GAP-ocrbench-v2 / README.md
elichen-skymizer's picture
Add deterministic seed-1234 subsample-50 and subsample-100 configs
ab410c8 verified
|
Raw
History Blame Contribute Delete
50.9 kB
metadata
dataset_info:
  - config_name: gemma-4-26b-a4b-it-ins-gen-2048-vt280
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 551577922
        num_examples: 984
    download_size: 550303774
    dataset_size: 551577922
  - config_name: gemma-4-26b-a4b-it-ins-gen-2048-vt280-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 44991381
        num_examples: 50
    download_size: 44532288
    dataset_size: 44991381
  - config_name: gemma-4-26b-a4b-it-ins-gen-2048-vt280-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 67486263
        num_examples: 100
    download_size: 66491202
    dataset_size: 67486263
  - config_name: gemma-4-26b-a4b-it-think-gen-4096-vt280
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 568422783
        num_examples: 968
    download_size: 557515959
    dataset_size: 568422783
  - config_name: gemma-4-26b-a4b-it-think-gen-4096-vt280-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 50077596
        num_examples: 50
    download_size: 48375003
    dataset_size: 50077596
  - config_name: gemma-4-26b-a4b-it-think-gen-4096-vt280-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 71985674
        num_examples: 100
    download_size: 68734443
    dataset_size: 71985674
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt140
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 554416038
        num_examples: 988
    download_size: 553013841
    dataset_size: 554416038
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt140-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 48474134
        num_examples: 50
    download_size: 48019160
    dataset_size: 48474134
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt140-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 74885796
        num_examples: 100
    download_size: 73942808
    dataset_size: 74885796
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt280
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 557444185
        num_examples: 991
    download_size: 556021362
    dataset_size: 557444185
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt280-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 15009800
        num_examples: 50
    download_size: 14487298
    dataset_size: 15009800
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt280-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 47507320
        num_examples: 100
    download_size: 46465726
    dataset_size: 47507320
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt560
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 561499908
        num_examples: 992
    download_size: 560088604
    dataset_size: 561499908
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt560-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 20210309
        num_examples: 50
    download_size: 19553738
    dataset_size: 20210309
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt560-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 43291046
        num_examples: 100
    download_size: 41879510
    dataset_size: 43291046
  - config_name: gemma-4-e4b-it-think-gen-4096-vt140
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 563100564
        num_examples: 989
    download_size: 560672462
    dataset_size: 563100564
  - config_name: gemma-4-e4b-it-think-gen-4096-vt140-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 42618123
        num_examples: 50
    download_size: 41876503
    dataset_size: 42618123
  - config_name: gemma-4-e4b-it-think-gen-4096-vt140-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 77884315
        num_examples: 100
    download_size: 76192397
    dataset_size: 77884315
  - config_name: gemma-4-e4b-it-think-gen-4096-vt280
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 567484600
        num_examples: 991
    download_size: 565066912
    dataset_size: 567484600
  - config_name: gemma-4-e4b-it-think-gen-4096-vt280-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 19897966
        num_examples: 50
    download_size: 18997534
    dataset_size: 19897966
  - config_name: gemma-4-e4b-it-think-gen-4096-vt280-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 47717656
        num_examples: 100
    download_size: 45779825
    dataset_size: 47717656
  - config_name: gemma-4-e4b-it-think-gen-4096-vt560
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 570082658
        num_examples: 989
    download_size: 567712549
    dataset_size: 570082658
  - config_name: gemma-4-e4b-it-think-gen-4096-vt560-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 24493589
        num_examples: 50
    download_size: 23312498
    dataset_size: 24493589
  - config_name: gemma-4-e4b-it-think-gen-4096-vt560-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 48024271
        num_examples: 100
    download_size: 45668638
    dataset_size: 48024271
  - config_name: qwen3.5-4b-ins-gen-2048
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 567412033
        num_examples: 986
    download_size: 566224571
    dataset_size: 567412033
  - config_name: qwen3.5-4b-ins-gen-2048-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 46492750
        num_examples: 50
    download_size: 45486266
    dataset_size: 46492750
  - config_name: qwen3.5-4b-ins-gen-2048-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 76072547
        num_examples: 100
    download_size: 73956783
    dataset_size: 76072547
  - config_name: qwen3.5-4b-think-gen-4096
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 590334925
        num_examples: 979
    download_size: 586369327
    dataset_size: 590334925
  - config_name: qwen3.5-4b-think-gen-4096-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 36385496
        num_examples: 50
    download_size: 34046251
    dataset_size: 36385496
  - config_name: qwen3.5-4b-think-gen-4096-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 84439384
        num_examples: 100
    download_size: 79351356
    dataset_size: 84439384
  - config_name: qwen3.6-35b-a3b-ins-gen-2048
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 567043588
        num_examples: 987
    download_size: 565893166
    dataset_size: 567043588
  - config_name: qwen3.6-35b-a3b-ins-gen-2048-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 54877308
        num_examples: 50
    download_size: 53565097
    dataset_size: 54877308
  - config_name: qwen3.6-35b-a3b-ins-gen-2048-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 82462210
        num_examples: 100
    download_size: 80154428
    dataset_size: 82462210
  - config_name: qwen3.6-35b-a3b-think-gen-4096
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 582835433
        num_examples: 983
    download_size: 580089408
    dataset_size: 582835433
  - config_name: qwen3.6-35b-a3b-think-gen-4096-subsample-50
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 41824920
        num_examples: 50
    download_size: 39798258
    dataset_size: 41824920
  - config_name: qwen3.6-35b-a3b-think-gen-4096-subsample-100
    features:
      - name: source
        dtype: string
      - name: item_id
        dtype: string
      - name: origin_id
        dtype: int64
      - name: sample_seed
        dtype: int64
      - name: question
        dtype: string
      - name: images
        list: image
      - name: num_images
        dtype: int32
      - name: ref_answer
        dtype: string
      - name: category
        dtype: string
      - name: input_ids
        list: int64
      - name: input_tokens_len
        dtype: int64
      - name: generated_texts
        dtype: string
      - name: generated_tokens_len
        dtype: int64
      - name: n_prefill_tokens
        dtype: int64
      - name: finish_reason
        dtype: string
      - name: seed
        dtype: int64
      - name: labels
        list: int64
      - name: image_fingerprint
        dtype: string
      - name: image_processor_config
        dtype: string
      - name: image_processor_config_hash
        dtype: string
      - name: sum_vision_tokens
        dtype: int64
      - name: max_vision_tokens
        dtype: int64
      - name: id
        dtype: string
      - name: truncated_by_cap
        dtype: bool
      - name: generation_model_name_or_path
        dtype: string
      - name: enable_thinking
        dtype: bool
    splits:
      - name: train
        num_bytes: 67103391
        num_examples: 100
    download_size: 63305602
    dataset_size: 67103391
configs:
  - config_name: gemma-4-26b-a4b-it-ins-gen-2048-vt280
    data_files:
      - split: train
        path: gemma-4-26b-a4b-it-ins-gen-2048-vt280/train-*
  - config_name: gemma-4-26b-a4b-it-ins-gen-2048-vt280-subsample-50
    data_files:
      - split: train
        path: gemma-4-26b-a4b-it-ins-gen-2048-vt280-subsample-50/train-*
  - config_name: gemma-4-26b-a4b-it-ins-gen-2048-vt280-subsample-100
    data_files:
      - split: train
        path: gemma-4-26b-a4b-it-ins-gen-2048-vt280-subsample-100/train-*
  - config_name: gemma-4-26b-a4b-it-think-gen-4096-vt280
    data_files:
      - split: train
        path: gemma-4-26b-a4b-it-think-gen-4096-vt280/train-*
  - config_name: gemma-4-26b-a4b-it-think-gen-4096-vt280-subsample-50
    data_files:
      - split: train
        path: gemma-4-26b-a4b-it-think-gen-4096-vt280-subsample-50/train-*
  - config_name: gemma-4-26b-a4b-it-think-gen-4096-vt280-subsample-100
    data_files:
      - split: train
        path: gemma-4-26b-a4b-it-think-gen-4096-vt280-subsample-100/train-*
