Datasets:

ZoeyZou
/

dual-agent-loss11-step192-generation-20260922

Languages:

English

Chinese

Tags:

evaluation

generation

qwen3-vl

Dataset card Files Files and versions

xet

Community

Dataset Viewer

Duplicate

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
snapshot_utc: string
model: string
experiment: string
not_single_shared_model: bool
completed_benchmarks: int64
complete_answers: int64
partial_answers: int64
total_unique_answers: int64
benchmarks: list<item: struct<benchmark: string, generated: int64, expected: int64, complete: bool, grading_comp (... 56 chars omitted)
  child 0, item: struct<benchmark: string, generated: int64, expected: int64, complete: bool, grading_complete: bool, (... 44 chars omitted)
      child 0, benchmark: string
      child 1, generated: int64
      child 2, expected: int64
      child 3, complete: bool
      child 4, grading_complete: bool
      child 5, source_file: string
      child 6, source_sha256: string
vlguard: struct<status: string, expected: int64, included: bool>
  child 0, status: string
  child 1, expected: int64
  child 2, included: bool
grading_status: string
scoring_protocol: string
images_included: bool
credentials_included: bool
checkpoint_included: bool
training_data_included: bool
github_code: string
original_snapshot_utc: string
prior_snapshot_revision: string
latest_update: string
mm_safety_truncated_answers: int64
HANDOFF.json: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/llava-wild/raw_predictions.jsonl.manifest.json: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/wildvision/predictions.jsonl: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: str
...
int64
  child 1, sha256: string
answers/mm-safety/raw_predictions.jsonl.manifest.json: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/mia-bench/predictions.jsonl.manifest.json: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/llava-wilder/predictions.json: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/mss-bench/predictions.json.manifest.json: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/llava-wilder/raw_predictions.jsonl: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/mm-safety/predictions.json.manifest.json: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/mm-vet-v2/predictions.json.manifest.json: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/llava-wild/predictions.json.manifest.json: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/llava-wilder/predictions.json.manifest.json: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/mia-bench/predictions.jsonl: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
answers/siuo/predictions.jsonl: struct<bytes: int64, sha256: string>
  child 0, bytes: int64
  child 1, sha256: string
to
{'HANDOFF.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'README.md': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answer_index.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/predictions.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/predictions.json.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/raw_predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/raw_predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/predictions.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/predictions.json.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/raw_predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/raw_predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mia-bench/predictions.json': {'bytes': Va
...
s-bench/predictions.json.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mss-bench/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mss-bench/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mss-bench/raw_predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mss-bench/raw_predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/siuo/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/siuo/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/wildvision/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/wildvision/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/predictions.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/predictions.json.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/raw_predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/raw_predictions.jsonl.manifest.json': {'bytes': Value('int64'), '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
              snapshot_utc: string
              model: string
              experiment: string
              not_single_shared_model: bool
              completed_benchmarks: int64
              complete_answers: int64
              partial_answers: int64
              total_unique_answers: int64
              benchmarks: list<item: struct<benchmark: string, generated: int64, expected: int64, complete: bool, grading_comp (... 56 chars omitted)
                child 0, item: struct<benchmark: string, generated: int64, expected: int64, complete: bool, grading_complete: bool, (... 44 chars omitted)
                    child 0, benchmark: string
                    child 1, generated: int64
                    child 2, expected: int64
                    child 3, complete: bool
                    child 4, grading_complete: bool
                    child 5, source_file: string
                    child 6, source_sha256: string
              vlguard: struct<status: string, expected: int64, included: bool>
                child 0, status: string
                child 1, expected: int64
                child 2, included: bool
              grading_status: string
              scoring_protocol: string
              images_included: bool
              credentials_included: bool
              checkpoint_included: bool
              training_data_included: bool
              github_code: string
              original_snapshot_utc: string
              prior_snapshot_revision: string
              latest_update: string
              mm_safety_truncated_answers: int64
              HANDOFF.json: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/llava-wild/raw_predictions.jsonl.manifest.json: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/wildvision/predictions.jsonl: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: str
              ...
              int64
                child 1, sha256: string
              answers/mm-safety/raw_predictions.jsonl.manifest.json: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/mia-bench/predictions.jsonl.manifest.json: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/llava-wilder/predictions.json: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/mss-bench/predictions.json.manifest.json: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/llava-wilder/raw_predictions.jsonl: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/mm-safety/predictions.json.manifest.json: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/mm-vet-v2/predictions.json.manifest.json: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/llava-wild/predictions.json.manifest.json: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/llava-wilder/predictions.json.manifest.json: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/mia-bench/predictions.jsonl: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              answers/siuo/predictions.jsonl: struct<bytes: int64, sha256: string>
                child 0, bytes: int64
                child 1, sha256: string
              to
              {'HANDOFF.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'README.md': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answer_index.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/predictions.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/predictions.json.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/raw_predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wild/raw_predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/predictions.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/predictions.json.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/raw_predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/llava-wilder/raw_predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mia-bench/predictions.json': {'bytes': Va
              ...
              s-bench/predictions.json.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mss-bench/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mss-bench/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mss-bench/raw_predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mss-bench/raw_predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/siuo/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/siuo/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/wildvision/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/wildvision/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/predictions.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/predictions.json.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/predictions.jsonl.manifest.json': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/raw_predictions.jsonl': {'bytes': Value('int64'), 'sha256': Value('string')}, 'answers/mm-safety/raw_predictions.jsonl.manifest.json': {'bytes': Value('int64'), '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.

