
  },
290
  "38262": {
291
  "t": "Protect ParallelInterface",
292
  "state": "MERGED",
293
- "created": "2025-05-21",
294
  "merged": "2025-05-21",
295
  "first": "2025-05-21",
296
- "last": "2025-05-21"
 
 
 
 
 
 
297
  },
298
  "38443": {
299
  "t": "Split `transformers chat` and `transformers serve` ",
300
  "state": "MERGED",
301
- "created": "2025-05-28",
302
  "merged": "2025-06-30",
303
  "first": "2025-06-24",
304
- "last": "2025-06-30"
 
 
 
 
 
 
305
  },
306
  "38645": {
307
  "t": "support loading qwen3 gguf",
308
  "state": "MERGED",
309
- "created": "2025-06-06",
310
  "merged": "2025-07-15",
311
  "first": "2025-06-06",
312
- "last": "2025-06-25"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
313
  },
314
  "38760": {
315
  "t": "Prepare for TF+Jax deprecation",
316
  "state": "MERGED",
317
- "created": "2025-06-11",
318
  "merged": "2025-06-11",
319
  "first": "2025-06-11",
320
- "last": "2025-06-11"
 
 
 
 
 
 
321
  },
322
  "38814": {
323
  "t": "\ud83c\udfaf Trackio integration",
324
  "state": "MERGED",
325
- "created": "2025-06-13",
326
  "merged": "2025-07-22",
327
  "first": "2025-06-13",
328
- "last": "2025-07-22"
 
 
 
 
 
 
329
  },
330
  "39011": {
331
  "t": "[docs] @auto_docstring",
332
  "state": "MERGED",
333
- "created": "2025-06-24",
334
  "merged": "2025-06-26",
335
  "first": "2025-06-26",
336
- "last": "2025-06-26"
 
 
 
 
 
 
337
  },
338
  "39155": {
339
  "t": "Responses API in `transformers serve`",
340
  "state": "MERGED",
341
- "created": "2025-07-01",
342
  "merged": "2025-07-16",
343
  "first": "2025-07-03",
344
- "last": "2025-07-16"
 
 
 
 
 
 
345
  },
346
  "39198": {
347
  "t": "CI workflow for performed test regressions",
348
  "state": "MERGED",
349
- "created": "2025-07-03",
350
  "merged": "2025-07-16",
351
  "first": "2025-07-03",
352
- "last": "2025-07-15"
 
 
 
 
 
 
353
  },
354
  "39454": {
355
  "t": "Transformers serve VLM",
356
  "state": "MERGED",
357
- "created": "2025-07-16",
358
  "merged": "2025-07-23",
359
  "first": "2025-07-18",
360
- "last": "2025-07-23"
 
 
 
 
 
 
361
  },
362
  "39457": {
363
  "t": "[Paged-Attention] Handle continuous batching for repetition penalty",
364
  "state": "MERGED",
365
- "created": "2025-07-16",
366
  "merged": "2025-07-22",
367
  "first": "2025-07-16",
368
- "last": "2025-07-22"
 
 
 
 
 
 
369
  },
370
  "39501": {
371
  "t": "Add ep",
372
  "state": "MERGED",
373
- "created": "2025-07-18",
374
  "merged": "2025-07-25",
375
  "first": "2025-07-17",
376
- "last": "2025-07-25"
 
 
 
 
 
 
377
  },
378
  "39562": {
379
  "t": "Fixes needed for n-d parallelism and TP",
380
  "state": "MERGED",
381
- "created": "2025-07-21",
382
  "merged": "2025-07-22",
383
  "first": "2025-07-21",
384
- "last": "2025-07-22"
 
 
 
 
 
 
385
  },
386
  "39583": {
387
  "t": "Update recent processors for vLLM backend",
388
  "state": "MERGED",
389
- "created": "2025-07-22",
390
  "merged": "2025-07-24",
391
  "first": "2025-07-22",
392
- "last": "2025-07-22"
 
 
 
 
 
 
393
  },
394
  "39595": {
395
  "t": "[Trackio] Allow single-gpu training and monitor power",
396
  "state": "MERGED",
397
- "created": "2025-07-23",
398
  "merged": "2025-07-23",
399
  "first": "2025-07-23",
400
- "last": "2025-07-23"
 
 
 
 
 
 
401
  },
402
  "39638": {
403
  "t": "Support loading Qwen3 MoE GGUF",
404
  "state": "MERGED",
405
- "created": "2025-07-24",
406
  "merged": "2025-07-29",
407
  "first": "2025-07-24",
408
- "last": "2025-07-29"
 
 
 
 
 
 
409
  },
410
  "39768": {
411
  "t": "Benchmarking improvements",
412
  "state": "MERGED",
413
- "created": "2025-07-29",
414
  "merged": "2025-08-15",
415
  "first": "2025-07-27",
416
- "last": "2025-08-15"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
417
  },
418
  "40426": {
419
  "t": "Continuous batching refactor",
420
  "state": "MERGED",
421
- "created": "2025-08-25",
422
  "merged": "2025-08-26",
423
  "first": "2025-08-26",
424
- "last": "2025-08-26"
 
 
 
 
 
 
425
  },
426
  "40619": {
427
  "t": "[CP] Add attention_mask to the buffer when the mask is causal ",
428
  "state": "MERGED",
429
- "created": "2025-09-02",
430
  "merged": "2025-09-03",
431
  "first": "2025-09-02",
432
- "last": "2025-09-03"
 
 
 
 
 
 
433
  },
434
  "40654": {
435
  "t": "Remove TF/Flax examples",
436
  "state": "MERGED",
437
- "created": "2025-09-03",
438
  "merged": "2025-09-03",
439
  "first": "2025-09-03",
440
- "last": "2025-09-03"
 
 
 
 
 
 
441
  },
442
  "40760": {
443
  "t": "\ud83d\udea8\ud83d\udea8\ud83d\udea8 Fully remove Tensorflow and Jax support library-wide",
444
  "state": "MERGED",
445
- "created": "2025-09-09",
446
  "merged": "2025-09-18",
447
  "first": "2025-09-18",
448
- "last": "2025-09-18"
 
 
 
 
 
 
449
  },
450
  "40936": {
451
  "t": "rm slow tokenizers",
452
  "state": "MERGED",
453
- "created": "2025-09-17",
454
  "merged": "2025-11-27",
455
  "first": "2025-10-10",
456
- "last": "2025-11-27"
 
 
 
 
 
 
457
  },
458
  "40997": {
459
  "t": "\ud83d\udea8 [v5] remove deprecated entry point",
460
  "state": "MERGED",
461
- "created": "2025-09-19",
462
  "merged": "2025-09-19",
463
  "first": "2025-09-19",
464
- "last": "2025-09-19"
 
 
 
 
 
 
465
  },
466
  "40998": {
467
  "t": "\ud83d\udea8 [v5] remove generate output retrocompatibility aliases",
468
  "state": "MERGED",
469
- "created": "2025-09-19",
470
  "merged": "2025-09-19",
471
  "first": "2025-09-19",
472
- "last": "2025-09-19"
 
 
 
 
 
 
473
  },
474
  "41017": {
475
  "t": "\ud83d\udea8 [V5] Remove deprecated training arguments ",
476
  "state": "MERGED",
477
- "created": "2025-09-20",
478
  "merged": "2025-09-24",
479
  "first": "2025-09-20",
480
- "last": "2025-09-20"
 
 
 
 
 
 
481
  },
482
  "41058": {
483
  "t": "Remove mention of TensorFlow/Flax/JAX from English documentation",
484
  "state": "MERGED",
485
- "created": "2025-09-22",
486
  "merged": "2025-09-23",
487
  "first": "2025-09-22",
488
- "last": "2025-09-22"
 
 
 
 
 
 
489
  },
490
  "41076": {
491
  "t": ":rotating_light: [`v5`] Remove headmasking",
492
  "state": "MERGED",
493
- "created": "2025-09-22",
494
  "merged": "2025-09-30",
495
  "first": "2025-09-19",
496
- "last": "2025-09-29"
 
 
 
 
 
 
497
  },
498
  "41111": {
499
  "t": "Support loading LFM2 GGUF",
500
  "state": "MERGED",
501
- "created": "2025-09-23",
502
  "merged": "2025-09-24",
503
  "first": "2025-09-23",
504
- "last": "2025-09-24"
 
 
 
 
 
 
505
  },
506
  "41123": {
507
  "t": "[v5] Remove deprecated prediction loop",
508
  "state": "MERGED",
509
- "created": "2025-09-24",
510
  "merged": "2025-09-30",
511
  "first": "2025-09-24",
512
- "last": "2025-09-30"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
513
  },
514
  "41253": {
515
  "t": "transformers serve quantization docs + some api fixes for bitsandbytes",
516
  "state": "MERGED",
517
- "created": "2025-10-01",
518
  "merged": "2025-10-23",
519
  "first": "2025-10-01",
520
- "last": "2025-10-23"
 
 
 
 
 
 
521
  },
522
  "41289": {
523
  "t": "Add modular detector",
524
  "state": "MERGED",
525
- "created": "2025-10-02",
526
  "merged": "2025-10-03",
527
  "first": "2025-10-02",
528
- "last": "2025-10-03"
 
 
 
 
 
 
529
  },
530
  "41370": {
531
  "t": "[`CB`] Refactors the way we access paged",
532
  "state": "MERGED",
533
- "created": "2025-10-06",
534
  "merged": "2025-10-06",
535
  "first": "2025-10-06",
536
- "last": "2025-10-06"
 
 
 
 
 
 
537
  },
538
  "41388": {
539
  "t": "\ud83d\udea8Default to fast image processors for all models",
540
  "state": "MERGED",
541
- "created": "2025-10-06",
542
  "merged": "2026-01-21",
543
  "first": "2025-10-15",
544
- "last": "2026-01-21"
 
 
 
 
 
 
545
  },
546
  "41446": {
547
  "t": "Enable non-streaming mode in `transformers serve`",
548
  "state": "MERGED",
549
- "created": "2025-10-08",
550
  "merged": "2025-10-15",
551
  "first": "2025-10-14",
552
- "last": "2025-10-14"
 
 
 
 
 
 
553
  },
554
  "41474": {
555
  "t": "\ud83d\udea8 [v5] Toggle the serialization format in processors",
556
  "state": "MERGED",
557
- "created": "2025-10-09",
558
  "merged": "2025-10-16",
559
  "first": "2025-10-09",
560
- "last": "2025-10-15"
 
 
 
 
 
 
561
  },
562
  "41580": {
563
  "t": "Refactor weight loading",
564
  "state": "MERGED",
565
- "created": "2025-10-14",
566
  "merged": "2025-11-13",
567
  "first": "2025-11-03",
568
- "last": "2025-11-13"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
569
  },
570
  "41633": {
571
  "t": "[v5] \ud83d\udea8Refactor subprocessors handling in processors",
572
  "state": "MERGED",
573
- "created": "2025-10-15",
574
  "merged": "2025-11-07",
575
  "first": "2025-10-15",
576
- "last": "2025-11-07"
 
 
 
 
 
 
577
  },
578
  "41655": {
579
  "t": "Switch to CB if cache_implementation == paged",
580
  "state": "MERGED",
581
- "created": "2025-10-16",
582
  "merged": "2025-10-16",
583
  "first": "2025-10-16",
584
- "last": "2025-10-16"
 
 
 
 
 
 
585
  },
586
  "41657": {
587
  "t": "Erroring when KernelConfig is passed without use_kernels = True",
588
  "state": "MERGED",
589
- "created": "2025-10-16",
590
  "merged": "2025-10-16",
591
  "first": "2025-10-16",
592
- "last": "2025-10-16"
 
 
 
 
 
 
593
  },
594
  "41773": {
595
  "t": "Simplify and standardize processor tests",
596
  "state": "MERGED",
597
- "created": "2025-10-21",
598
  "merged": "2025-11-26",
599
  "first": "2025-10-15",
600
- "last": "2025-11-26"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
601
  },
602
  "41916": {
603
  "t": "feat(ci): add continuous batching to benchmarks",
604
  "state": "MERGED",
605
- "created": "2025-10-28",
606
  "merged": "2025-11-07",
607
  "first": "2025-11-07",
608
- "last": "2025-11-07"
 
 
 
 
 
 
609
  },
610
  "41990": {
611
  "t": "feat(kernels): add opt-out flag to disable kernels hub usage through the lib",
612
  "state": "MERGED",
613
- "created": "2025-11-03",
614
  "merged": "2025-11-12",
615
  "first": "2025-11-10",
616
- "last": "2025-11-12"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
617
  },
618
  "42191": {
619
  "t": "Much more efficient and clear weight initialization and tie weights",
620
  "state": "MERGED",
621
- "created": "2025-11-13",
622
  "merged": "2025-11-14",
623
  "first": "2025-11-14",
624
- "last": "2025-11-14"
 
 
 
 
 
 
625
  },
626
  "42254": {
627
  "t": "Update conversion mapping to separate renaming from converting",
628
  "state": "MERGED",
629
- "created": "2025-11-18",
630
  "merged": "2025-11-20",
631
  "first": "2025-11-18",
632
- "last": "2025-11-20"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
633
  },
634
  "42436": {
635
  "t": "Fix mixed torch.Tensor and DTensor in generate when use FSDP2 + LoRA",
636
  "state": "MERGED",
637
- "created": "2025-11-27",
638
  "merged": "2025-12-03",
639
  "first": "2025-12-03",
640
- "last": "2025-12-03"
 
 
 
 
 
 
641
  },
642
  "42634": {
643
  "t": "allow registration of custom checkpoint conversion mappings",
644
  "state": "MERGED",
645
- "created": "2025-12-04",
646
  "merged": "2025-12-11",
647
  "first": "2025-12-10",
648
- "last": "2025-12-11"
 
 
 
 
 
 
649
  },
650
  "42697": {
651
  "t": "batched and grouped experts implementations",
652
  "state": "MERGED",
653
- "created": "2025-12-08",
654
  "merged": "2026-01-05",
655
  "first": "2025-12-06",
656
- "last": "2026-01-01"
 
 
 
 
 
 
657
  },
658
  "42932": {
659
  "t": "[docs] inference engines",
660
  "state": "MERGED",
661
- "created": "2025-12-17",
662
  "merged": "2026-01-07",
663
  "first": "2026-01-07",
664
- "last": "2026-01-07"
 
 
 
 
 
 
665
  },
666
  "43176": {
667
  "t": "[docs] tensorrt-llm",
668
  "state": "MERGED",
669
- "created": "2026-01-08",
670
  "merged": "2026-01-13",
671
  "first": "2026-01-13",
672
- "last": "2026-01-13"
 
 
 
 
 
 
673
  },
674
  "43185": {
675
  "t": "[docs] llama.cpp",
676
  "state": "MERGED",
677
- "created": "2026-01-09",
678
  "merged": "2026-01-09",
679
  "first": "2026-01-09",
680
- "last": "2026-01-09"
 
 
 
 
 
 
681
  },
682
  "43227": {
683
  "t": "GptOss experts implementation",
684
  "state": "MERGED",
685
- "created": "2026-01-12",
686
  "merged": "2026-01-22",
687
  "first": "2026-01-12",
688
- "last": "2026-01-21"
 
 
 
 
 
 
689
  },
690
  "43241": {
691
  "t": "[docs] deploying",
692
  "state": "MERGED",
693
- "created": "2026-01-13",
694
  "merged": "2026-02-05",
695
  "first": "2026-01-30",
696
- "last": "2026-01-30"
 
 
 
 
 
 
697
  },
698
  "43294": {
699
  "t": "[docs] pretraining",
700
  "state": "MERGED",
701
- "created": "2026-01-14",
702
  "merged": "2026-01-20",
703
  "first": "2026-01-14",
704
- "last": "2026-01-20"
 
 
 
 
 
 
705
  },
706
  "43312": {
707
  "t": "[docs] finetuning",
708
  "state": "MERGED",
709
- "created": "2026-01-15",
710
  "merged": "2026-01-20",
711
  "first": "2026-01-20",
712
- "last": "2026-01-20"
 
 
 
 
 
 
713
  },
714
  "43521": {
715
  "t": "[CB] Minor perf improvements and ty compatibility",
716
  "state": "MERGED",
717
- "created": "2026-01-27",
718
  "merged": "2026-01-28",
719
  "first": "2026-01-27",
720
- "last": "2026-01-28"
 
 
 
 
 
 
721
  },
722
  "43807": {
723
  "t": "Refactor trainer init",
724
  "state": "MERGED",
725
- "created": "2026-02-06",
726
  "merged": "2026-02-10",
727
  "first": "2026-02-06",
728
- "last": "2026-02-10"
 
 
 
 
 
 
729
  },
730
  "43917": {
731
  "t": "Model patching API",
732
  "state": "MERGED",
733
- "created": "2026-02-11",
734
  "merged": "2026-03-02",
735
  "first": "2026-02-11",
736
- "last": "2026-02-25"
 
 
 
 
 
 
737
  },
738
  "43977": {
739
  "t": "stable grouped_mm API",
740
  "state": "MERGED",
741
- "created": "2026-02-13",
742
  "merged": "2026-02-16",
743
  "first": "2026-02-13",
744
- "last": "2026-02-16"
 
 
 
 
 
 
745
  },
746
  "44043": {
747
  "t": "`grouped_mm` fallback",
748
  "state": "MERGED",
749
- "created": "2026-02-16",
750
  "merged": "2026-02-18",
751
  "first": "2026-02-16",
752
- "last": "2026-02-17"
 
 
 
 
 
 
753
  },
754
  "44152": {
755
  "t": "AutoGrad support for grouped_mm fallback",
756
  "state": "MERGED",
757
- "created": "2026-02-19",
758
  "merged": "2026-02-20",
759
  "first": "2026-02-19",
760
- "last": "2026-02-20"
 
 
 
 
 
 
761
  },
762
  "44174": {
763
  "t": "Expand model-structure lint rules with a fast AST-based, ruff-like framework",
764
  "state": "MERGED",
765
- "created": "2026-02-20",
766
  "merged": "2026-03-12",
767
  "first": "2026-03-11",
768
- "last": "2026-03-12"
 
 
 
 
 
 
769
  },
770
  "44300": {
771
  "t": "Dynamic weight conversion is recursive",
772
  "state": "MERGED",
773
- "created": "2026-02-26",
774
  "merged": "2026-03-26",
775
  "first": "2026-02-26",
776
- "last": "2026-03-26"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
777
  },
778
  "44469": {
779
  "t": "[remote code/vllm] Fix incorrect tied weights",
780
  "state": "MERGED",
781
- "created": "2026-03-05",
782
  "merged": "2026-03-05",
783
  "first": "2026-03-05",
784
- "last": "2026-03-05"
 
 
 
 
 
 
785
  },
786
  "44558": {
787
  "t": "Add /v1/completions endpoint (OpenAI legacy completions API) to `transformers serve`",
788
  "state": "MERGED",
789
- "created": "2026-03-10",
790
  "merged": "2026-04-22",
791
  "first": "2026-04-22",
792
- "last": "2026-04-22"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
793
  },
794
  "44699": {
795
  "t": "Fix several based models' pipeline parallel support",
796
  "state": "MERGED",
797
- "created": "2026-03-14",
798
  "merged": "2026-03-20",
799
  "first": "2026-03-14",
800
- "last": "2026-03-20"
 
 
 
 
 
 
801
  },
802
  "44939": {
803
  "t": "refactor: mlinter as its own package",
804
  "state": "MERGED",
805
- "created": "2026-03-23",
806
  "merged": "2026-03-24",
807
  "first": "2026-03-23",
808
- "last": "2026-03-23"
 
 
 
 
 
 
809
  },
810
  "44950": {
811
  "t": "\ud83d\udea8 [Cache] Native mamba & hybrid cache",
812
  "state": "MERGED",
813
- "created": "2026-03-23",
814
  "merged": "2026-03-31",
815
  "first": "2026-03-23",
816
- "last": "2026-03-31"
 
 
 
 
 
 
817
  },
818
  "45279": {
819
  "t": "add expert parallelism for gemma-4-26B-A4B-it",
820
  "state": "MERGED",
821
- "created": "2026-04-07",
822
  "merged": "2026-04-22",
823
  "first": "2026-04-21",
824
- "last": "2026-04-22"
 
 
 
 
 
 
825
  },
826
  "45436": {
827
  "t": "Add expert parallelism (EP) config support for Qwen3 MoE",
828
  "state": "MERGED",
829
- "created": "2026-04-14",
830
  "merged": "2026-04-22",
831
  "first": "2026-04-21",
832
- "last": "2026-04-22"
 
 
 
 
 
 
833
  },
834
  "45473": {
835
  "t": "Fix EP: RouterParallel shape, tp_plan property, grouped_mm sentinels",
836
  "state": "MERGED",
837
- "created": "2026-04-16",
838
  "merged": "2026-04-21",
839
  "first": "2026-04-16",
840
- "last": "2026-04-20"
 
 
 
 
 
 
841
  },
842
  "45475": {
843
  "t": "chore(qa): split out mlinter",
844
  "state": "MERGED",
845
- "created": "2026-04-16",
846
  "merged": "2026-04-20",
847
  "first": "2026-04-17",
848
- "last": "2026-04-20"
 
 
 
 
 
 
849
  },
850
  "45585": {
851
  "t": "qa: bumped mlinter and allow local override",
852
  "state": "MERGED",
853
- "created": "2026-04-22",
854
  "merged": "2026-04-23",
855
  "first": "2026-04-23",
856
- "last": "2026-04-23"
 
 
 
 
 
 
857
  },
858
  "45621": {
859
  "t": "Better Grouped GEMM + EP",
860
  "state": "MERGED",
861
- "created": "2026-04-24",
862
  "merged": "2026-05-04",
863
  "first": "2026-04-23",
864
- "last": "2026-05-04"
 
 
 
 
 
 
865
  },
866
  "45634": {
867
  "t": "DeepGEMM BF16 + mixed FP8/FP4 + MegaMoE + refactor",
868
  "state": "MERGED",
869
- "created": "2026-04-24",
870
  "merged": "2026-06-02",
871
  "first": "2026-04-23",
872
- "last": "2026-06-02"
 
 
 
 
 
 
873
  },
874
  "45795": {
875
  "t": "[docs] adding audio/video processors",
876
  "state": "MERGED",
877
- "created": "2026-05-05",
878
  "merged": "2026-05-18",
879
  "first": "2026-05-05",
880
- "last": "2026-05-18"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
881
  },
882
  "46246": {
883
  "t": "[`Revert`] FSDP+Dtensor refactor related changes",
884
  "state": "MERGED",
885
- "created": "2026-05-27",
886
  "merged": "2026-05-28",
887
  "first": "2026-05-27",
888
- "last": "2026-05-27"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
889
  },
890
  "46410": {
891
  "t": "Add workflow to post CI Grafana dashboard link to PR",
892
  "state": "MERGED",
893
- "created": "2026-06-04",
894
  "merged": "2026-06-04",
895
  "first": "2026-06-04",
896
- "last": "2026-06-04"
 
 
 
 
 
 
897
  },
898
  "46412": {
899
  "t": "Improve CI dashboard comment: rename and deduplicate",
900
  "state": "MERGED",
901
- "created": "2026-06-04",
902
  "merged": "2026-06-04",
903
  "first": "2026-06-04",
904
- "last": "2026-06-04"
 
 
 
 
 
 
905
  },
906
  "46415": {
907
  "t": "ci: mention code quality failure in CI dashboard comment",
908
  "state": "MERGED",
909
- "created": "2026-06-04",
910
  "merged": "2026-06-04",
911
  "first": "2026-06-04",
912
- "last": "2026-06-04"
 
 
 
 
 
 
913
  },
914
  "46500": {
915
  "t": "Fix `{image/video/audio}_token_ids` in `ProcessorMixin`",
916
  "state": "MERGED",
917
- "created": "2026-06-08",
918
  "merged": "2026-06-08",
919
  "first": "2026-06-08",
920
- "last": "2026-06-08"
 
 
 
 
 
 
921
  },
922
  "46721": {
923
  "t": "ci: disable CircleCI by replacing config with no-op",
924
  "state": "MERGED",
925
- "created": "2026-06-17",
926
  "merged": "2026-06-17",
927
  "first": "2026-06-17",
928
- "last": "2026-06-17"
 
 
 
 
 
 
929
  },
930
  "46774": {
931
  "t": "Insert a Grafana badge in the PR",
932
  "state": "MERGED",
933
- "created": "2026-06-19",
934
  "merged": "2026-06-28",
935
  "first": "2026-06-28",
936
- "last": "2026-06-28"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
937
  },
938
  "47005": {
939
  "t": "Add tiny_model_id support to ProcessorTesterMixin for memory-sensitive tests",
940
  "state": "MERGED",
941
- "created": "2026-07-01",
942
  "merged": "2026-07-03",
943
  "first": "2026-07-01",
944
- "last": "2026-07-03"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
945
  },
946
  "47172": {
947
  "t": "Switch mlinter to 0.1.2",
948
  "state": "MERGED",
949
- "created": "2026-07-08",
950
  "merged": "2026-07-09",
951
  "first": "2026-07-09",
952
- "last": "2026-07-09"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
953
  },
954
  "47289": {
955
  "t": "Pipeline parallel naive inference",
956
  "state": "MERGED",
957
- "created": "2026-07-13",
958
  "merged": "2026-08-25",
959
  "first": "2026-07-01",
960
- "last": "2026-08-21"
 
 
 
 
 
 
961
  },
962
  "47357": {
963
  "t": "Add FSDP CI and end-to-end FSDP tests + save fsdp",
964
  "state": "MERGED",
965
- "created": "2026-07-16",
966
  "merged": "2026-07-23",
967
  "first": "2026-07-16",
968
- "last": "2026-07-23"
 
 
 
 
 
 
969
  },
970
  "47579": {
971
  "t": "\ud83d\udea8 TP dtensor API inference + training",
972
  "state": "MERGED",
973
- "created": "2026-07-27",
974
  "merged": "2026-08-21",
975
  "first": "2026-06-17",
976
- "last": "2026-08-21"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
977
  },
978
  "47819": {
979
  "t": "Post mlinter findings as inline PR review comments",
980
  "state": "MERGED",
981
- "created": "2026-08-06",
982
  "merged": "2026-08-06",
983
  "first": "2026-08-06",
984
- "last": "2026-08-06"
 
 
 
 
 
 
985
  },
986
  "47863": {
987
  "t": "Bump default flash-attn2 hub kernel version to v3",
988
  "state": "MERGED",
989
- "created": "2026-08-10",
990
  "merged": "2026-08-17",
991
  "first": "2026-08-10",
992
- "last": "2026-08-17"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
993
  },
994
  "47918": {
995
  "t": "Moving mlinter to 0.1.4",
996
  "state": "MERGED",
997
- "created": "2026-08-12",
998
  "merged": "2026-08-18",
999
  "first": "2026-08-17",
1000
- "last": "2026-08-18"
 
 
 
 
 
 
1001
  },
1002
  "48056": {
1003
  "t": "[CircleCI] Enable CI for private forks, no-op for public repo",
1004
  "state": "MERGED",
1005
- "created": "2026-08-18",
1006
  "merged": "2026-08-18",
1007
  "first": "2026-08-18",
1008
- "last": "2026-08-18"
 
 
 
 
 
 
1009
  },
1010
  "48117": {
1011
  "t": "Enable mlinter findings artifact for inline PR reviews",
1012
  "state": "MERGED",
1013
- "created": "2026-08-19",
1014
  "merged": "2026-08-19",
1015
  "first": "2026-08-19",
1016
- "last": "2026-08-19"
 
 
 
 
 
 
1017
  },
1018
  "48167": {
1019
  "t": "Apply context parallelism to the evaluation path",
1020
  "state": "MERGED",
1021
- "created": "2026-08-20",
1022
  "merged": "2026-08-21",
1023
  "first": "2026-08-20",
1024
- "last": "2026-08-20"
 
 
 
 
 
 
1025
  },
1026
  "48271": {
1027
  "t": "draft modular v2 + transformers format",
1028
  "state": "OPEN",
1029
- "created": "2026-08-25",
1030
  "merged": "",
1031
  "first": "2026-08-25",
1032
- "last": "2026-08-26"
1033
- },
1034
- "41897": {
1035
- "t": "[FPQuant] MXFP8 and MXFP4 backwards support",
1036
- "state": "MERGED",
1037
- "created": "2025-10-27",
1038
- "merged": "2025-11-04",
1039
- "first": "2025-10-27",
1040
- "last": "2025-10-28"
1041
  }
1042
  }
 
2
  "28847": {
3
  "t": "Fast image processor",
4
  "state": "MERGED",
 
5
  "merged": "2024-06-11",
6
  "first": "2024-06-10",
7
+ "last": "2024-06-11",
8
+ "by": "amyeroberts",
9
+ "add": 1643,
10
+ "del": 811,
11
+ "files": 64,
12
+ "commits": 40,
13
+ "body": "Adds a ViTImageProcessorFast class, which uses a torchvision backend to speed up image processor import requests from PIL import Image from transformers import AutoImageProcessor # Load the image image = Image.open(requests.get(\"http://images.cocodataset.org/val2017/000000039769.jpg\", stream=True).r"
14
  },
15
  "31247": {
16
  "t": "Add torch.compile Support For Mamba",
17
  "state": "MERGED",
 
18
  "merged": "2024-07-18",
19
  "first": "2024-06-04",
20
+ "last": "2024-07-16",
21
+ "by": "zhenglongjiepheonix",
22
+ "add": 225,
23
+ "del": 85,
24
+ "files": 4,
25
+ "commits": 41,
26
+ "body": "torch.compile support for mamba! Closes #31246"
27
+ },
28
+ "32168": {
29
+ "t": "Make static cache compatible with torch.export",
30
+ "state": "MERGED",
31
+ "merged": "2024-07-29",
32
+ "first": "2024-07-26",
33
+ "last": "2024-07-29",
34
+ "by": "guangy10",
35
+ "add": 80,
36
+ "del": 10,
37
+ "files": 2,
38
+ "commits": 2,
39
+ "body": "This PR is to make the static cache impl compatible with both torch.compile() and torch.export() (new). Making a model exportable so that it can be further lowered and run on edge devices through ExecuTorch, which is a native PyTorch framework for on-device inference. An example of lower Gemma-2b to"
40
  },
41
  "32227": {
42
  "t": ">3-5x faster torch.compile forward compilation for autoregressive decoder models",
43
  "state": "MERGED",
 
44
  "merged": "2024-07-31",
45
  "first": "2024-07-25",
46
+ "last": "2024-07-31",
47
+ "by": "fxmarty",
48
+ "add": 1602,
49
+ "del": 424,
50
+ "files": 22,
51
+ "commits": 19,
52
+ "body": "As detailed in #29114 and #29165, the current implementation of GenerationMixin.generate passes 2D integer tensors of variable length as attention_mask to the models forward, that are transformed into 4D tensors of static shape in the models forward for decoder models. This is an issue when using to"
53
  },
54
  "33248": {
55
  "t": "Modular `transformers`: modularity and inheritance for new model additions",
56
  "state": "MERGED",
 
57
  "merged": "2024-09-24",
58
  "first": "2024-09-02",
59
+ "last": "2024-09-24",
60
+ "by": "ArthurZucker",
61
+ "add": 6504,
62
+ "del": 778,
63
+ "files": 41,
64
+ "commits": 167,
65
+ "body": "Modular transformers: supporting inheritance, while keeping our philosophy This PR introduces Modularity for transformers, which has always been prohibited when working with transformers (see blog post for the accompanying design philosophy). The core idea behind this PR is to facilitate model addit"
66
+ },
67
+ "33707": {
68
+ "t": "Generate using exported model and enable gemma2-2b in ExecuTorch",
69
+ "state": "MERGED",
70
+ "merged": "2024-10-11",
71
+ "first": "2024-10-03",
72
+ "last": "2024-10-11",
73
+ "by": "guangy10",
74
+ "add": 168,
75
+ "del": 0,
76
+ "files": 3,
77
+ "commits": 5,
78
+ "body": "Adding generate support for exported model. Adding gemma2-2b to ExecuTorch with tests. Adding an integration test for gemma-2b that we've enabled already. Additional Test in ExecuTorch Running gemma2-2b E2E: cmake-out/examples/models/llama2/llama_main --tokenizer_path=tokenizer_gemma2.bin --model_pa"
79
  },
80
  "33730": {
81
  "t": "Make audio classification pipeline spec-compliant and add test",
82
  "state": "MERGED",
 