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt140
    data_files:
      - split: train
        path: gemma-4-e4b-it-ins-gen-2048-vt140/train-*
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt140-subsample-50
    data_files:
      - split: train
        path: gemma-4-e4b-it-ins-gen-2048-vt140-subsample-50/train-*
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt140-subsample-100
    data_files:
      - split: train
        path: gemma-4-e4b-it-ins-gen-2048-vt140-subsample-100/train-*
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt280
    data_files:
      - split: train
        path: gemma-4-e4b-it-ins-gen-2048-vt280/train-*
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt280-subsample-50
    data_files:
      - split: train
        path: gemma-4-e4b-it-ins-gen-2048-vt280-subsample-50/train-*
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt280-subsample-100
    data_files:
      - split: train
        path: gemma-4-e4b-it-ins-gen-2048-vt280-subsample-100/train-*
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt560
    data_files:
      - split: train
        path: gemma-4-e4b-it-ins-gen-2048-vt560/train-*
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt560-subsample-50
    data_files:
      - split: train
        path: gemma-4-e4b-it-ins-gen-2048-vt560-subsample-50/train-*
  - config_name: gemma-4-e4b-it-ins-gen-2048-vt560-subsample-100
    data_files:
      - split: train
        path: gemma-4-e4b-it-ins-gen-2048-vt560-subsample-100/train-*
  - config_name: gemma-4-e4b-it-think-gen-4096-vt140
    data_files:
      - split: train
        path: gemma-4-e4b-it-think-gen-4096-vt140/train-*
  - config_name: gemma-4-e4b-it-think-gen-4096-vt140-subsample-50
    data_files:
      - split: train
        path: gemma-4-e4b-it-think-gen-4096-vt140-subsample-50/train-*
  - config_name: gemma-4-e4b-it-think-gen-4096-vt140-subsample-100
    data_files:
      - split: train
        path: gemma-4-e4b-it-think-gen-4096-vt140-subsample-100/train-*
  - config_name: gemma-4-e4b-it-think-gen-4096-vt280
    data_files:
      - split: train
        path: gemma-4-e4b-it-think-gen-4096-vt280/train-*
  - config_name: gemma-4-e4b-it-think-gen-4096-vt280-subsample-50
    data_files:
      - split: train
        path: gemma-4-e4b-it-think-gen-4096-vt280-subsample-50/train-*
  - config_name: gemma-4-e4b-it-think-gen-4096-vt280-subsample-100
    data_files:
      - split: train
        path: gemma-4-e4b-it-think-gen-4096-vt280-subsample-100/train-*
  - config_name: gemma-4-e4b-it-think-gen-4096-vt560
    data_files:
      - split: train
        path: gemma-4-e4b-it-think-gen-4096-vt560/train-*
  - config_name: gemma-4-e4b-it-think-gen-4096-vt560-subsample-50
    data_files:
      - split: train
        path: gemma-4-e4b-it-think-gen-4096-vt560-subsample-50/train-*
  - config_name: gemma-4-e4b-it-think-gen-4096-vt560-subsample-100
    data_files:
      - split: train
        path: gemma-4-e4b-it-think-gen-4096-vt560-subsample-100/train-*
  - config_name: qwen3.5-4b-ins-gen-2048
    data_files:
      - split: train
        path: qwen3.5-4b-ins-gen-2048/train-*
  - config_name: qwen3.5-4b-ins-gen-2048-subsample-50
    data_files:
      - split: train
        path: qwen3.5-4b-ins-gen-2048-subsample-50/train-*
  - config_name: qwen3.5-4b-ins-gen-2048-subsample-100
    data_files:
      - split: train
        path: qwen3.5-4b-ins-gen-2048-subsample-100/train-*
  - config_name: qwen3.5-4b-think-gen-4096
    data_files:
      - split: train
        path: qwen3.5-4b-think-gen-4096/train-*
  - config_name: qwen3.5-4b-think-gen-4096-subsample-50
    data_files:
      - split: train
        path: qwen3.5-4b-think-gen-4096-subsample-50/train-*
  - config_name: qwen3.5-4b-think-gen-4096-subsample-100
    data_files:
      - split: train
        path: qwen3.5-4b-think-gen-4096-subsample-100/train-*
  - config_name: qwen3.6-35b-a3b-ins-gen-2048
    data_files:
      - split: train
        path: qwen3.6-35b-a3b-ins-gen-2048/train-*
  - config_name: qwen3.6-35b-a3b-ins-gen-2048-subsample-50
    data_files:
      - split: train
        path: qwen3.6-35b-a3b-ins-gen-2048-subsample-50/train-*
  - config_name: qwen3.6-35b-a3b-ins-gen-2048-subsample-100
    data_files:
      - split: train
        path: qwen3.6-35b-a3b-ins-gen-2048-subsample-100/train-*
  - config_name: qwen3.6-35b-a3b-think-gen-4096
    data_files:
      - split: train
        path: qwen3.6-35b-a3b-think-gen-4096/train-*
  - config_name: qwen3.6-35b-a3b-think-gen-4096-subsample-50
    data_files:
      - split: train
        path: qwen3.6-35b-a3b-think-gen-4096-subsample-50/train-*
  - config_name: qwen3.6-35b-a3b-think-gen-4096-subsample-100
    data_files:
      - split: train
        path: qwen3.6-35b-a3b-think-gen-4096-subsample-100/train-*