双模型1:1，最终step192 Policy：生成回答

本回答集来自已完成的双模型4B:4B、GSN/BOC/Replay20、judgment:generation=1:1实验。不是正在训练的单模型1:3实验，也不是先前step40模型。

尚未GPT评分。现包含9个完整benchmark，共8,990条回答：原有8项3,950条保持不变，MM-Safety更新为完整5,040条。VLGuard不在本次上传范围内。

Benchmark	已生成/总数	完整性
mia-bench	400/400	完整
llava-wild	60/60	完整
llava-wilder	128/128	完整
mss-bench	1960/1960	完整
mm-vet	218/218	完整
mm-vet-v2	517/517	完整
wildvision	500/500	完整
siuo	167/167	完整
mm-safety	5040/5040	完整生成，待GPT评分

文件与复核

answers/<benchmark>/：已完成输出逐字节保留，包括生成manifest及完整MIA400/Wilder128。
answers/mm-safety/：完整5,040条原始/合并回答及生成manifest。此前608条部分快照保留在历史提交 1214992bc8196282417cfeb727f19c728cb9a992，当前版本已替换。
answer_index.jsonl：每个 (model, benchmark, uid) 恰好一条回答的便捷索引；正式评分使用原文件及manifest。
HANDOFF.json：模型、范围、数量和快照时间。FILES_SHA256.json：文件SHA256与字节数。
MM-Vet的samples.jsonl实际是JSON数组，使用json.load；predictions.json为ID到回答的字典。

下载

本仓库当前为公开仓库，按用户确认保持公开，可直接下载：

from huggingface_hub import snapshot_download
snapshot_download('ZoeyZou/dual-agent-loss11-step192-generation-20260922',
                  repo_type='dataset', local_dir='step192-answers')

后续评分

评分代码沿用 initial-comments-rlhf-experiments 中的 handoff/generation_suite.py 和 frozen c9a60fb7 evaluator；仓库名称不代表本次模型是initial-comments实验。

本仓库交付回答和生成契约，不包含原图。record的绝对image_path仅用于来源追踪；评分机应按manifest固定dataset revision准备原图、题目元数据及参考答案，并配置image-root，按UID与image_sha256核对映射。可以用generation_suite.py prepare准备数据；将完整answers目录放到 <work>/evaluation/4b-loss11/global_step_192/policy/generation/。

每bench只评分一次，MM-Vet两版也一次。MM-Safety已通过5,040个固定UID及顺序校验，可以进行全量评分；生成完成不代表评分完成。合并使用 (model, benchmark, uid, answer_sha256)，保留评分入口、模型、prompt及图片hash，不混用其他checkpoint分数。

不包含checkpoint、训练数据、API/HF凭证或聊天日志。

MM-Safety本次更新

生成任务于 2026-09-22T10:23:29.807432+00:00 正常完成。5,040条唯一回答、固定UID顺序、模型身份、dataset revision与文件SHA256均已核对。共有534条回答达到生成token上限，原样保留并通过hit_token_limit标记；未补跑、未编造评分。此更新只修改MM-Safety文件和总索引/清单，原有8个benchmark输出不变。

Downloads last month: 42

Total file size:

112 MB