83
  "merged": "2024-09-27",
84
  "first": "2024-09-27",
85
+ "last": "2024-09-27",
86
+ "by": "Rocketknight1",
87
+ "add": 130,
88
+ "del": 3,
89
+ "files": 3,
90
+ "commits": 18,
91
+ "body": "cc @LysandreJik, this test splits the large PR at #33452 into manageable pieces. This first PR will be the largest, since it includes the testing utils, as well as testing compliance for AudioClassificationPipeline. The other single-pipeline PRs will be much smaller, since they can just import the s"
92
  },
93
  "33769": {
94
  "t": "Make ASR pipeline compliant with Hub spec + add tests",
95
  "state": "MERGED",
 
96
  "merged": "2024-10-01",
97
  "first": "2024-09-27",
98
+ "last": "2024-10-01",
99
+ "by": "Rocketknight1",
100
+ "add": 42,
101
+ "del": 9,
102
+ "files": 5,
103
+ "commits": 8,
104
+ "body": "As well as making the ASR pipeline compliant, this PR also refactors out the output test, so we can just import that in future PRs as well."
105
  },
106
  "33771": {
107
  "t": "[`AutoDocstring`] Based on inspect parsing of the signature",
108
  "state": "MERGED",
 
109
  "merged": "2025-05-08",
110
  "first": "2024-10-10",
111
+ "last": "2025-05-08",
112
+ "by": "ArthurZucker",
113
+ "add": 18216,
114
+ "del": 46742,
115
+ "files": 405,
116
+ "commits": 125,
117
+ "body": "Config stay explicit any place that had non explicit (dynamic docstring) gets replaced we can leverage inheritance for forward dynamic typing I think"
118
  },
119
  "33899": {
120
  "t": "Image pipelines spec compliance",
121
  "state": "MERGED",
 
122
  "merged": "2024-10-08",
123
  "first": "2024-10-02",
124
+ "last": "2024-10-08",
125
+ "by": "Rocketknight1",
126
+ "add": 153,
127
+ "del": 44,
128
+ "files": 13,
129
+ "commits": 15,
130
+ "body": "This PR checks for Hub spec compliance for a lot of image pipelines that have very similar inputs. In most cases, no changes are required except deprecating the timeout argument, and renaming the images input to inputs."
131
+ },
132
+ "34101": {
133
+ "t": "Llama3 and Llama2 are ExecuTorch compatible",
134
+ "state": "MERGED",
135
+ "merged": "2024-10-17",
136
+ "first": "2024-10-11",
137
+ "last": "2024-10-11",
138
+ "by": "guangy10",
139
+ "add": 69,
140
+ "del": 0,
141
+ "files": 1,
142
+ "commits": 1,
143
+ "body": "Llama 2&# is compatible with ExecuTorch. Note that LLama2&3 in ExecuTorch repo has been fully optimized for SOTA perf using its own model definition and optimization. You can read details in https://github.com/pytorch/executorch/tree/main/examples/models/llama2. The work here is to make the Llama mo"
144
+ },
145
+ "34102": {
146
+ "t": "Qwen2.5 is ExecuTorch Compatible",
147
+ "state": "MERGED",
148
+ "merged": "2024-10-22",
149
+ "first": "2024-10-11",
150
+ "last": "2024-10-11",
151
+ "by": "guangy10",
152
+ "add": 55,
153
+ "del": 0,
154
+ "files": 1,
155
+ "commits": 1,
156
+ "body": "Qwen2.5 is compatible with ExecuTorch. Before submitting Did you read the contributor guideline, Pull Request section? Was this discussed/approved via a Github issue or the forum? Please add a link to it if that's the case. #33833 Did you write any new necessary tests? Who can review? Anyone in the "
157
  },
158
  "34184": {
159
  "t": "Simplify Tensor Parallel implementation with PyTorch TP",
160
  "state": "MERGED",
 
161
  "merged": "2024-11-18",
162
  "first": "2024-10-15",
163
+ "last": "2024-11-18",
164
+ "by": "kwen2501",
165
+ "add": 357,
166
+ "del": 92,
167
+ "files": 18,
168
+ "commits": 26,
169
+ "body": "This PR uses the torch.distributed.tensor.parallel subpackage to implement Tensor Parallel for Llama (as an example). The motivation is multi-fold: to make modeling code simple as single-worker case: all manual TP implementations under if self.config.pretraining_tp > 1 can be removed. to make tensor"
170
  },
171
  "34194": {
172
  "t": "feat: add support for tensor parallel training workflow with accelerate",
173
  "state": "MERGED",
 
174
  "merged": "2025-02-18",
175
  "first": "2025-02-18",
176
+ "last": "2025-02-18",
177
+ "by": "kmehant",
178
+ "add": 133,
179
+ "del": 6,
180
+ "files": 8,
181
+ "commits": 8,
182
+ "body": "1. Add apply_tensor_parallel API to apply TP plan to Llama and Granite models Already merged in to HF/transformers. 2. Introduce tp_size user facing argument to be further consumed by accelerate (see huggingface/accelerate#3173) 3. Allows for e2e TP training. Please review in conjunction with huggin"
183
  },
184
  "34288": {
185
  "t": "Sync video classification pipeline with huggingface_hub spec",
186
  "state": "MERGED",
 
187
  "merged": "2024-10-22",
188
  "first": "2024-10-21",
189
+ "last": "2024-10-22",
190
+ "by": "Rocketknight1",
191
+ "add": 61,
192
+ "del": 7,
193
+ "files": 3,
194
+ "commits": 2,
195
+ "body": ""
196
  },
197
  "34487": {
198
  "t": "Large modular logic refactoring",
199
  "state": "MERGED",
 
200
  "merged": "2024-11-01",
201
  "first": "2024-10-31",
202
+ "last": "2024-11-01",
203
+ "by": "Cyrilvallez",
204
+ "add": 2668,
205
+ "del": 1600,
206
+ "files": 19,
207
+ "commits": 40,
208
+ "body": "This PR largely rework the logic we use in the modular converter. It is (hopefully) clearer and maintainable. Instead of going in all directions, adding stuff, then deleting it if not needed, we now do the following: visit all the modular file (record imports/functions/classes/assignments nodes) cre"
209
  },
210
  "34544": {
211
  "t": "Test: generate with `torch.compile(model.forward)` as a fast test",
212
  "state": "MERGED",
 
213
  "merged": "2025-01-28",
214
  "first": "2024-10-31",
215
+ "last": "2025-01-28",
216
+ "by": "gante",
217
+ "add": 105,
218
+ "del": 53,
219
+ "files": 25,
220
+ "commits": 13,
221
+ "body": "Follow-up to #34464 This PR: Converts test_generate_compile_model_forward to a fast test. This means we will check generate with torch.compile(model.forward) at each commit on ALL models that support StaticCache \ud83d\udc9b Fixes failing cases of test_generate_compile_model_forward whenever possible Tags mode"
222
  },
223
  "34564": {
224
  "t": "Add TimmWrapper",
225
  "state": "MERGED",
 
226
  "merged": "2024-12-11",
227
  "first": "2024-11-01",
228
+ "last": "2024-12-11",
229
+ "by": "qubvel",
230
+ "add": 1432,
231
+ "del": 129,
232
+ "files": 25,
233
+ "commits": 92,
234
+ "body": "Adds a TimmWrapper set of classes such that timm models can be loaded in as transformer models into the library. Continue of #33687 General Usage import torch from urllib.request import urlopen from PIL import Image from transformers import AutoConfig, AutoModelForImageClassification, AutoImageProce"
235
  },
236
  "34729": {
237
  "t": "feat: allow to use hf-models",
238
  "state": "MERGED",
 
239
  "merged": "2024-11-19",
240
  "first": "2024-11-14",
241
+ "last": "2024-11-14",
242
+ "by": "cgebbe",
243
+ "add": 0,
244
+ "del": 5,
245
+ "files": 1,
246
+ "commits": 1,
247
+ "body": "Problem: This doesn't currently work, but should in my opinion? import transformers model = transformers.AutoBackbone.from_pretrained( \"hf-hub:bioptimus/H-optimus-0\", use_timm_backbone=True, use_pretrained_backbone=True, ) Solution approach: Just remove the two lines which explicitly check whether t"
248
  },
249
  "34800": {
250
  "t": "protect tensor parallel usage",
251
  "state": "MERGED",
 
252
  "merged": "2024-11-19",
253
  "first": "2024-11-19",
254
+ "last": "2024-11-19",
255
+ "by": "ArthurZucker",
256
+ "add": 11,
257
+ "del": 5,
258
+ "files": 2,
259
+ "commits": 1,
260
+ "body": "Fixes #34795"
261
  },
262
  "34896": {
263
  "t": "[`GPTNeoX`] Flex Attention + Refactor",
264
  "state": "MERGED",
 
265
  "merged": "2024-12-04",
266
  "first": "2024-11-23",
267
+ "last": "2024-12-02",
268
+ "by": "vasqu",
269
+ "add": 371,
270
+ "del": 249,
271
+ "files": 6,
272
+ "commits": 20,
273
+ "body": "Adds flex attention and the refactor according to #34809 However, I discovered several issues in the current version of gemma2 (#34282): It seems like that flex attention needs a transpose afterwards like sdpa Loading flex attn with from pretrained didn't work and hence, current tests use another at"
274
+ },
275
+ "35124": {
276
+ "t": "Add common test for `torch.export` and fix some vision models",
277
+ "state": "MERGED",
278
+ "merged": "2025-02-11",
279
+ "first": "2024-12-05",
280
+ "last": "2025-02-11",
281
+ "by": "qubvel",
282
+ "add": 305,
283
+ "del": 151,
284
+ "files": 77,
285
+ "commits": 46,
286
+ "body": "Add a common slow test to check if a model can be exported with no issues using torch.export.export Add an optional test, to enable it please set test_torch_exportable = True flag for model-specific test. Enable test for vision and video models Fix most of the vision models The main fixes include: U"
287
  },
288
  "35184": {
289
  "t": "Add compile test for fast image processor",
290
  "state": "MERGED",
 
291
  "merged": "2024-12-23",
292
  "first": "2024-12-11",
293
+ "last": "2024-12-18",
294
+ "by": "yonigozlan",
295
+ "add": 63,
296
+ "del": 4,
297
+ "files": 3,
298
+ "commits": 4,
299
+ "body": "Add a test in ImageProcessingTestMixin, which is only ran on fast image processors. The test compiles a fast image processor, run a processing pass and compares the result with an eager image processor. Who can review?"
300
  },
301
  "35187": {
302
  "t": "make LlamaModel._update_causal_mask torch compilable",
303
  "state": "MERGED",
 
304
  "merged": "2024-12-23",
305
  "first": "2024-12-23",
306
+ "last": "2024-12-23",
307
+ "by": "winglian",
308
+ "add": 33,
309
+ "del": 33,
310
+ "files": 33,
311
+ "commits": 3,
312
+ "body": "This doesn't really have a performance improvement, but at least keeps torch.compile from having a graph break here. Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull Request section? Was th"
313
  },
314
  "35235": {
315
  "t": "\ud83d\udea8All attention refactor\ud83d\udea8",
316
  "state": "MERGED",
 
317
  "merged": "2024-12-18",
318
  "first": "2024-12-16",
319
+ "last": "2024-12-18",
320
+ "by": "ArthurZucker",
321
+ "add": 5635,
322
+ "del": 9778,
323
+ "files": 107,
324
+ "commits": 99,
325
+ "body": "Todo in this PR: Cohere Chameleon DBRX Gemma Gemma2 GLM (modular donc rien \u00e0 faire je crois) gpt_neoX et GPT2 Granite Jamba JetMoe Mimi Mistral Mixtral Mllama Moshi Nemotron OPT Phi Ph3 PhiMoe Qwen2 qwen2Moe qwen2VL SableML StartCoder2 -> Modular normalement oK Idefics1,2,3 Olmo Olmo2 Siglip Whisper"
326
  },
327
  "35342": {
328
  "t": "Cleaner attention interfaces",
329
  "state": "MERGED",
 
330
  "merged": "2024-12-20",
331
  "first": "2024-12-19",
332
+ "last": "2024-12-20",
333
+ "by": "Cyrilvallez",
334
+ "add": 30,
335
+ "del": 12,
336
+ "files": 4,
337
+ "commits": 5,
338
+ "body": "Improve the attention interfaces with comments/dtype fix"
339
  },
340
  "35567": {
341
  "t": "Trainer Refactor: Part 1",
342
  "state": "MERGED",
 
343
  "merged": "2025-01-29",
344
  "first": "2025-01-07",
345
+ "last": "2025-01-23",
346
+ "by": "muellerzr",
347
+ "add": 147,
348
+ "del": 132,
349
+ "files": 3,
350
+ "commits": 15,
351
+ "body": "This is the first chunk in my attempt to clean up the trainer without breaking core capabilities (and integration capabilities/projects). The aim is for readability rather than enhancement of features. I've started this process by: Refactoring how we get the \"core\" training values (max_steps, update"
352
  },
353
  "35658": {
354
  "t": "[`Compile`] Only test compiling model forward pass",
355
  "state": "MERGED",
 
356
  "merged": "2025-01-13",
357
  "first": "2025-01-13",
358
+ "last": "2025-01-13",
359
+ "by": "ArthurZucker",
360
+ "add": 9,
361
+ "del": 34,
362
+ "files": 7,
363
+ "commits": 2,
364
+ "body": "Instead of testing the full generate compile."
365
  },
366
  "35794": {
367
  "t": "Auto-add `timm` tag to timm-wrapper models.",
368
  "state": "MERGED",
 
369
  "merged": "2025-01-21",
370
  "first": "2025-01-20",
371
+ "last": "2025-01-20",
372
+ "by": "pcuenca",
373
+ "add": 1,
374
+ "del": 0,
375
+ "files": 1,
376
+ "commits": 1,
377
+ "body": "Works for fine-tuned or exported models: from transformers import AutoModelForImageClassification checkpoint = \"timm/vit_base_patch16_224.augreg2_in21k_ft_in1k\" model = AutoModelForImageClassification.from_pretrained(checkpoint) model.push_to_hub(\"pcuenq/tw1\") The uploaded model will now show snippe"
378
  },
379
  "36091": {
380
  "t": "Add pipeline parallel plan to `PretrainedConfig` and `PreTrainedModel`",
381
  "state": "MERGED",
 
382
  "merged": "2025-02-12",
383
  "first": "2025-02-07",
384
+ "last": "2025-02-10",
385
+ "by": "hmellor",
386
+ "add": 188,
387
+ "del": 0,
388
+ "files": 50,
389
+ "commits": 25,
390
+ "body": "Fixes # (issue) Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull Request section? Was this discussed/approved via a Github issue or the forum? Please add a link to it if that's the case. Di"
391
+ },
392
+ "36486": {
393
+ "t": "Export T5 (encoder-decoder) to ExecuTorch",
394
+ "state": "MERGED",
395
+ "merged": "2025-03-31",
396
+ "first": "2025-03-06",
397
+ "last": "2025-03-06",
398
+ "by": "guangy10",
399
+ "add": 324,
400
+ "del": 0,
401
+ "files": 2,
402
+ "commits": 1,
403
+ "body": "This PR enables exporting T5 model to ExecuTorch, which has been asked by many OSS users. We will need to export T5 encoder and decoder separately (i.e. to separate .pte files when lowering to ExecuTorch), and compose the encoder-decoder for specific task (e.g. summarization) in the ExecuTorch runti"
404
  },
405
  "36835": {
406
  "t": "Enhance Model Loading By Providing Parallelism, Uses Optional Env Flag",
407
  "state": "MERGED",
 
408
  "merged": "2025-05-23",
409
  "first": "2025-03-19",
410
+ "last": "2025-05-23",
411
+ "by": "inf3rnus",
412
+ "add": 234,
413
+ "del": 76,
414
+ "files": 4,
415
+ "commits": 34,
416
+ "body": "modeling_utils.py has been modified to allow for the parallelized loading of weights. Two env vars have been introduced, HF_ENABLE_PARALLEL_LOADING which allows weights to be loaded in a multiprocessing pool and HF_PARALLEL_LOADING_WORKERS which specifies how many child processes you want to spawn t"
417
  },
418
  "36853": {
419
  "t": "Use `deformable_detr` kernel from the Hub",
420
  "state": "MERGED",
 
421
  "merged": "2025-03-21",
422
  "first": "2025-03-21",
423
+ "last": "2025-03-21",
424
+ "by": "danieldk",
425
+ "add": 405,
426
+ "del": 3834,
427
+ "files": 21,
428
+ "commits": 3,
429
+ "body": "Remove the deformable_detr kernel from kernels/ and use the pre-built kernel from the Hub instead. Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull Request section? Was this discussed/appro"
430
  },
431
  "36934": {
432
  "t": "Aligning modling code for GPT2 to work with vLLM (fallback)",
433
  "state": "MERGED",
 
