CHANGELOG.md

Changelog

All changes we make to the assignment code or PDF will be documented in this file.

[26.0.1] - 2026-05-07

Changed

• Automatically make folder for local data

[26.0.0] - 2026-05-06

Changed

• Replaced raw 5,000-WET filtering with English-only WET files from 2,500 raw WETs.
• Update final training to 8 B200 GPUs, 16,384 steps and 8.6B tokens.

[1.0.4] - 2025-05-19

Changed

• code: Halve training tokens for the leaderboard run

[1.0.3] - 2025-05-18

Changed

• code: update Paloma validation set file name to tokenized_paloma_c4_100_domains_validation.bin, as it is a binary file loaded with np.fromfile("/data/paloma/tokenized_paloma_c4_100_domains_validation.bin", dtype=np.uint16)
• handout: add guidance to load the validation set with np.fromfile, and update references to new file name

[1.0.2] - 2025-05-12

Changed

• code: update README.md to clarify that students should use the provided training script, not their own train script
• code: update dependencies (pyproject.toml and uv.lock) with packages for WARC processing: fastwarc and tldextract
• handout: add hint to use fastwarc for WARC record iteration earlier in assignment
• handout: fix Together cluster paths to hatespeech and nsfw classifiers

[1.0.1] - 2025-05-11

Changed

• handout: clarify that students should use the provided training script, not their own train script
• handout: change references to WARC files in the final filtering step to WET files
• handout: provide hints on helpful classes to process the WET files

[1.0.0] - 2025-05-07

Added

• code: script to get all assets

Changed

• code: improve supplied training script
• handout: update data to 2025
• handout: use WET files instead of WARC files for most tasks
• code: update deployment to use uv

[0.0.4] - 2024-05-29

Added

• handout: make sure to specify in problem train_model that we provide a training script.

Changed

Fixed

[0.0.3] - 2024-05-26

Added

Changed

Fixed

• handout: add --device cuda to training command

[0.0.2] - 2024-05-19

Added

• handout: added usage example for parallelism with concurrent.futures and submitit.
• handout: added points to each of the problems

Changed

Fixed

[0.0.1] - 2024-05-14

Added

Changed

Fixed

• code: fix type signature of run_mask_emails, run_mask_phone_numbers, and run_mask_ips adapters.
• code: fix expected labels in NSFW classifier test.
• handout: fix typo in mention of adapter run_classify_quality for problem quality_classifier.
• handout: fix link to Dolma NSFW and hatespeech classifiers, since the HF links point to the same model binary

[0.0.0] - 2024-05-10

Initial release.