Upload folder using huggingface_hub
Browse files- SUBMISSION.md +14 -0
- board.json +1 -0
SUBMISSION.md
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Submission packet
|
| 2 |
+
|
| 3 |
+
Dataset: https://huggingface.co/datasets/pragnyanramtha/splice-contextbench
|
| 4 |
+
|
| 5 |
+
Track: Agent + Backbone
|
| 6 |
+
Row name: **Splice + DeepSeek-V4-Pro-0813**
|
| 7 |
+
Model: `fireworks_ai/accounts/fireworks/models/deepseek-v4-pro-0813`
|
| 8 |
+
|
| 9 |
+
This is a **budget slice**, not a 1,136-task full-board claim. Live #1 (mini-SWE + DeepSeek-V4-Pro) is 57.5% Pass@1 / Line F1 0.338 on the full set.
|
| 10 |
+
|
| 11 |
+
ContextBench authors (no public submit form): `{he.ye, zhaoyang.chu.25}@ucl.ac.uk`
|
| 12 |
+
Board: https://contextbench.github.io/
|
| 13 |
+
|
| 14 |
+
Do not send until django-16263 (and further fail-first rows) are scored into the dataset.
|
board.json
CHANGED
|
@@ -3,6 +3,7 @@
|
|
| 3 |
"model": "fireworks_ai/accounts/fireworks/models/deepseek-v4-pro-0813",
|
| 4 |
"slice": "contextbench-lite-fail-first",
|
| 5 |
"note": "Budget slice. Not the 1136-task ContextBench board.",
|
|
|
|
| 6 |
"published_backbone_full_set": {
|
| 7 |
"name": "mini-SWE + DeepSeek-V4-Pro",
|
| 8 |
"pass_at_1": 0.575,
|
|
|
|
| 3 |
"model": "fireworks_ai/accounts/fireworks/models/deepseek-v4-pro-0813",
|
| 4 |
"slice": "contextbench-lite-fail-first",
|
| 5 |
"note": "Budget slice. Not the 1136-task ContextBench board.",
|
| 6 |
+
"hf_dataset": "https://huggingface.co/datasets/pragnyanramtha/splice-contextbench",
|
| 7 |
"published_backbone_full_set": {
|
| 8 |
"name": "mini-SWE + DeepSeek-V4-Pro",
|
| 9 |
"pass_at_1": 0.575,
|