434
  "merged": "2025-05-02",
435
  "first": "2025-03-24",
436
+ "last": "2025-05-01",
437
+ "by": "ariG23498",
438
+ "add": 13,
439
+ "del": 0,
440
+ "files": 3,
441
+ "commits": 23,
442
+ "body": "This PR changes the modeling code for GPT2 to support working on vLLM using the transformers fallback backend. The changes are as follows: Introduction of kwargs which is used to propagate information about attention_indices in vLLM Adding a base_model_tp_plan, which is currently empty. This is a re"
443
+ },
444
+ "37124": {
445
+ "t": "Make canine model exportable by removing unncessary complicated logic",
446
+ "state": "MERGED",
447
+ "merged": "2025-04-01",
448
+ "first": "2025-03-31",
449
+ "last": "2025-03-31",
450
+ "by": "tugsbayasgalan",
451
+ "add": 2,
452
+ "del": 2,
453
+ "files": 1,
454
+ "commits": 2,
455
+ "body": "In this code, char_seq_length is guaranteed to be an integer, so there is no need to cast it to tensor. By removing this, canine model is now exportable. Fixes # (issue) Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read th"
456
  },
457
  "37376": {
458
  "t": "Add moe kernels",
459
  "state": "MERGED",
 
460
  "merged": "2025-04-11",
461
  "first": "2025-04-08",
462
+ "last": "2025-04-11",
463
+ "by": "ArthurZucker",
464
+ "add": 9,
465
+ "del": 0,
466
+ "files": 2,
467
+ "commits": 10,
468
+ "body": "Add MoE kernels"
469
  },
470
  "37381": {
471
  "t": "Fix mask handling for flex attention in llama/gemma2/mistral/qwen2",
472
  "state": "MERGED",
 
473
  "merged": "2025-04-14",
474
  "first": "2025-04-08",
475
+ "last": "2025-04-12",
476
+ "by": "flukeskywalker",
477
+ "add": 423,
478
+ "del": 177,
479
+ "files": 59,
480
+ "commits": 4,
481
+ "body": "Issue In gemma2 and mistral modelling code, where the _update_causal_mask() methods were not handling the case of the atn_implementation being the recently added \"flex_attention\". Main Fix For Mistral and Gemma2: Added logic for converting attention mask passed as as a tensor to a BlockMask required"
482
+ },
483
+ "37728": {
484
+ "t": "Gemma3 is Torch Exportable",
485
+ "state": "MERGED",
486
+ "merged": "2025-04-28",
487
+ "first": "2025-04-25",
488
+ "last": "2025-04-25",
489
+ "by": "guangy10",
490
+ "add": 369,
491
+ "del": 10,
492
+ "files": 9,
493
+ "commits": 2,
494
+ "body": "Initial effort to add torch.expport support for the Gemma3 model! Gemma3 provides a 1b variant that is suitable for ExecuTorch to bring it for on-device use-case. This PR is focusing on creating the export recipe and validate the exported model can produce same output as eager. Expand support to oth"
495
  },
496
  "37877": {
497
  "t": "parallelism goes brrr",
498
  "state": "MERGED",
 
499
  "merged": "2025-05-20",
500
  "first": "2025-04-29",
501
+ "last": "2025-05-20",
502
+ "by": "nmntz",
503
+ "add": 1515,
504
+ "del": 136,
505
+ "files": 6,
506
+ "commits": 59,
507
+ "body": "Fixes # (issue) Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull Request section? Was this discussed/approved via a Github issue or the forum? Please add a link to it if that's the case. Di"
508
  },
509
  "37915": {
510
  "t": "[transformers x vLLM] standardize processors",
511
  "state": "MERGED",
 
512
  "merged": "2025-05-27",
513
  "first": "2025-05-01",
514
+ "last": "2025-05-26",
515
+ "by": "zucchini-nlp",
516
+ "add": 1167,
517
+ "del": 79,
518
+ "files": 28,
519
+ "commits": 19,
520
+ "body": "Part of #37780. The design was tested on different model types: Oke, I verified that the inference works for all models, unless I forgot about some new ones. Here is the list I tested. A few models (blip2, gotOcr, gemma3) won't be supported in the first release. Gemma3 is already planned after we me"
521
  },
522
  "38090": {
523
  "t": "Refactor `get_XXX_dataloader` from Trainer",
524
  "state": "MERGED",
 
525
  "merged": "2025-05-19",
526
  "first": "2025-05-12",
527
+ "last": "2025-05-13",
528
+ "by": "yaswanth19",
529
+ "add": 72,
530
+ "del": 80,
531
+ "files": 1,
532
+ "commits": 3,
533
+ "body": "Refactors dataloader functions in Trainer"
534
  },
535
  "38108": {
536
  "t": "\ud83d\udd34\ud83d\udd34\ud83d\udd34 [`Attention`] Refactor Attention Interface for Bart-based Models",
537
  "state": "MERGED",
 
538
  "merged": "2025-05-22",
539
  "first": "2025-05-12",
540
+ "last": "2025-05-22",
541
+ "by": "vasqu",
542
+ "add": 8035,
543
+ "del": 5932,
544
+ "files": 75,
545
+ "commits": 71,
546
+ "body": "This PR is gonna tackle two things in general: Flex Attention for all base attention types (encoder, decoder, encoder-decoder cross) New Attention Interface for a bunch of models (mostly based on Bart's implementation) As a bonus, some models have already been refactored into modular if seen fit, e."
547
  },
548
  "38205": {
549
  "t": "Add kernelize to transformers",
550
  "state": "MERGED",
 
551
  "merged": "2025-06-24",
552
  "first": "2025-06-24",
553
+ "last": "2025-06-24",
554
+ "by": "MekkCyber",
555
+ "add": 13,
556
+ "del": 43,
557
+ "files": 4,
558
+ "commits": 11,
559
+ "body": "Instead of dynamically switching the forward methods using a decorator, we are exploring a new approach that performs this replacement statically within modeling_utils.py. This allows us to modify the forward methods at load time, which makes the kernels compile compatible. Also there is no need to "
560
  },
561
  "38242": {
562
  "t": "Add support for auto_docstring with model outputs",
563
  "state": "MERGED",
 
564
  "merged": "2025-06-23",
565
  "first": "2025-05-20",
566
+ "last": "2025-06-23",
567
+ "by": "yonigozlan",
568
+ "add": 6972,
569
+ "del": 8993,
570
+ "files": 177,
571
+ "commits": 21,
572
+ "body": "As the title says. However not sure if worth it, as there are a lot of different descriptions currently for the same args. The descriptions that are different from the one in ModelOutputArgs are left in the ModelOutput docstring, which is why the there's not that much more deletions than additions i"
573
  },
574
  "38262": {
575
  "t": "Protect ParallelInterface",
576
  "state": "MERGED",
 
577
  "merged": "2025-05-21",
578
  "first": "2025-05-21",
579
+ "last": "2025-05-21",
580
+ "by": "ArthurZucker",
581
+ "add": 22,
582
+ "del": 17,
583
+ "files": 3,
584
+ "commits": 2,
585
+ "body": ""
586
  },
587
  "38443": {
588
  "t": "Split `transformers chat` and `transformers serve` ",
589
  "state": "MERGED",
 
590
  "merged": "2025-06-30",
591
  "first": "2025-06-24",
592
+ "last": "2025-06-30",
593
+ "by": "LysandreJik",
594
+ "add": 915,
595
+ "del": 310,
596
+ "files": 9,
597
+ "commits": 25,
598
+ "body": "This PR splits the transformers chat frontend from its backend (now moved to transformers serve). We take this opportunity to have transformers serve act as a server with an OpenAI-compatible API, via http, using SSE to stream tokens. You can take it for a spin either by spawning a server: transform"
599
  },
600
  "38645": {
601
  "t": "support loading qwen3 gguf",
602
  "state": "MERGED",
 
603
  "merged": "2025-07-15",
604
  "first": "2025-06-06",
605
+ "last": "2025-06-25",
606
+ "by": "44670",
607
+ "add": 31,
608
+ "del": 0,
609
+ "files": 2,
610
+ "commits": 6,
611
+ "body": "This PR adds gguf loading support for qwen3 models. Before submitting [No] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). [Yes] Did you read the contributor guideline, Pull Request section? [No afaik] Was this discussed/approved via a Github issue or"
612
+ },
613
+ "38646": {
614
+ "t": "Unbreak optimum-executorch",
615
+ "state": "MERGED",
616
+ "merged": "2025-06-13",
617
+ "first": "2025-06-10",
618
+ "last": "2025-06-12",
619
+ "by": "guangy10",
620
+ "add": 64,
621
+ "del": 39,
622
+ "files": 9,
623
+ "commits": 3,
624
+ "body": "Revert minimal changes made from #37866 that breaks export to ExecuTorch in huggingface/optimum-executorch when developing from latest transformers trunk TODO: Will update with tests shortly Before submitting Did you read the contributor guideline, Pull Request section? Was this discussed/approved v"
625
  },
626
  "38760": {
627
  "t": "Prepare for TF+Jax deprecation",
628
  "state": "MERGED",
 
629
  "merged": "2025-06-11",
630
  "first": "2025-06-11",
631
+ "last": "2025-06-11",
632
+ "by": "Rocketknight1",
633
+ "add": 4,
634
+ "del": 37,
635
+ "files": 2,
636
+ "commits": 2,
637
+ "body": "This PR removes a lot of TF + Jax testing in repo-consistency and .circleci. In a follow-up PR, I'll delete a lot of the test files, but this needs to be in a separate PR because repo-consistency checks run against the target branch, not the PR branch, and so they will fail if we remove other tests "
638
  },
639
  "38814": {
640
  "t": "\ud83c\udfaf Trackio integration",
641
  "state": "MERGED",
 
642
  "merged": "2025-07-22",
643
  "first": "2025-06-13",
644
+ "last": "2025-07-22",
645
+ "by": "qgallouedec",
646
+ "add": 150,
647
+ "del": 9,
648
+ "files": 7,
649
+ "commits": 13,
650
+ "body": "from trl import SFTTrainer, SFTConfig from datasets import load_dataset trainer = SFTTrainer( model=\"Qwen/Qwen2.5-0.5B\", args=SFTConfig(report_to=\"trackio\"), train_dataset=load_dataset(\"trl-lib/Capybara\", split=\"train\"), ) trainer.train() Before submitting This PR fixes a typo or improves the docs ("
651
  },
652
  "39011": {
653
  "t": "[docs] @auto_docstring",
654
  "state": "MERGED",
 
655
  "merged": "2025-06-26",
656
  "first": "2025-06-26",
657
+ "last": "2025-06-26",
658
+ "by": "stevhliu",
659
+ "add": 107,
660
+ "del": 103,
661
+ "files": 3,
662
+ "commits": 2,
663
+ "body": "Refactors the @auto_docstring docs: adds a link to it from the modular_transformers.md doc to help with documentation when contributing a model reorganizes sections to start with practical examples of how to use the decorator, documenting arguments, and checking the docstrings first remove the emoji"
664
  },
665
  "39155": {
666
  "t": "Responses API in `transformers serve`",
667
  "state": "MERGED",
 
668
  "merged": "2025-07-16",
669
  "first": "2025-07-03",
670
+ "last": "2025-07-16",
671
+ "by": "LysandreJik",
672
+ "add": 948,
673
+ "del": 391,
674
+ "files": 8,
675
+ "commits": 10,
676
+ "body": "This PR offers an initial scaffolding of transformers serve so that it has both /v1/chat/completions and /v1/responses endpoints. So far, it only implements a streaming version of both endpoints. To implement the responses endpoint, I'm basing myself off of the \"Streaming\" tab in the OpenAPI API ref"
677
  },
678
  "39198": {
679
  "t": "CI workflow for performed test regressions",
680
  "state": "MERGED",
 
681
  "merged": "2025-07-16",
682
  "first": "2025-07-03",
683
+ "last": "2025-07-15",
684
+ "by": "ahadnagy",
685
+ "add": 183,
686
+ "del": 13,
687
+ "files": 3,
688
+ "commits": 5,
689
+ "body": "This PR adds a new workflow to the CI that dumps a report about the differences in performed tests. We noticed that the test counts sometimes fluctuate quite a bit between CI runs, and this report will help in narrowing down the root cause and find regressions (removed test, catastrophic failure dur"
690
  },
691
  "39454": {
692
  "t": "Transformers serve VLM",
693
  "state": "MERGED",
 
694
  "merged": "2025-07-23",
695
  "first": "2025-07-18",
696
+ "last": "2025-07-23",
697
+ "by": "LysandreJik",
698
+ "add": 268,
699
+ "del": 75,
700
+ "files": 2,
701
+ "commits": 8,
702
+ "body": "Add VLM support for transformers serve. In case you're interested in playing around with it, Open WebUI is a good tool to work with. Launching transformers serve with VLM support: cd <PATH_TO_TRANSFORMERS> git checkout transformers-serve-vlm transformers serve --enable_cors To install Open WebUI: pi"
703
  },
704
  "39457": {
705
  "t": "[Paged-Attention] Handle continuous batching for repetition penalty",
706
  "state": "MERGED",
 
707
  "merged": "2025-07-22",
708
  "first": "2025-07-16",
709
+ "last": "2025-07-22",
710
+ "by": "kashif",
711
+ "add": 80,
712
+ "del": 5,
713
+ "files": 3,
714
+ "commits": 9,
715
+ "body": "We handle the repetition penalty for continuous batching Fixes # (issue) Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull Request section? Was this discussed/approved via a Github issue or "
716
  },
717
  "39501": {
718
  "t": "Add ep",
719
  "state": "MERGED",
 
720
  "merged": "2025-07-25",
721
  "first": "2025-07-17",
722
+ "last": "2025-07-25",
723
+ "by": "ArthurZucker",
724
+ "add": 436,
725
+ "del": 186,
726
+ "files": 9,
727
+ "commits": 25,
728
+ "body": "Add support for expert parallel!"
729
  },
730
  "39562": {
731
  "t": "Fixes needed for n-d parallelism and TP",
732
  "state": "MERGED",
 
733
  "merged": "2025-07-22",
734
  "first": "2025-07-21",
735
+ "last": "2025-07-22",
736
+ "by": "winglian",
737
+ "add": 3,
738
+ "del": 3,
739
+ "files": 1,
740
+ "commits": 2,
741
+ "body": "related: #38949 Fixes # (issue) Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull Request section? Was this discussed/approved via a Github issue or the forum? Please add a link to it if tha"
742
  },
743
  "39583": {
744
  "t": "Update recent processors for vLLM backend",
745
  "state": "MERGED",
 
746
  "merged": "2025-07-24",
747
  "first": "2025-07-22",
748
+ "last": "2025-07-22",
749
+ "by": "zucchini-nlp",
750
+ "add": 79,
751
+ "del": 9,
752
+ "files": 6,
753
+ "commits": 2,
754
+ "body": "As peer title, models needed tiny changes here and there"
755
  },
756
  "39595": {
757
  "t": "[Trackio] Allow single-gpu training and monitor power",
758
  "state": "MERGED",
 
759
  "merged": "2025-07-23",
760
  "first": "2025-07-23",
761
+ "last": "2025-07-23",
762
+ "by": "qgallouedec",
763
+ "add": 9,
764
+ "del": 6,
765
+ "files": 1,
766
+ "commits": 1,
767
+ "body": ""
768
  },
769
  "39638": {
770
  "t": "Support loading Qwen3 MoE GGUF",
771
  "state": "MERGED",
 
772
  "merged": "2025-07-29",
773
  "first": "2025-07-24",
774
+ "last": "2025-07-29",
775
+ "by": "ctcanbol",
776
+ "add": 31,
777
+ "del": 0,
778
+ "files": 2,
779
+ "commits": 7,
780
+ "body": "Currently, GGUF versions of Qwen3 MoE models raises \"GGUF model with architecture qwen3moe is not supported yet\" error. This PR resolves this issue. Fixes #39721 Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contri"
781
  },
782
  "39768": {
783
  "t": "Benchmarking improvements",
784
  "state": "MERGED",
 
