Skip to content

Make SFT metadata filters configurable for short-task datasets#115

Open
hans-yang-tps wants to merge 1 commit into
NVIDIA:mainfrom
hans-yang-tps:data/sft-metadata-filter-knobs
Open

Make SFT metadata filters configurable for short-task datasets#115
hans-yang-tps wants to merge 1 commit into
NVIDIA:mainfrom
hans-yang-tps:data/sft-metadata-filter-knobs

Conversation

@hans-yang-tps

Copy link
Copy Markdown

Summary

  • Make SFT converter metadata filters configurable with min_window_frames=None and max_duration_s=None.
  • Plumb the same knobs through SFT and Transfer SFT metadata loaders so converter and training counts stay aligned.
  • Add regression coverage for disabling duration/window-frame prefilters for short-task datasets.

Testing

  • pytest -q cosmos_framework/scripts/curator_to_sft_jsonl_test.py cosmos_framework/scripts/captions_to_sft_jsonl_test.py cosmos_framework/data/generator/local_datasets/sft_dataset_caption_test.py cosmos_framework/data/generator/local_datasets/transfer_sft_dataset_test.py --disable-warnings
  • ruff check cosmos_framework/scripts/curator_to_sft_jsonl.py cosmos_framework/scripts/curator_to_sft_jsonl_test.py cosmos_framework/scripts/captions_to_sft_jsonl.py cosmos_framework/scripts/captions_to_sft_jsonl_test.py cosmos_framework/data/generator/local_datasets/sft_dataset.py cosmos_framework/data/generator/local_datasets/sft_dataset_caption_test.py cosmos_framework/data/generator/local_datasets/transfer_sft_dataset.py cosmos_framework/data/generator/local_datasets/transfer_sft_dataset_test.py
  • ruff format --check cosmos_framework/scripts/curator_to_sft_jsonl.py cosmos_framework/scripts/curator_to_sft_jsonl_test.py cosmos_framework/scripts/captions_to_sft_jsonl.py cosmos_framework/scripts/captions_to_sft_jsonl_test.py cosmos_framework/data/generator/local_datasets/sft_dataset.py cosmos_framework/data/generator/local_datasets/sft_dataset_caption_test.py cosmos_framework/data/generator/local_datasets/transfer_sft_dataset.py cosmos_framework/data/generator/local_datasets/transfer_sft_dataset_test.py

@hans-yang-tps
hans-yang-tps marked this pull request as ready for review July 17, 2026 07:53
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant