Hugging Face's logo

zotowata
/
pc-sho-dlm-code

Model card Files
xet
 Community
pc-sho-dlm-code / train_2b.py

Commit History

  1. Allow checkpoint resume across shorter context configs

    zotowata
    Verified signature
    99e3c47

  2. Add eval, sharding, and curriculum controls

    zotowata
    Verified signature
    fc1c409

  3. Add buffered shuffle to pretrain streams

    zotowata
    Verified signature
    b6a7818

  1. Add TensorBoard training metrics

    zotowata
    Verified signature
    4b1a5dd

  2. Use SmolLM tokenizer by default and validate vocab

    zotowata
    Verified signature
    73c832f

  3. Sync 2B training job files

    zotowata
    Verified signature
    2a5274f

  4. Multi-GPU via Local SGD: split batch across GPUs, average params each step⁠

    Zae +1: Claude Opus 4.6 (1M context)
    f2306ae

  5. Remove DataParallel (incompatible with unified mode in-place updates)⁠

    Zae +1: Claude Opus 4.6 (1M context)
    e20d10c

  1. Upload train_2b.py with huggingface_hub

    zotowata
    Unverified signature
    e0cb0d3