785
  "merged": "2025-08-15",
786
  "first": "2025-07-27",
787
+ "last": "2025-08-15",
788
+ "by": "ahadnagy",
789
+ "add": 719,
790
+ "del": 435,
791
+ "files": 7,
792
+ "commits": 7,
793
+ "body": "The goal of this PR is to start improving the benchmarking infrastructure of Transformers. In the short-term we'd like to build a dataset similar to what Diffusers did (https://huggingface.co/datasets/diffusers/benchmarks) and put a dashboard on Spaces. This PR starts with the data acquisition part "
794
+ },
795
+ "39836": {
796
+ "t": "Support input_embeds in torch exportable decoders",
797
+ "state": "MERGED",
798
+ "merged": "2025-08-07",
799
+ "first": "2025-08-01",
800
+ "last": "2025-08-05",
801
+ "by": "jackzhxng",
802
+ "add": 324,
803
+ "del": 84,
804
+ "files": 11,
805
+ "commits": 13,
806
+ "body": "Allows specifying inputs_embeds in TorchExportableModule's in order to support export of multimodal model's text decoders. Adds config and generation_config to the constructors to support multimodal models since the TorchExportableModule will wrap the nested text decoder model, which doesn't have it"
807
  },
808
  "40426": {
809
  "t": "Continuous batching refactor",
810
  "state": "MERGED",
 
811
  "merged": "2025-08-26",
812
  "first": "2025-08-26",
813
+ "last": "2025-08-26",
814
+ "by": "remi-or",
815
+ "add": 2174,
816
+ "del": 1547,
817
+ "files": 10,
818
+ "commits": 27,
819
+ "body": "This PR refactors the files related to continuous batching and changes a few parts of the logic: the way the max_batch_tokens and num_blocks is computed; the static inputs of the CB are now sliced to avoid passing an arbitrary amount of padding in the forward the example file for CB has been reworke"
820
  },
821
  "40619": {
822
  "t": "[CP] Add attention_mask to the buffer when the mask is causal ",
823
  "state": "MERGED",
 
824
  "merged": "2025-09-03",
825
  "first": "2025-09-02",
826
+ "last": "2025-09-03",
827
+ "by": "kashif",
828
+ "add": 25,
829
+ "del": 14,
830
+ "files": 1,
831
+ "commits": 3,
832
+ "body": "Ensured that the attention_mask is always checked for causality before being added to the buffer, and that this validation is performed only once for performance. Modified the logic so that the attention_mask is appended to buffers only after successful validation, preventing non-causal masks from b"
833
  },
834
  "40654": {
835
  "t": "Remove TF/Flax examples",
836
  "state": "MERGED",
 
837
  "merged": "2025-09-03",
838
  "first": "2025-09-03",
839
+ "last": "2025-09-03",
840
+ "by": "Rocketknight1",
841
+ "add": 1,
842
+ "del": 21961,
843
+ "files": 74,
844
+ "commits": 3,
845
+ "body": "As the title says, the frameworks have been deprecated, so we should probably drop the examples as well! One thing to note: check_copies.py had a check_full_copies function that checked when files should be exact copies of each other. This was only used for TF/Flax examples. I've removed the FULL_CO"
846
  },
847
  "40760": {
848
  "t": "\ud83d\udea8\ud83d\udea8\ud83d\udea8 Fully remove Tensorflow and Jax support library-wide",
849
  "state": "MERGED",
 
850
  "merged": "2025-09-18",
851
  "first": "2025-09-18",
852
+ "last": "2025-09-18",
853
+ "by": "Cyrilvallez",
854
+ "add": 3792,
855
+ "del": 181269,
856
+ "files": 854,
857
+ "commits": 35,
858
+ "body": "Apart from obvious tf/jax support, I believe the following should be the only potential breaking changes to torch-only code: pipelines do not take framework argument anymore onnx config methods do not take framework argument anymore It may break current torch code if users do framework=\"pt\" explicit"
859
  },
860
  "40936": {
861
  "t": "rm slow tokenizers",
862
  "state": "MERGED",
 
863
  "merged": "2025-11-27",
864
  "first": "2025-10-10",
865
+ "last": "2025-11-27",
866
+ "by": "itazap",
867
+ "add": 16908,
868
+ "del": 55397,
869
+ "files": 443,
870
+ "commits": 277,
871
+ "body": "Tokenization Just as we moved towards a single backend library for model definition, we want Tokenizer to be a lot more intuitive. With v5, you can now initialize an empty LlamaTokenizer and train it directly on your new task! Defining a new tokenizer object should be as simple as this: from transfo"
872
  },
873
  "40997": {
874
  "t": "\ud83d\udea8 [v5] remove deprecated entry point",
875
  "state": "MERGED",
 
876
  "merged": "2025-09-19",
877
  "first": "2025-09-19",
878
+ "last": "2025-09-19",
879
+ "by": "gante",
880
+ "add": 99,
881
+ "del": 111,
882
+ "files": 30,
883
+ "commits": 3,
884
+ "body": "Removes the transformers-cli entrypoint, which is deprecated in favor of transformers"
885
  },
886
  "40998": {
887
  "t": "\ud83d\udea8 [v5] remove generate output retrocompatibility aliases",
888
  "state": "MERGED",
 
889
  "merged": "2025-09-19",
890
  "first": "2025-09-19",
891
+ "last": "2025-09-19",
892
+ "by": "gante",
893
+ "add": 0,
894
+ "del": 66,
895
+ "files": 3,
896
+ "commits": 2,
897
+ "body": "See title :)"
898
  },
899
  "41017": {
900
  "t": "\ud83d\udea8 [V5] Remove deprecated training arguments ",
901
  "state": "MERGED",
 
902
  "merged": "2025-09-24",
903
  "first": "2025-09-20",
904
+ "last": "2025-09-20",
905
+ "by": "cyyever",
906
+ "add": 12,
907
+ "del": 234,
908
+ "files": 5,
909
+ "commits": 4,
910
+ "body": "Remove deprecated training arguments from TrainingArguments"
911
  },
912
  "41058": {
913
  "t": "Remove mention of TensorFlow/Flax/JAX from English documentation",
914
  "state": "MERGED",
 
915
  "merged": "2025-09-23",
916
  "first": "2025-09-22",
917
+ "last": "2025-09-22",
918
+ "by": "cyyever",
919
+ "add": 3,
920
+ "del": 4,
921
+ "files": 2,
922
+ "commits": 1,
923
+ "body": "There is a remaining one. Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull Request section? Was this discussed/approved via a Github issue or the forum? Please add a link to it if that's th"
924
  },
925
  "41076": {
926
  "t": ":rotating_light: [`v5`] Remove headmasking",
927
  "state": "MERGED",
 
928
  "merged": "2025-09-30",
929
  "first": "2025-09-19",
930
+ "last": "2025-09-29",
931
+ "by": "vasqu",
932
+ "add": 1154,
933
+ "del": 8688,
934
+ "files": 497,
935
+ "commits": 24,
936
+ "body": "As per title, time to deprecate this This includes anything in the docs, tests, and modeling xxx head mask xxx_headmasking xxx_head_mask"
937
  },
938
  "41111": {
939
  "t": "Support loading LFM2 GGUF",
940
  "state": "MERGED",
 
941
  "merged": "2025-09-24",
942
  "first": "2025-09-23",
943
+ "last": "2025-09-24",
944
+ "by": "HaroldBenoit",
945
+ "add": 57,
946
+ "del": 0,
947
+ "files": 3,
948
+ "commits": 6,
949
+ "body": "Currently, GGUF versions of LFM2 models raises \"GGUF model with architecture lfm2 is not supported yet\" error. This PR resolves this issue. Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull "
950
  },
951
  "41123": {
952
  "t": "[v5] Remove deprecated prediction loop",
953
  "state": "MERGED",
 
954
  "merged": "2025-09-30",
955
  "first": "2025-09-24",
956
+ "last": "2025-09-30",
957
+ "by": "SunMarc",
958
+ "add": 6,
959
+ "del": 530,
960
+ "files": 11,
961
+ "commits": 5,
962
+ "body": "This PR removes the legacy prediction loop that was deprecated 4 years ago. I think that it should be fine to deprecate it even without any warning for v5 which it would have been better to have. I checked files on github and no one really seemed to use the legacy one. Also SequentialDistributedSamp"
963
+ },
964
+ "41138": {
965
+ "t": "Make quantizers good citizens loading-wise",
966
+ "state": "MERGED",
967
+ "merged": "2025-09-29",
968
+ "first": "2025-09-25",
969
+ "last": "2025-09-29",
970
+ "by": "Cyrilvallez",
971
+ "add": 356,
972
+ "del": 538,
973
+ "files": 15,
974
+ "commits": 21,
975
+ "body": "The quantizers should NEVER need to access the state_dict during param_needs_quantization or create_quantized_param. This is mostly for the following reasons: Most models have a lot of different state dict (due to restricting the size of each file to 5GB), so relying on other params of the state dic"
976
  },
977
  "41253": {
978
  "t": "transformers serve quantization docs + some api fixes for bitsandbytes",
979
  "state": "MERGED",
 
980
  "merged": "2025-10-23",
981
  "first": "2025-10-01",
982
+ "last": "2025-10-23",
983
+ "by": "SunMarc",
984
+ "add": 39,
985
+ "del": 31,
986
+ "files": 2,
987
+ "commits": 18,
988
+ "body": "This PR adds some docs for transformers serve x quantization and fixes the quantization api for bitsandbytes. Following PR will add more support for other methods and I will add a section in each quantization method to show what is the preferred setup (compile ? dtype ?) when using serve !"
989
  },
990
  "41289": {
991
  "t": "Add modular detector",
992
  "state": "MERGED",
 
993
  "merged": "2025-10-03",
994
  "first": "2025-10-02",
995
+ "last": "2025-10-03",
996
+ "by": "molbap",
997
+ "add": 677,
998
+ "del": 1,
999
+ "files": 3,
1000
+ "commits": 7,
1001
+ "body": "Adds a new util to parse a new modeling_* file, whether or not it is from transformers, and lists the closest possible matches from existing models in the library using embeddings from Qwen3. Why? Because most models should be able to be defined as modulars from each other. It's not exactly the case"
1002
  },
1003
  "41370": {
1004
  "t": "[`CB`] Refactors the way we access paged",
1005
  "state": "MERGED",
 
1006
  "merged": "2025-10-06",
1007
  "first": "2025-10-06",
1008
+ "last": "2025-10-06",
1009
+ "by": "ArthurZucker",
1010
+ "add": 33,
1011
+ "del": 43,
1012
+ "files": 8,
1013
+ "commits": 6,
1014
+ "body": "The user decides on the function, but CB handles itself which \"interface\" wrapper to use. This should make stuff easier"
1015
  },
1016
  "41388": {
1017
  "t": "\ud83d\udea8Default to fast image processors for all models",
1018
  "state": "MERGED",
 
1019
  "merged": "2026-01-21",
1020
  "first": "2025-10-15",
1021
+ "last": "2026-01-21",
1022
+ "by": "yonigozlan",
1023
+ "add": 404,
1024
+ "del": 259,
1025
+ "files": 64,
1026
+ "commits": 81,
1027
+ "body": "Following the trial testing with Qwen_VL image processors, this extends defaulting to fast image processors even for checkpoints saved with a slow one to all models. Also made sure that all processors use AutoImageProcessor to instantiate their image_processor_class. On that point, defining default "
1028
  },
1029
  "41446": {
1030
  "t": "Enable non-streaming mode in `transformers serve`",
1031
  "state": "MERGED",
 
1032
  "merged": "2025-10-15",
1033
  "first": "2025-10-14",
1034
+ "last": "2025-10-14",
1035
+ "by": "LysandreJik",
1036
+ "add": 211,
1037
+ "del": 73,
1038
+ "files": 4,
1039
+ "commits": 3,
1040
+ "body": "Needs this to be merged first: #41444 Tests and docs need to be added before undraft"
1041
  },
1042
  "41474": {
1043
  "t": "\ud83d\udea8 [v5] Toggle the serialization format in processors",
1044
  "state": "MERGED",
 
1045
  "merged": "2025-10-16",
1046
  "first": "2025-10-09",
1047
+ "last": "2025-10-15",
1048
+ "by": "zucchini-nlp",
1049
+ "add": 137,
1050
+ "del": 227,
1051
+ "files": 23,
1052
+ "commits": 11,
1053
+ "body": "As per title, in v5 we will start saving in the new nested format. The loading is still supported though for old-format configs Also, we don't need an optional_attributes field that does not do what it is called. Since it is only chat template, let's just manually pop and set it. For audio tokenizer"
1054
  },
1055
  "41580": {
1056
  "t": "Refactor weight loading",
1057
  "state": "MERGED",
 
1058
  "merged": "2025-11-13",
1059
  "first": "2025-11-03",
1060
+ "last": "2025-11-13",
1061
+ "by": "ArthurZucker",
1062
+ "add": 8733,
1063
+ "del": 7189,
1064
+ "files": 586,
1065
+ "commits": 387,
1066
+ "body": "CORE REFACTORING, loading, converting, logging More helpful debugging report when loading weights If you just want to fuse qkv: It can. You just need to make sure you change the model code and pouf! WeightConverter( [\"self_attn.q_proj\", \"self_attn.k_proj\", \"self_attn.v_proj\"], \"self_attn.qkv_proj\", "
1067
+ },
1068
+ "41627": {
1069
+ "t": "[`Executorch`] Simplify for encoder models",
1070
+ "state": "MERGED",
1071
+ "merged": "2025-10-16",
1072
+ "first": "2025-10-15",
1073
+ "last": "2025-10-15",
1074
+ "by": "vasqu",
1075
+ "add": 25,
1076
+ "del": 189,
1077
+ "files": 6,
1078
+ "commits": 2,
1079
+ "body": "Now that we include a fast path using no vmapping, we can revert the extra treatment. Followup to #41586"
1080
  },
1081
  "41633": {
1082
  "t": "[v5] \ud83d\udea8Refactor subprocessors handling in processors",
1083
  "state": "MERGED",
 
1084
  "merged": "2025-11-07",
1085
  "first": "2025-10-15",
1086
+ "last": "2025-11-07",
1087
+ "by": "yonigozlan",
1088
+ "add": 368,
1089
+ "del": 839,
1090
+ "files": 149,
1091
+ "commits": 29,
1092
+ "body": "Refactor the handling of subprocessors in processors. Main change is that we deduce the subprocessors from the init signature instead of having to manually add \"subprocessor_class\" attributes. This means we can remove all attributes attribute in processors, along with all \"subprocessor\"_class attrib"
1093
  },
1094
  "41655": {
1095
  "t": "Switch to CB if cache_implementation == paged",
1096
  "state": "MERGED",
 
1097
  "merged": "2025-10-16",
1098
  "first": "2025-10-16",
1099
+ "last": "2025-10-16",
1100
+ "by": "remi-or",
1101
+ "add": 75,
1102
+ "del": 3,
1103
+ "files": 2,
1104
+ "commits": 7,
1105
+ "body": "This PR fixes one of the issues raised in #41501 This PR adds paged as a static cache implementation and switches from generate to generate_batch if it's the chosen implementation. It also raises errors if an argument for generate is passed to generate_batch but is not supported, and warn the user t"
1106
  },
1107
  "41657": {
1108
  "t": "Erroring when KernelConfig is passed without use_kernels = True",
1109
  "state": "MERGED",
 
1110
  "merged": "2025-10-16",
1111
  "first": "2025-10-16",
1112
+ "last": "2025-10-16",
1113
+ "by": "MekkCyber",
1114
+ "add": 6,
1115
+ "del": 0,
1116
+ "files": 1,
1117
+ "commits": 3,
1118
+ "body": "as per title"
1119
  },
1120
  "41773": {
1121
  "t": "Simplify and standardize processor tests",
1122
  "state": "MERGED",
 
1123
  "merged": "2025-11-26",
1124
  "first": "2025-10-15",
1125
+ "last": "2025-11-26",
1126
+ "by": "yonigozlan",
1127
+ "add": 1925,
1128
+ "del": 5197,
1129
+ "files": 113,
1130
+ "commits": 52,
1131
+ "body": "Improve ProcessorTestMixin to standardize processor tests, especially the setup part. Requires #41633"
1132
+ },
1133
+ "41897": {
1134
+ "t": "[FPQuant] MXFP8 and MXFP4 backwards support",
1135
+ "state": "MERGED",
1136
+ "merged": "2025-11-04",
1137
+ "first": "2025-10-27",
1138
+ "last": "2025-10-28",
1139
+ "by": "BlackSamorez",
1140
+ "add": 20,
1141
+ "del": 5,
1142
+ "files": 5,
1143
+ "commits": 6,
1144
+ "body": "This PR adds MXFP4 and MXFP8 backwards support in combination with MXFP4 forward, allowing for lightning-fast QAT on Blackwell GPUs. It's blocked by QuTLASS v0.2.0 and fp_quant v0.3.0 release we're hoping to release within a day or two. Before submitting This PR fixes a typo or improves the docs (yo"
1145
  },
1146
  "41916": {
1147
  "t": "feat(ci): add continuous batching to benchmarks",
1148
  "state": "MERGED",
 
