Dataset Viewer

Split (1)

train

The dataset viewer is not available for this split.

Cannot load the dataset split (in streaming mode) to extract the first rows.

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
Snow rocky mountains peaks canyon. snow blanketed rocky mountains surround and shadow deep canyons.  (... 74 chars omitted): string
Splash of turquoise water in extreme slow motion, alpha channel included.-0.mp4: string
A 3D model of a 1800s victorian house.-0.mp4: string
this is how I do makeup in the morning.-0.mp4: string
A raccoon that looks like a turtle, digital art.-0.mp4: string
Busy freeway at night.-0.mp4: string
Sewing machine, old sewing machine working.-0.mp4: string
A panda cooking in the kitchen-0.mp4: string
Clown fish swimming through the coral reef-0.mp4: string
Hyper-realistic spaceship landing on Mars-0.mp4: string
source: string
vbench_commit: string
selection_ids_in_generation_order: list<item: int64>
  child 0, item: int64
method: string
selection_ids_in_stratum_order: list<item: int64>
  child 0, item: int64
full93: struct<n: int64, bf16_mean: double, mxfp8_mean: double, mean_of_reported_deltas: double, difference_ (... 31 chars omitted)
  child 0, n: int64
  child 1, bf16_mean: double
  child 2, mxfp8_mean: double
  child 3, mean_of_reported_deltas: double
  child 4, difference_of_rounded_score_means: double
retrieved_date: timestamp[s]
prompt_sha256: string
mini10: struct<n: int64, bf16_mean: double, mxfp8_mean: double, mean_of_reported_deltas: double, difference_ (... 31 chars omitted)
  child 0, n: int64
  child 1, bf16_mean: double
  child 2, mxfp8_mean: double
  child 3, mean_of_reported_deltas: double
  child 4, difference_of_rounded_score_means: double
rounding_note: string
to
{'source': Value('string'), 'retrieved_date': Value('timestamp[s]'), 'vbench_commit': Value('string'), 'selection_ids_in_stratum_order': List(Value('int64')), 'selection_ids_in_generation_order': List(Value('int64')), 'method': Value('string'), 'full93': {'n': Value('int64'), 'bf16_mean': Value('float64'), 'mxfp8_mean': Value('float64'), 'mean_of_reported_deltas': Value('float64'), 'difference_of_rounded_score_means': Value('float64')}, 'mini10': {'n': Value('int64'), 'bf16_mean': Value('float64'), 'mxfp8_mean': Value('float64'), 'mean_of_reported_deltas': Value('float64'), 'difference_of_rounded_score_means': Value('float64')}, 'rounding_note': Value('string'), 'prompt_sha256': Value('string')}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 147, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 127, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 483, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2840, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2373, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2398, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2378, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2306, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              Snow rocky mountains peaks canyon. snow blanketed rocky mountains surround and shadow deep canyons.  (... 74 chars omitted): string
              Splash of turquoise water in extreme slow motion, alpha channel included.-0.mp4: string
              A 3D model of a 1800s victorian house.-0.mp4: string
              this is how I do makeup in the morning.-0.mp4: string
              A raccoon that looks like a turtle, digital art.-0.mp4: string
              Busy freeway at night.-0.mp4: string
              Sewing machine, old sewing machine working.-0.mp4: string
              A panda cooking in the kitchen-0.mp4: string
              Clown fish swimming through the coral reef-0.mp4: string
              Hyper-realistic spaceship landing on Mars-0.mp4: string
              source: string
              vbench_commit: string
              selection_ids_in_generation_order: list<item: int64>
                child 0, item: int64
              method: string
              selection_ids_in_stratum_order: list<item: int64>
                child 0, item: int64
              full93: struct<n: int64, bf16_mean: double, mxfp8_mean: double, mean_of_reported_deltas: double, difference_ (... 31 chars omitted)
                child 0, n: int64
                child 1, bf16_mean: double
                child 2, mxfp8_mean: double
                child 3, mean_of_reported_deltas: double
                child 4, difference_of_rounded_score_means: double
              retrieved_date: timestamp[s]
              prompt_sha256: string
              mini10: struct<n: int64, bf16_mean: double, mxfp8_mean: double, mean_of_reported_deltas: double, difference_ (... 31 chars omitted)
                child 0, n: int64
                child 1, bf16_mean: double
                child 2, mxfp8_mean: double
                child 3, mean_of_reported_deltas: double
                child 4, difference_of_rounded_score_means: double
              rounding_note: string
              to
              {'source': Value('string'), 'retrieved_date': Value('timestamp[s]'), 'vbench_commit': Value('string'), 'selection_ids_in_stratum_order': List(Value('int64')), 'selection_ids_in_generation_order': List(Value('int64')), 'method': Value('string'), 'full93': {'n': Value('int64'), 'bf16_mean': Value('float64'), 'mxfp8_mean': Value('float64'), 'mean_of_reported_deltas': Value('float64'), 'difference_of_rounded_score_means': Value('float64')}, 'mini10': {'n': Value('int64'), 'bf16_mean': Value('float64'), 'mxfp8_mean': Value('float64'), 'mean_of_reported_deltas': Value('float64'), 'difference_of_rounded_score_means': Value('float64')}, 'rounding_note': Value('string'), 'prompt_sha256': Value('string')}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

YAML Metadata Warning:empty or missing yaml metadata in repo card

Check out the documentation for more information.