1149
  "merged": "2025-11-07",
1150
  "first": "2025-11-07",
1151
+ "last": "2025-11-07",
1152
+ "by": "McPatate",
1153
+ "add": 190,
1154
+ "del": 107,
1155
+ "files": 10,
1156
+ "commits": 15,
1157
+ "body": ""
1158
  },
1159
  "41990": {
1160
  "t": "feat(kernels): add opt-out flag to disable kernels hub usage through the lib",
1161
  "state": "MERGED",
 
1162
  "merged": "2025-11-12",
1163
  "first": "2025-11-10",
1164
+ "last": "2025-11-12",
1165
+ "by": "mfuntowicz",
1166
+ "add": 138,
1167
+ "del": 2,
1168
+ "files": 2,
1169
+ "commits": 7,
1170
+ "body": "This PR attempts to bring an opt-out mecanism to disable hub kernels (aka kernels) globally through transformers. The default value for the USE_HUB_KERNELS environment variable defaults to YES to maintain backward compatiblity with the current machinery. In more details, it works by proxying the ker"
1171
+ },
1172
+ "41992": {
1173
+ "t": "[PoC] HF exporters",
1174
+ "state": "MERGED",
1175
+ "merged": "2026-07-02",
1176
+ "first": "2026-03-19",
1177
+ "last": "2026-07-02",
1178
+ "by": "IlyasMoutawwakil",
1179
+ "add": 5468,
1180
+ "del": 264,
1181
+ "files": 89,
1182
+ "commits": 406,
1183
+ "body": "HfExporters: Native, Unified export for PyTorch / ONNX / ExecuTorch Edit: pieces of this PR were merged separately (#42697, #42317), so this one is now focused on the HfExporters machinery itself. What this adds A native, in-Transformers export pipeline \u2014 one base class (HfExporter), three subclasse"
1184
+ },
1185
+ "42043": {
1186
+ "t": "Fix bnb for the weights refactor",
1187
+ "state": "MERGED",
1188
+ "merged": "2025-11-18",
1189
+ "first": "2025-11-03",
1190
+ "last": "2025-11-18",
1191
+ "by": "SunMarc",
1192
+ "add": 627,
1193
+ "del": 185,
1194
+ "files": 13,
1195
+ "commits": 396,
1196
+ "body": "This PR fixes bnb support (8bit + 4bit) in the new weight loading logic. Testing from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig model_name = \"meta-llama/Llama-3.2-3B-Instruct\" quantization_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type=\"nf4\") #mo"
1197
  },
1198
  "42191": {
1199
  "t": "Much more efficient and clear weight initialization and tie weights",
1200
  "state": "MERGED",
 
1201
  "merged": "2025-11-14",
1202
  "first": "2025-11-14",
1203
+ "last": "2025-11-14",
1204
+ "by": "Cyrilvallez",
1205
+ "add": 3341,
1206
+ "del": 5443,
1207
+ "files": 416,
1208
+ "commits": 45,
1209
+ "body": "Follow-up to #41580. This PR is focused on speed and efficiency, as well as clarity. It avoids the following issues/bottlenecks: No longer need to switch dynamically the class of the Parameters (which is hard to understand and may lead to issues with quantization etc, as it's not always clear if the"
1210
  },
1211
  "42254": {
1212
  "t": "Update conversion mapping to separate renaming from converting",
1213
  "state": "MERGED",
 
1214
  "merged": "2025-11-20",
1215
  "first": "2025-11-18",
1216
+ "last": "2025-11-20",
1217
+ "by": "ArthurZucker",
1218
+ "add": 525,
1219
+ "del": 329,
1220
+ "files": 5,
1221
+ "commits": 34,
1222
+ "body": "Update the conversion: to support torchao"
1223
+ },
1224
+ "42426": {
1225
+ "t": "Clean bnb integration using weight converter",
1226
+ "state": "MERGED",
1227
+ "merged": "2025-11-26",
1228
+ "first": "2025-11-26",
1229
+ "last": "2025-11-26",
1230
+ "by": "SunMarc",
1231
+ "add": 141,
1232
+ "del": 104,
1233
+ "files": 4,
1234
+ "commits": 6,
1235
+ "body": "This PR cleans bnb integration, we don't need to deal with missing keys."
1236
  },
1237
  "42436": {
1238
  "t": "Fix mixed torch.Tensor and DTensor in generate when use FSDP2 + LoRA",
1239
  "state": "MERGED",
 
1240
  "merged": "2025-12-03",
1241
  "first": "2025-12-03",
1242
+ "last": "2025-12-03",
1243
+ "by": "Xiao-Chenguang",
1244
+ "add": 2,
1245
+ "del": 0,
1246
+ "files": 1,
1247
+ "commits": 3,
1248
+ "body": "model.generate raise an Fix mixed torch.Tensor and DTensor when FSDP2 and LoRA is applied. The trainer class is the base class for trl trainers like PPOTrainer and GRPOTrainer in which model.generate is called during the training loop. When FSDP2 is used, all gather and reshard is automatically done"
1249
  },
1250
  "42634": {
1251
  "t": "allow registration of custom checkpoint conversion mappings",
1252
  "state": "MERGED",
 
1253
  "merged": "2025-12-11",
1254
  "first": "2025-12-10",
1255
+ "last": "2025-12-11",
1256
+ "by": "winglian",
1257
+ "add": 52,
1258
+ "del": 1,
1259
+ "files": 2,
1260
+ "commits": 5,
1261
+ "body": "For custom modeling, this provides a clean route to adding new or overwriting existing mappings with custom mappings. Additionally, it properly memoizes the cached mappings. Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you rea"
1262
  },
1263
  "42697": {
1264
  "t": "batched and grouped experts implementations",
1265
  "state": "MERGED",
 
1266
  "merged": "2026-01-05",
1267
  "first": "2025-12-06",
1268
+ "last": "2026-01-01",
1269
+ "by": "IlyasMoutawwakil",
1270
+ "add": 1060,
1271
+ "del": 374,
1272
+ "files": 65,
1273
+ "commits": 74,
1274
+ "body": "I have started experimenting with pure pytorch MoE implementations following the HF exporters PR while trying to find a traceable/exportable variant for onnx/openvino. In this PR I copy the attn_implementation API into a similar experts_implementation API, and added two new implementations: batched_"
1275
  },
1276
  "42932": {
1277
  "t": "[docs] inference engines",
1278
  "state": "MERGED",
 
1279
  "merged": "2026-01-07",
1280
  "first": "2026-01-07",
1281
+ "last": "2026-01-07",
1282
+ "by": "stevhliu",
1283
+ "add": 248,
1284
+ "del": 208,
1285
+ "files": 5,
1286
+ "commits": 3,
1287
+ "body": "starts off a new section dedicated to our integrations with inference engines"
1288
  },
1289
  "43176": {
1290
  "t": "[docs] tensorrt-llm",
1291
  "state": "MERGED",
 
1292
  "merged": "2026-01-13",
1293
  "first": "2026-01-13",
1294
+ "last": "2026-01-13",
1295
+ "by": "stevhliu",
1296
+ "add": 49,
1297
+ "del": 8,
1298
+ "files": 2,
1299
+ "commits": 2,
1300
+ "body": "adds tensorrt-llm guide showing how it uses transformers to the new community integrations section"
1301
  },
1302
  "43185": {
1303
  "t": "[docs] llama.cpp",
1304
  "state": "MERGED",
 
1305
  "merged": "2026-01-09",
1306
  "first": "2026-01-09",
1307
+ "last": "2026-01-09",
1308
+ "by": "stevhliu",
1309
+ "add": 72,
1310
+ "del": 6,
1311
+ "files": 2,
1312
+ "commits": 1,
1313
+ "body": "adds llama.cpp guide explaining how it uses transformers to the new community integrations section create separate section for deployment (more to follow soon here!)"
1314
  },
1315
  "43227": {
1316
  "t": "GptOss experts implementation",
1317
  "state": "MERGED",
 
1318
  "merged": "2026-01-22",
1319
  "first": "2026-01-12",
1320
+ "last": "2026-01-21",
1321
+ "by": "IlyasMoutawwakil",
1322
+ "add": 359,
1323
+ "del": 223,
1324
+ "files": 8,
1325
+ "commits": 35,
1326
+ "body": "[wip] Fixes #43193 @vasqu @mattteochen Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull Request section? Was this discussed/approved via a Github issue or the forum? Please add a link to it"
1327
  },
1328
  "43241": {
1329
  "t": "[docs] deploying",
1330
  "state": "MERGED",
 
1331
  "merged": "2026-02-05",
1332
  "first": "2026-01-30",
1333
+ "last": "2026-01-30",
1334
+ "by": "stevhliu",
1335
+ "add": 230,
1336
+ "del": 24,
1337
+ "files": 8,
1338
+ "commits": 3,
1339
+ "body": "adds ecosystem integration docs for deploying with Candle, ExecuTorch, and MLX"
1340
  },
1341
  "43294": {
1342
  "t": "[docs] pretraining",
1343
  "state": "MERGED",
 
1344
  "merged": "2026-01-20",
1345
  "first": "2026-01-14",
1346
+ "last": "2026-01-20",
1347
+ "by": "stevhliu",
1348
+ "add": 106,
1349
+ "del": 0,
1350
+ "files": 3,
1351
+ "commits": 3,
1352
+ "body": "adds ecosystem integration docs for pretraining frameworks (nanotron, torchtitan)"
1353
  },
1354
  "43312": {
1355
  "t": "[docs] finetuning",
1356
  "state": "MERGED",
 
1357
  "merged": "2026-01-20",
1358
  "first": "2026-01-20",
1359
+ "last": "2026-01-20",
1360
+ "by": "stevhliu",
1361
+ "add": 196,
1362
+ "del": 9,
1363
+ "files": 4,
1364
+ "commits": 4,
1365
+ "body": "adds ecosystem integration docs for axolotl, trl, and unsloth"
1366
  },
1367
  "43521": {
1368
  "t": "[CB] Minor perf improvements and ty compatibility",
1369
  "state": "MERGED",
 
1370
  "merged": "2026-01-28",
1371
  "first": "2026-01-27",
1372
+ "last": "2026-01-28",
1373
+ "by": "remi-or",
1374
+ "add": 228,
1375
+ "del": 169,
1376
+ "files": 8,
1377
+ "commits": 21,
1378
+ "body": "Summary This PR introduces minor changes to the continuous batching feature: Performance optimize the get_seqlens_k function to avoid looping over all cache managers logits indexing is now done inside the cuda graph, as there was no apparent reason it should be done outside fuse the get_read/write_i"
1379
  },
1380
  "43807": {
1381
  "t": "Refactor trainer init",
1382
  "state": "MERGED",
 
1383
  "merged": "2026-02-10",
1384
  "first": "2026-02-06",
1385
+ "last": "2026-02-10",
1386
+ "by": "SunMarc",
1387
+ "add": 319,
1388
+ "del": 250,
1389
+ "files": 5,
1390
+ "commits": 16,
1391
+ "body": "This PR simplifies Trainer __init__: Quantization validation extracted PEFT unwrapping deduplicated Liger Kernel extracted \u2014 apply_liger_kernel Label smoother simplified Validations grouped \u2014 _validate_args() method consolidates three scattered validation blocks (arg checks, optimizer checks, datase"
1392
  },
1393
  "43917": {
1394
  "t": "Model patching API",
1395
  "state": "MERGED",
 
1396
  "merged": "2026-03-02",
1397
  "first": "2026-02-11",
1398
+ "last": "2026-02-25",
1399
+ "by": "IlyasMoutawwakil",
1400
+ "add": 1157,
1401
+ "del": 3,
1402
+ "files": 5,
1403
+ "commits": 27,
1404
+ "body": "Integrates a library wide monkey patching API to fix #43284 and allow things like restructuring the expert weights and fusing the qkv projections. A concrete and complete example: Details from typing import Unpack import torch import torch.nn as nn from transformers import AutoModelForCausalLM, Conc"
1405
  },
1406
  "43977": {
1407
  "t": "stable grouped_mm API",
1408
  "state": "MERGED",
 
1409
  "merged": "2026-02-16",
1410
  "first": "2026-02-13",
1411
+ "last": "2026-02-16",
1412
+ "by": "IlyasMoutawwakil",
1413
+ "add": 48,
1414
+ "del": 23,
1415
+ "files": 5,
1416
+ "commits": 9,
1417
+ "body": "Fixes # (issue) Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull Request section? Was this discussed/approved via a Github issue or the forum? Please add a link to it if that's the case. Di"
1418
  },
1419
  "44043": {
1420
  "t": "`grouped_mm` fallback",
1421
  "state": "MERGED",
 
1422
  "merged": "2026-02-18",
1423
  "first": "2026-02-16",
1424
+ "last": "2026-02-17",
1425
+ "by": "IlyasMoutawwakil",
1426
+ "add": 170,
1427
+ "del": 162,
1428
+ "files": 31,
1429
+ "commits": 8,
1430
+ "body": "With this PR, grouped_mm impl will always be default when the model uses experts impl (even in old torch versions) and will no longer fail if you are torch compiling an fp16/fp32 model, the impl will succeed in all cases, except cuda-graphs (common failure case in all grouped_mm impls). Before submi"
1431
  },
1432
  "44152": {
1433
  "t": "AutoGrad support for grouped_mm fallback",
1434
  "state": "MERGED",
 
1435
  "merged": "2026-02-20",
1436
  "first": "2026-02-19",
1437
+ "last": "2026-02-20",
1438
+ "by": "IlyasMoutawwakil",
1439
+ "add": 79,
1440
+ "del": 45,
1441
+ "files": 3,
1442
+ "commits": 6,
1443
+ "body": "Fixes # (issue) Before submitting This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case). Did you read the contributor guideline, Pull Request section? Was this discussed/approved via a Github issue or the forum? Please add a link to it if that's the case. Di"
1444
  },
1445
  "44174": {
1446
  "t": "Expand model-structure lint rules with a fast AST-based, ruff-like framework",
1447
  "state": "MERGED",
 
1448
  "merged": "2026-03-12",
1449
  "first": "2026-03-11",
1450
+ "last": "2026-03-12",
1451
+ "by": "tarekziade",
1452
+ "add": 1367,
1453
+ "del": 24,
1454
+ "files": 4,
1455
+ "commits": 12,
1456
+ "body": "This draft expands utils/check_modeling_structure.py into a rule-driven linter for model code, with new checks and tests, while keeping runtime very low. Key features: The checker is intentionally AST-only (no heavy imports/execution), so it stays fast (<3 seconds against all models) It's built in a"
1457
  },
1458
  "44300": {
1459
  "t": "Dynamic weight conversion is recursive",
1460
  "state": "MERGED",
 
1461
  "merged": "2026-03-26",
1462
  "first": "2026-02-26",
1463
+ "last": "2026-03-26",
1464
+ "by": "zucchini-nlp",
1465
+ "add": 2539,
1466
+ "del": 470,
1467
+ "files": 18,
1468
+ "commits": 48,
1469
+ "body": "The recursive feature is needed for me in #44252 to allow timm backbone define its conversion only once. Also it currently allows to delete \"t5gemma2\" from conversion, allowing its backbones to rename weight"
1470
+ },
1471
+ "44348": {
1472
+ "t": "Enable MetalConfig to load pre-quantized MLX models from HuggingFace Hub",
1473
+ "state": "OPEN",
1474
+ "merged": "",
1475
+ "first": "2026-02-28",
1476
+ "last": "2026-02-28",
1477
+ "by": "n0kovo",
1478
+ "add": 341,
1479
+ "del": 32,
1480
+ "files": 4,
1481
+ "commits": 4,
1482
+ "body": "Summary Most quantized models for Apple Silicon on the Hub are in MLX format. The MetalConfig quantization backend supports on-the-fly quantization of standard checkpoints but cannot load pre-quantized MLX models. This PR fixes the five issues blocking that: quantizers/auto.py: Detect MLX affine qua"
1483
  },
1484
  "44469": {
1485
  "t": "[remote code/vllm] Fix incorrect tied weights",
1486
  "state": "MERGED",
 
1487
  "merged": "2026-03-05",
1488
  "first": "2026-03-05",
1489
+ "last": "2026-03-05",
1490
+ "by": "Cyrilvallez",
1491
+ "add": 4,
1492
+ "del": 2,
1493
+ "files": 1,
1494
+ "commits": 1,
1495
+ "body": "For remote code that behave correctly with tied weights, we need to keep the same behavior as for the main lib, i.e. not remove them from tied weights (as tied weights are marked as missing to avoid inits!!)"
1496
  },
1497
  "44558": {
1498
  "t": "Add /v1/completions endpoint (OpenAI legacy completions API) to `transformers serve`",
1499
  "state": "MERGED",
 
1500
  "merged": "2026-04-22",
1501
  "first": "2026-04-22",
1502
+ "last": "2026-04-22",
1503
+ "by": "rain-1",
1504
+ "add": 567,
1505
+ "del": 1,
1506
+ "files": 5,
1507
+ "commits": 1,
1508
+ "body": "Adds support for the legacy text completions endpoint, which accepts a freeform text prompt (no chat template) and returns generated text in choices[].text. Supports both streaming and non-streaming modes, suffix for fill-in-the-middle insertion, and proper finish_reason detection. What does this PR"
1509
+ },
1510
+ "44576": {
1511
+ "t": "Disable async loading when quantizing on the fly",
1512
+ "state": "MERGED",
1513
+ "merged": "2026-03-16",
1514
+ "first": "2026-03-10",
1515
+ "last": "2026-03-16",
1516
+ "by": "SunMarc",
1517
+ "add": 16,
1518
+ "del": 5,
1519
+ "files": 1,
1520
+ "commits": 11,
1521
+ "body": "Fixes #44387. This PR disable async loading when we want to quantize the model. it is actually faster than doing a semaphore. If a quantizer happens to quantize faster than materializing the weights, we might consider adding adding a semaphore for that. For future reference: I tried to Materialize a"
1522
  },
1523
  "44699": {
1524
  "t": "Fix several based models' pipeline parallel support",
1525
  "state": "MERGED",
 
1526
  "merged": "2026-03-20",
1527
  "first": "2026-03-14",
1528
+ "last": "2026-03-20",
1529
+ "by": "hmellor",
1530
+ "add": 219,
1531
+ "del": 148,
1532
+ "files": 52,
1533
+ "commits": 6,
1534
+ "body": "These models have base_model_pp_plans but currently do not work because the base model's forward pass depends on all the layers being Qwen2VLDecoderLayer. i.e. if one of the layers is removed/replaced with Identity, decoder_layer.attention_type will fail with an attribute error. This PR fixes the is"
1535
  },
1536
  "44939": {
1537
  "t": "refactor: mlinter as its own package",
1538
  "state": "MERGED",
 
1539
  "merged": "2026-03-24",
1540
  "first": "2026-03-23",
1541
+ "last": "2026-03-23",
1542
+ "by": "tarekziade",
1543
+ "add": 2038,
1544
+ "del": 1446,
1545
+ "files": 25,
1546
+ "commits": 8,
1547
+ "body": "Refactored and cleaned up model linter separated package one rule per module refactored legacy checks into their own rules simplified pattern, duplication removal"
1548
  },
1549
  "44950": {
1550
  "t": "\ud83d\udea8 [Cache] Native mamba & hybrid cache",
1551
  "state": "MERGED",
 
1552
  "merged": "2026-03-31",
1553
  "first": "2026-03-23",
1554
+ "last": "2026-03-31",
1555
+ "by": "Cyrilvallez",
1556
+ "add": 1346,
1557
+ "del": 4113,
1558
+ "files": 64,
1559
+ "commits": 59,
1560
+ "body": "As per the title. This PR finally makes mamba layer caches first class citizen, and adds native support for them. It supports the following layers combinations: all mamba layers alternating attention layer/mamba layer layers that are BOTH mamba and attention (zamba models) For this, it adds the 2 fo"
1561
  },
1562
  "45279": {
1563
  "t": "add expert parallelism for gemma-4-26B-A4B-it",
1564
  "state": "MERGED",
 
1565
  "merged": "2026-04-22",
1566
  "first": "2026-04-21",
1567
+ "last": "2026-04-22",
1568
+ "by": "sywangyi",
1569
+ "add": 19,
1570
+ "del": 3,
1571
+ "files": 2,
1572
+ "commits": 2,
1573
+ "body": "Fixes # (issue) Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by our ability to review and respond to them. As a result, we ask that new users do not submit pure code agent PRs "
1574
  },
1575
  "45436": {
1576
  "t": "Add expert parallelism (EP) config support for Qwen3 MoE",
1577
  "state": "MERGED",
 
1578
  "merged": "2026-04-22",
1579
  "first": "2026-04-21",
1580
+ "last": "2026-04-22",
1581
+ "by": "AmineDiro",
1582
+ "add": 35,
1583
+ "del": 0,
1584
+ "files": 5,
1585
+ "commits": 11,
1586
+ "body": "Summary Added base_model_ep_plan to Qwen3MoeConfig enabling expert parallelism via DistributedConfig(enable_expert_parallel=True) Depends on #45473 Test plan Tested on 8\u00d7H100 with torchrun --nproc_per_node=8 using Qwen/Qwen3-30B-A3B (128 experts, 4 KV heads): TP=2, DP=4: forward \u2713, backward \u2713, exper"
1587
  },
1588
  "45473": {
1589
  "t": "Fix EP: RouterParallel shape, tp_plan property, grouped_mm sentinels",
1590
  "state": "MERGED",
 
1591
  "merged": "2026-04-21",
1592
  "first": "2026-04-16",
1593
+ "last": "2026-04-20",
1594
+ "by": "AmineDiro",
1595
+ "add": 190,
1596
+ "del": 48,
1597
+ "files": 5,
1598
+ "commits": 7,
1599
+ "body": "While benchmarking Qwen3-30B-A3B SFT training with Expert Parallelism (EP) using TRL, I found three bugs that combine to produce silently wrong results or NaN loss. Every existing test uses tp_plan=\"auto\" which bypasses RouterParallel and EP sentinel handling entirely if I understand correctly. Thes"
1600
  },
1601
  "45475": {
1602
  "t": "chore(qa): split out mlinter",
1603
  "state": "MERGED",
 
1604
  "merged": "2026-04-20",
1605
  "first": "2026-04-17",
1606
+ "last": "2026-04-20",
1607
+ "by": "tarekziade",
1608
+ "add": 54,
1609
+ "del": 3405,
1610
+ "files": 29,
1611
+ "commits": 10,
1612
+ "body": "Split out the mlinter tool see https://github.com/huggingface/transformers-mlinter We want to be able to: use it from other CI projects remove the ability to alter the linter from Transformers PRs This change should keep the exact same behavior than before"
1613
  },
1614
  "45585": {
1615
  "t": "qa: bumped mlinter and allow local override",
1616
  "state": "MERGED",
 
1617
  "merged": "2026-04-23",
1618
  "first": "2026-04-23",
1619
+ "last": "2026-04-23",
1620
+ "by": "tarekziade",
1621
+ "add": 323,
1622
+ "del": 28,
1623
+ "files": 6,
1624
+ "commits": 5,
1625
+ "body": "bumps mlinter to 0.1.1 add a local config that can be used to quickly override the linter rules"
1626
  },
1627
  "45621": {
1628
  "t": "Better Grouped GEMM + EP",
1629
  "state": "MERGED",
 
1630
  "merged": "2026-05-04",
1631
  "first": "2026-04-23",
1632
+ "last": "2026-05-04",
1633
+ "by": "IlyasMoutawwakil",
1634
+ "add": 394,
1635
+ "del": 219,
1636
+ "files": 5,
1637
+ "commits": 36,
1638
+ "body": "The idea is that we shouldn't be clamping experts ids at all, clamping them makes their tokens get prjected as if they were routed to the last expert, instead, we should let the sentinels be: moved to the queue by the sorting (num_experts/sentinel is the biggest value) dropped by the histogram (beca"
1639
  },
1640
  "45634": {
1641
  "t": "DeepGEMM BF16 + mixed FP8/FP4 + MegaMoE + refactor",
1642
  "state": "MERGED",
 
1643
  "merged": "2026-06-02",
1644
  "first": "2026-04-23",
1645
+ "last": "2026-06-02",
1646
+ "by": "IlyasMoutawwakil",
1647
+ "add": 1627,
1648
+ "del": 486,
1649
+ "files": 16,
1650
+ "commits": 164,
1651
+ "body": "Based on #45621 GroupedLinear co-authored-by @sywangyi Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by our ability to review and respond to them. As a result, we ask that new u"
1652
  },
1653
  "45795": {
1654
  "t": "[docs] adding audio/video processors",
1655
  "state": "MERGED",
 
1656
  "merged": "2026-05-18",
1657
  "first": "2026-05-05",
1658
+ "last": "2026-05-18",
1659
+ "by": "stevhliu",
1660
+ "add": 255,
1661
+ "del": 9,
1662
+ "files": 3,
1663
+ "commits": 4,
1664
+ "body": "follow up to this comment about separate docs for how to add audio and video processors"
1665
+ },
1666
+ "45975": {
1667
+ "t": "GGUF: optional Metal dequant fast path via kernels-community",
1668
+ "state": "OPEN",
1669
+ "merged": "",
1670
+ "first": "2026-05-14",
1671
+ "last": "2026-05-14",
1672
+ "by": "ArthurZucker",
1673
+ "add": 127,
1674
+ "del": 6,
1675
+ "files": 3,
1676
+ "commits": 3,
1677
+ "body": "Summary Adds an optional Metal dequant fast path in GGUFDequantize. When the kernels package is installed and kernels-community/gguf-dequant is reachable, MPS GGUF loads (Q4_0 / Q8_0 / Q4_K / Q5_K / Q6_K / IQ4_NL / IQ4_XS) route through one Metal compute kernel per tensor \u2014 2.3\u20137.5\u00d7 faster than the "
1678
  },
1679
  "46246": {
1680
  "t": "[`Revert`] FSDP+Dtensor refactor related changes",
1681
  "state": "MERGED",
 
1682
  "merged": "2026-05-28",
1683
  "first": "2026-05-27",
1684
+ "last": "2026-05-27",
1685
+ "by": "vasqu",
1686
+ "add": 3011,
1687
+ "del": 6332,
1688
+ "files": 286,
1689
+ "commits": 8,
1690
+ "body": "As per title \ud83d\udc40 Discussed internally"
1691
+ },
1692
+ "46325": {
1693
+ "t": "feat(quantizers): support non-weight param names in TorchAo safetensors loading",
1694
+ "state": "MERGED",
1695
+ "merged": "2026-06-08",
1696
+ "first": "2026-06-01",
1697
+ "last": "2026-06-08",
1698
+ "by": "agesf",
1699
+ "add": 52,
1700
+ "del": 17,
1701
+ "files": 2,
1702
+ "commits": 4,
1703
+ "body": "Add _discover_quantized_param_names() to dynamically extract parameter names from safetensors metadata instead of hardcoding 'weight' Add aten.normal_ compat patch for TorchAO tensor subclasses What does this PR do? Allows loading quantized checkpoints that have non-standard parameter names."
1704
+ },
1705
+ "46357": {
1706
+ "t": "[docs] gguf mps",
1707
+ "state": "OPEN",
1708
+ "merged": "",
1709
+ "first": "2026-05-18",
1710
+ "last": "2026-06-02",
1711
+ "by": "stevhliu",
1712
+ "add": 71,
1713
+ "del": 0,
1714
+ "files": 1,
1715
+ "commits": 2,
1716
+ "body": "adds docs for: #45975, dequant fast path through a Metal kernel #45977, native quantized and on-the-fly quantization inference on a Metal kernel. i don't think the docs here overlap with the existing docs in your PR :)"
1717
  },
1718
  "46410": {
1719
  "t": "Add workflow to post CI Grafana dashboard link to PR",
1720
  "state": "MERGED",
 
1721
  "merged": "2026-06-04",
1722
  "first": "2026-06-04",
1723
+ "last": "2026-06-04",
1724
+ "by": "ydshieh",
1725
+ "add": 39,
1726
+ "del": 0,
1727
+ "files": 1,
1728
+ "commits": 1,
1729
+ "body": "Summary Adds pr-ci-post-dashboard-link.yml workflow that triggers via workflow_run on completion of \"PR CI\" Uses actions/github-script to find the PR by head SHA and post a comment with the Grafana observability dashboard link Works for both same-repo and fork PRs since workflow_run always runs in t"
1730
  },
1731
  "46412": {
1732
  "t": "Improve CI dashboard comment: rename and deduplicate",
1733
  "state": "MERGED",
 
1734
  "merged": "2026-06-04",
1735
  "first": "2026-06-04",
1736
+ "last": "2026-06-04",
1737
+ "by": "ydshieh",
1738
+ "add": 18,
1739
+ "del": 1,
1740
+ "files": 1,
1741
+ "commits": 1,
1742
+ "body": "Summary Renames comment from **CI Observability Dashboard:** to **CI Dashboard:** Before posting, deletes any existing dashboard comment on the PR (handles both old and new text) to avoid duplicates on CI reruns"
1743
  },
1744
  "46415": {
1745
  "t": "ci: mention code quality failure in CI dashboard comment",
1746
  "state": "MERGED",
 
1747
  "merged": "2026-06-04",
1748
  "first": "2026-06-04",
1749
+ "last": "2026-06-04",
1750
+ "by": "ydshieh",
1751
+ "add": 16,
1752
+ "del": 1,
1753
+ "files": 1,
1754
+ "commits": 1,
1755
+ "body": "Summary When check_code_quality fails, the CI dashboard comment posted to the PR now includes a warning explaining that all test jobs were skipped The Grafana link is always posted regardless, so partial OTEL data is still accessible How it works After the \"PR CI\" workflow completes, listJobsForWork"
1756
  },
1757
  "46500": {
1758
  "t": "Fix `{image/video/audio}_token_ids` in `ProcessorMixin`",
1759
  "state": "MERGED",
 
1760
  "merged": "2026-06-08",
1761
  "first": "2026-06-08",
1762
+ "last": "2026-06-08",
1763
+ "by": "hmellor",
1764
+ "add": 18,
1765
+ "del": 0,
1766
+ "files": 1,
1767
+ "commits": 1,
1768
+ "body": "#45493 converted these attributes to properties. This breaks custom models which set {image/video/audio}_token_ids directly because: There is no setter, so these properties are read-only The getter will only ever read from image_token_id This PR maintains BC by: Adding a setter for these properties "
1769
  },
1770
  "46721": {
1771
  "t": "ci: disable CircleCI by replacing config with no-op",
1772
  "state": "MERGED",
 
1773
  "merged": "2026-06-17",
1774
  "first": "2026-06-17",
1775
+ "last": "2026-06-17",
1776
+ "by": "ydshieh",
1777
+ "add": 241,
1778
+ "del": 228,
1779
+ "files": 3,
1780
+ "commits": 3,
1781
+ "body": "Replaces the full CircleCI config with a minimal no-op job that immediately passes, and disables the CircleCI failure summary workflow. Why not just delete the config? Removing config.yml causes CircleCI to post a failing \"no configuration found\" check. Replacing it with a no-op keeps the check gree"
1782
  },
1783
  "46774": {
1784
  "t": "Insert a Grafana badge in the PR",
1785
  "state": "MERGED",
 
1786
  "merged": "2026-06-28",
1787
  "first": "2026-06-28",
1788
+ "last": "2026-06-28",
1789
+ "by": "tarekziade",
1790
+ "add": 714,
1791
+ "del": 73,
1792
+ "files": 3,
1793
+ "commits": 4,
1794
+ "body": "Moved the script to Python, in utils, and insert a badge in the PR"
1795
+ },
1796
+ "46792": {
1797
+ "t": "Enable CB paged decode fast path for MPS hub kernels + sensible block_size default",
1798
+ "state": "OPEN",
1799
+ "merged": "",
1800
+ "first": "2026-06-21",
1801
+ "last": "2026-06-26",
1802
+ "by": "ArthurZucker",
1803
+ "add": 52,
1804
+ "del": 7,
1805
+ "files": 4,
1806
+ "commits": 3,
1807
+ "body": "Summary Three changes to make continuous batching (CB) work well on Apple Silicon (MPS) with the metal-flash-sdpa kernel: 1. Enable paged decode fast path for hub kernels (initialization.py) ensure_decode_fast_path_is_available previously only allowed the paged decode fast path (block tables + flash"
1808
  },
1809
  "47005": {
1810
  "t": "Add tiny_model_id support to ProcessorTesterMixin for memory-sensitive tests",
1811
  "state": "MERGED",
 