Wan2.2 dense93 mini10 v1

用于快速比较不同推理配方的 imaging_quality；从已有 93 条中挑选 10 条。这是自定义筛选集，未经独立 datatype 或 seed 的泛化验证，不是官方 VBench mini benchmark。

文件

prompts.txt：10 条原始英文 prompt，每行一条，保持原 dense93 顺序。
prompts.json：VBench custom_input 的 {video_filename: prompt} 映射，seed 对应视频索引仍为 -0。
selection.csv：分层顺序、原始编号、公开分数、挑选理由。
dense93_scores.csv：从 issue 评论提取的完整 93 条公开分数，用于复核。
analysis.json：分层规则、统计、来源、原始版本和 prompt 校验值。
VBENCH_LICENSE：VBench 的 Apache 2.0 许可证。Prompt 文本未修改，列表经过筛选。

选择方法

将 MXFP8 配方分数减 BF16 分数的差值从低到高排序，按人数分成 10 层（每层 9 或 10 条），每层选 1 条。层内考虑接近典型差值以及场景和画风互补，不重复堆叠海洋动物或幻想风景。这是人工分层筛选，没有随机抽样，也没有穷举拟合全量均分。第一层均值为 -22.71，选择水花 -22.6；不将 -42.6 的极端退化作为这一层的唯一代表。其他层保留中等退化、近乎持平和改善样本。后续固定这 10 条，不随待测 datatype 更换。

在现有结果上回算

集合	BF16 均分	MXFP8 配方均分	公开 delta 列的平均
93 条	70.658	68.073	-2.580
10 条	70.580	67.870	-2.670

源表的 score 和 delta 分别四舍五入至 0.1，因此 mini 两列均分相减为 -2.710，与 delta 列均值 -2.670 略有不同。报告时宜写约 -2.7 分；不能把这部分差异视为显著差异。这里只做了同一批数据的回算，没有生成新视频，也没有重新评分或肉眼检查视频。 10 条不能保证检出未纳入的极端失败，也不能据此证明所有 datatype 的排序可靠。本集没有独立的 2D 油画、烟花、长时序场景专项覆盖；正式结论回到全量集。

使用

在已有 bench-omni-wan checkout 中，用同一完整生成配置执行：

python generate.py YOUR_DATATYPE.env \
  --prompt-file /home/user2/wan22-vbench-mini10/prompts.txt \
  --num-prompts 10 \
  --output-dir /path/to/output/UNIQUE_DATATYPE_mini10

每种配置使用单独输出目录，避免生成器的 resume 行为跳过旧视频。异机运行请先复制 prompts.txt 并修改路径。评分沿用已验证的 VBench 环境、模型权重、预处理模式，以及 imaging_quality/custom_input，prompt_file 使用 prompts.json。同机器相同并发下，视频生成工作量为原来的 10/93（约 10.8%）；初始化与模型加载等固定开销不会同比减少。

固定 seed=42、1280x720、81 帧、16 fps、40 steps、guidance 4.0/3.0、boundary_ratio=0.875、flow_shift=5.0 以及原 negative prompt。若要单独归因到 datatype，还应固定硬件、attention、cache 设置、软件版本；改变其他因素时结果按整个配方解释。除均分外，保存每条分差与最差样本，避免改善和退化互相抵消。分数接近时不据 10 条单 seed 强行排序，应对候选配方使用配对额外 seed 或全量复核。

来源与限制

完整结果：https://github.com/yiliu30/bench-omni-wan/issues/4#issuecomment-5673012918
生成规范与配方差异：https://github.com/yiliu30/bench-omni-wan/issues/4
VBench prompt 列表：https://github.com/Vchitect/VBench/blob/fd18b3d055cb0fc6f066ca90fe2c3c8cbb698490/vbench/VBench_full_info.json

已验证公开 93 条 prompt 与上述官方版本的 imaging_quality 列表逐条一致。现有样本仅包含 BF16/B300 与 MXFP8 linear + SageV3/XPU 两个配方；XPU 还混用了三种 kernel build。因此这份 mini 集是基于现有配方差异的候选，不能称为已验证的纯 datatype 敏感性基准。

Downloads last month: 15

Total file size:

29.4 kB