1812
  "merged": "2026-07-03",
1813
  "first": "2026-07-01",
1814
+ "last": "2026-07-03",
1815
+ "by": "ydshieh",
1816
+ "add": 82,
1817
+ "del": 30,
1818
+ "files": 3,
1819
+ "commits": 4,
1820
+ "body": "Summary Adds tiny_model_id attribute to ProcessorTesterMixin allowing test classes to specify a lightweight Hub repo (e.g. containing only a tiny tokenizer) for use in most tests, while keeping model_id as the full Hub repo used only when explicitly requested via get_processor(use_full=True) / get_c"
1821
+ },
1822
+ "47096": {
1823
+ "t": "Exportable kimi",
1824
+ "state": "MERGED",
1825
+ "merged": "2026-07-31",
1826
+ "first": "2026-07-06",
1827
+ "last": "2026-07-30",
1828
+ "by": "IlyasMoutawwakil",
1829
+ "add": 517,
1830
+ "del": 293,
1831
+ "files": 19,
1832
+ "commits": 13,
1833
+ "body": "Fixes # (issue) Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. We are currently bottlenecked by our ability to review and respond to them. As a result, we ask that new users do not submit pure code agent PRs "
1834
  },
1835
  "47172": {
1836
  "t": "Switch mlinter to 0.1.2",
1837
  "state": "MERGED",
 
1838
  "merged": "2026-07-09",
1839
  "first": "2026-07-09",
1840
+ "last": "2026-07-09",
1841
+ "by": "tarekziade",
1842
+ "add": 165,
1843
+ "del": 3,
1844
+ "files": 4,
1845
+ "commits": 4,
1846
+ "body": "runs on current mlinter 0.1.2 models that fail to pass the linter TRF014 (trust_remote_code) \u2502 auto, legitimate loader exemption we can keep that TRF016 (dead do_* flags) \u2502 kimi_k25, minimax_m3_vl TRF018 (super()._init_weights) | cosmos3_omni, diffusion_gemma, eomt_dinov3, exaone4_5, gemma4, higgs_a"
1847
+ },
1848
+ "47201": {
1849
+ "t": "Make executorch exporter tests always use xnnpack backend",
1850
+ "state": "MERGED",
1851
+ "merged": "2026-07-09",
1852
+ "first": "2026-07-09",
1853
+ "last": "2026-07-09",
1854
+ "by": "tarekziade",
1855
+ "add": 19,
1856
+ "del": 6,
1857
+ "files": 3,
1858
+ "commits": 19,
1859
+ "body": "The ExecuTorch export tests target CPU (via xnnpack on CPU), and its deps now live in the CPU torch-light CI image. Mark test_executorch_export with a new require_non_gpu decorator so they run on the CPU image, not the GPU nightly. Lower DynamoExporter.min_versions torch 2.11.0 -> 2.10.0 so the test"
1860
+ },
1861
+ "47243": {
1862
+ "t": "Executorch exporter fixes",
1863
+ "state": "MERGED",
1864
+ "merged": "2026-08-03",
1865
+ "first": "2026-07-10",
1866
+ "last": "2026-08-03",
1867
+ "by": "IlyasMoutawwakil",
1868
+ "add": 1894,
1869
+ "del": 314,
1870
+ "files": 28,
1871
+ "commits": 41,
1872
+ "body": "Been hunting them since we introduced their skips in the original HF exporters PR. Also adds static cache export support, for in-place cache modification with cache position. Also adds multi-token decoding support, so that the decode graph can do prefill (empty cache), single token decode and multi-"
1873
  },
1874
  "47289": {
1875
  "t": "Pipeline parallel naive inference",
1876
  "state": "MERGED",
 
1877
  "merged": "2026-08-25",
1878
  "first": "2026-07-01",
1879
+ "last": "2026-08-21",
1880
+ "by": "3outeille",
1881
+ "add": 758,
1882
+ "del": 10,
1883
+ "files": 9,
1884
+ "commits": 317,
1885
+ "body": "Summary Adding a very naive pipeline parallel inference engine as a baseline (no specific schedule such as gpipe, 1f1b yet) supports tied/untied weight embeddings To make it work seamless with generate(), we are broadcasting every logits to every ranks. In long term, we want only the last stage to h"
1886
  },
1887
  "47357": {
1888
  "t": "Add FSDP CI and end-to-end FSDP tests + save fsdp",
1889
  "state": "MERGED",
 
1890
  "merged": "2026-07-23",
1891
  "first": "2026-07-16",
1892
+ "last": "2026-07-23",
1893
+ "by": "3outeille",
1894
+ "add": 960,
1895
+ "del": 82,
1896
+ "files": 16,
1897
+ "commits": 40,
1898
+ "body": "Summary Part of FSDP orchestration stack (2/2). Replaces #46990. #47352 (1/2) \u2192 this PR (2/2) Add FSDP DCP (distributed_checkpoint=True) paths in save_pretrained Add FSDP gather into single checkpoint saving Add FSDPTesterMixin and tests/test_fsdp_mixin.py (load/save/generation) Add base_model_fsdp_"
1899
  },
1900
  "47579": {
1901
  "t": "\ud83d\udea8 TP dtensor API inference + training",
1902
  "state": "MERGED",
 
1903
  "merged": "2026-08-21",
1904
  "first": "2026-06-17",
1905
+ "last": "2026-08-21",
1906
+ "by": "3outeille",
1907
+ "add": 1234,
1908
+ "del": 2161,
1909
+ "files": 27,
1910
+ "commits": 244,
1911
+ "body": "Summary This PR replaces the legacy Transformers tensor-parallel implementation with a DTensor-native backend while preserving efficient inference for kernels that operate on local tensors + loading time. We outsource all the placement and parameters sharding logic to Dtensor which will enable us to"
1912
+ },
1913
+ "47711": {
1914
+ "t": "Fix ImportError in transformers.exporters on torch < 2.8",
1915
+ "state": "MERGED",
1916
+ "merged": "2026-08-03",
1917
+ "first": "2026-08-02",
1918
+ "last": "2026-08-03",
1919
+ "by": "Neal006",
1920
+ "add": 1,
1921
+ "del": 1,
1922
+ "files": 1,
1923
+ "commits": 2,
1924
+ "body": "Fixes #47710. import transformers.exporters raises ImportError on torch 2.5, 2.6 and 2.7, all inside the declared torch>=2.5 support range. exporter_executorch.py imports guard_or_true at module scope, and that symbol was added in torch 2.8 where it replaced definitely_true. is_torch_available() che"
1925
+ },
1926
+ "47779": {
1927
+ "t": "Refactor GGUF to speed-up inference",
1928
+ "state": "OPEN",
1929
+ "merged": "",
1930
+ "first": "2026-07-31",
1931
+ "last": "2026-08-31",
1932
+ "by": "SunMarc",
1933
+ "add": 2435,
1934
+ "del": 137,
1935
+ "files": 35,
1936
+ "commits": 43,
1937
+ "body": "This PR updates the GGUF in transformers in order to take advantages of some features that landed for the past few months and are now stable: conversion using WeightConverter, instead of our custom tensor processor. Support for loading quantized models. Before we were dequantizing by default. Add su"
1938
  },
1939
  "47819": {
1940
  "t": "Post mlinter findings as inline PR review comments",
1941
  "state": "MERGED",
 
1942
  "merged": "2026-08-06",
1943
  "first": "2026-08-06",
1944
+ "last": "2026-08-06",
1945
+ "by": "ydshieh",
1946
+ "add": 198,
1947
+ "del": 6,
1948
+ "files": 2,
1949
+ "commits": 2,
1950
+ "body": "What this does Adds a post-mlinter-review job to pr-ci-post-dashboard-link.yml that: Downloads the mlinter-findings artifact produced by the check_code_quality job Posts the findings as a single PR review with inline comments anchored to the diff The review includes a summary table (rule \u2192 count) an"
1951
  },
1952
  "47863": {
1953
  "t": "Bump default flash-attn2 hub kernel version to v3",
1954
  "state": "MERGED",
 
1955
  "merged": "2026-08-17",
1956
  "first": "2026-08-10",
1957
+ "last": "2026-08-17",
1958
+ "by": "jiqing-feng",
1959
+ "add": 19,
1960
+ "del": 3,
1961
+ "files": 1,
1962
+ "commits": 6,
1963
+ "body": "What load_and_register_attn_kernel hardcodes version=1 as the default major version for every attention kernel loaded from the Hub. Pin kernels-community/flash-attn2 to v3 through a small per-repo table instead. Why The v1 branch of kernels-community/flash-attn2 only ships builds up to Torch 2.12, s"
1964
+ },
1965
+ "47871": {
1966
+ "t": "Make muse glimmer exportable",
1967
+ "state": "MERGED",
1968
+ "merged": "2026-08-12",
1969
+ "first": "2026-08-10",
1970
+ "last": "2026-08-11",
1971
+ "by": "IlyasMoutawwakil",
1972
+ "add": 160,
1973
+ "del": 286,
1974
+ "files": 5,
1975
+ "commits": 3,
1976
+ "body": "Fixes # (issue) Code Agent Policy The Transformers repo is currently being overwhelmed by a large number of PRs and issue comments written by code agents. These often are low-quality, or fix extremely minor issues that occur rarely or never in practice. As a result, we're instituting a rule that fir"
1977
+ },
1978
+ "47900": {
1979
+ "t": "Fix DynamicCache reconstruction during ExecuTorch export",
1980
+ "state": "MERGED",
1981
+ "merged": "2026-08-21",
1982
+ "first": "2026-08-11",
1983
+ "last": "2026-08-21",
1984
+ "by": "eladsegal",
1985
+ "add": 97,
1986
+ "del": 25,
1987
+ "files": 3,
1988
+ "commits": 19,
1989
+ "body": "Fixes incorrect cache-update graphs when exporting models with DynamicCache to ExecuTorch. torch.export uses the registered pytree functions to flatten and reconstruct DynamicCache during export. The current pytree registration keeps populated key/value tensors, but loses each layer's type and windo"
1990
  },
1991
  "47918": {
1992
  "t": "Moving mlinter to 0.1.4",
1993
  "state": "MERGED",
 
1994
  "merged": "2026-08-18",
1995
  "first": "2026-08-17",
1996
+ "last": "2026-08-18",
1997
+ "by": "tarekziade",
1998
+ "add": 2249,
1999
+ "del": 131,
2000
+ "files": 54,
2001
+ "commits": 13,
2002
+ "body": "Bumps transformers-mlinter to 0.1.4 to pick up the new rules (TRF020 to TRF057), and resolves everything those rules flag. The base pinned transformers-mlinter==0.1.2, whose utils/rules.toml carried only TRF001\u2013TRF019. This bump adds 37 new rules in the TRF020\u2013TRF057 range, and every finding they re"
2003
  },
2004
  "48056": {
2005
  "t": "[CircleCI] Enable CI for private forks, no-op for public repo",
2006
  "state": "MERGED",
 
2007
  "merged": "2026-08-18",
2008
  "first": "2026-08-18",
2009
+ "last": "2026-08-18",
2010
+ "by": "ydshieh",
2011
+ "add": 201,
2012
+ "del": 14,
2013
+ "files": 2,
2014
+ "commits": 5,
2015
+ "body": "Summary Public repo (huggingface/transformers): genuine noop workflow \u2014 CircleCI does nothing (GitHub Actions handles CI here) Private fork (e.g. new model additions): full CI runs automatically, gated by [run-circleci] in the commit message to control costs Add peft_integration CircleCI job to matc"
2016
  },
2017
  "48117": {
2018
  "t": "Enable mlinter findings artifact for inline PR reviews",
2019
  "state": "MERGED",
 
2020
  "merged": "2026-08-19",
2021
  "first": "2026-08-19",
2022
+ "last": "2026-08-19",
2023
+ "by": "ydshieh",
2024
+ "add": 1,
2025
+ "del": 1,
2026
+ "files": 1,
2027
+ "commits": 1,
2028
+ "body": "Summary Last piece of the mlinter inline review pipeline, verified in #48109. Adds --output-json mlinter-findings.json to the mlinter invocation so the check_code_quality job uploads a mlinter-findings artifact. The post-mlinter-review job (already on main) then downloads it and posts inline review "
2029
  },
2030
  "48167": {
2031
  "t": "Apply context parallelism to the evaluation path",
2032
  "state": "MERGED",
 
2033
  "merged": "2026-08-21",
2034
  "first": "2026-08-20",
2035
+ "last": "2026-08-20",
2036
+ "by": "qgallouedec",
2037
+ "add": 4,
2038
+ "del": 2,
2039
+ "files": 1,
2040
+ "commits": 2,
2041
+ "body": "With context parallelism enabled, eval_loss is reported as cp_size times its true value. Training loss is correct; only evaluation is wrong. Same everything, varying only cp_size: Run it once per cp_size (no config file needed): FSDP=\"--use_fsdp --fsdp_version 2 --fsdp_auto_wrap_policy TRANSFORMER_B"
2042
  },
2043
  "48271": {
2044
  "t": "draft modular v2 + transformers format",
2045
  "state": "OPEN",
 
2046
  "merged": "",
2047
  "first": "2026-08-25",
2048
+ "last": "2026-08-26",
2049
+ "by": "ArthurZucker",
2050
+ "add": 15323,
2051
+ "del": 3245,
2052
+ "files": 175,
2053
+ "commits": 21,
2054
+ "body": ""
 
 
2055
  }
2056
  }
releases.json ADDED
@@ -0,0 +1 @@
 
 
1
+ [{"n": "v4.42.0", "d": "2024-06-27", "big": false}, {"n": "v4.43.0", "d": "2024-07-23", "big": false}, {"n": "v4.44.0", "d": "2024-08-06", "big": false}, {"n": "v4.45.0", "d": "2024-09-25", "big": false}, {"n": "v4.46.0", "d": "2024-10-24", "big": false}, {"n": "v4.47.0", "d": "2024-12-05", "big": false}, {"n": "v4.48.0", "d": "2025-01-10", "big": false}, {"n": "v4.49.0", "d": "2025-02-17", "big": false}, {"n": "v4.50.0", "d": "2025-03-21", "big": false}, {"n": "v4.51.0", "d": "2025-04-05", "big": false}, {"n": "v4.52.0", "d": "2025-05-20", "big": false}, {"n": "v4.53.0", "d": "2025-06-26", "big": false}, {"n": "v4.54.0", "d": "2025-07-25", "big": false}, {"n": "v4.55.0", "d": "2025-08-05", "big": false}, {"n": "v4.56.0", "d": "2025-08-29", "big": false}, {"n": "v4.57.0", "d": "2025-10-03", "big": false}, {"n": "v5.0.0rc0", "d": "2025-12-01", "big": true}, {"n": "v5.0.0rc1", "d": "2025-12-11", "big": false}, {"n": "v5.0.0rc2", "d": "2026-01-07", "big": false}, {"n": "v5.0.0rc3", "d": "2026-01-14", "big": false}, {"n": "v5.0.0", "d": "2026-01-26", "big": true}, {"n": "v5.1.0", "d": "2026-02-05", "big": false}, {"n": "v5.2.0", "d": "2026-02-16", "big": false}, {"n": "v5.3.0", "d": "2026-03-04", "big": false}, {"n": "v5.4.0", "d": "2026-03-27", "big": false}, {"n": "v5.5.0", "d": "2026-04-02", "big": false}, {"n": "v5.6.0", "d": "2026-04-22", "big": false}, {"n": "v5.7.0", "d": "2026-04-28", "big": false}, {"n": "v5.8.0", "d": "2026-05-05", "big": false}, {"n": "v5.9.0", "d": "2026-05-20", "big": false}, {"n": "v5.10.0", "d": "2026-06-03", "big": false}, {"n": "v5.11.0", "d": "2026-06-10", "big": false}, {"n": "v5.12.0", "d": "2026-06-12", "big": false}, {"n": "v5.13.0", "d": "2026-07-03", "big": false}, {"n": "v5.14.0", "d": "2026-07-15", "big": false}, {"n": "v5.15.0", "d": "2026-08-10", "big": false}, {"n": "v5.16.0", "d": "2026-08-26", "big": false}]