Commit History

  1. fixes circular import

    6995e18
  2. add configs to _init_.py

    63ce7d6
  3. solves import error

    c238a70
  4. non destructive dataset operations

    cf628aa
  5. adds imports and forces monitoring

    190d843
  6. adds scheduler stuff and hopes for the best with track tonic

    d455d12
  7. adds local and remote training monitors to config

    71db310
  8. improve dataset utils non destructive writes

    924581c
  9. adds parameters to medical config

    81f39f1
  10. adds parameters to medical config

    b4f1cb3
  11. adds parameters to medical config

    c68717e
  12. adds improved dev and system prompts to o1-medical

    fc29a51
  13. adds improved launch for reasoning gpt-oss configs and new config for medical reasoning

    0ded6bb
  14. adds improved dataset utils in tracktonic

    b11b94b
  15. adds optimizations for faster training

    3331c7f
  16. adds a100 memory optimized

    fa9560d
  17. adds template files , adds non destructive dataset updates

    d47568c
  18. adds flash attention 3 kernel

    cb276d8

  1. adapt lr_scheduler according to trl version

    976e218
  2. sets min_lr

    7f45871
  3. increases batchsize and gradient accumulation steps in memory optimized

    4e59f6d
  4. increases max samples in memory optimized

    8b56686
  5. adds better launch.sh and eval / test splits auto

    0fa6045
  6. adds repoid only based on repo name, adds version-robust sfttrainer

    665844a
  7. coerce akk numeric config values to safe values

    c346dad
  8. adds defensive programming (boo) and adaptations based on transformer versions

    97dacc7
  9. improves launcher with model family and defaults based on options, updates trl trainer , removes trl config paths by switching to trainingarguments class , tokenizer parameters updated to sfttrainer , resolves evaluation_strategy error

    598357a
  10. hide all tokens in logs, never persist to disk, remove max_seq_length from config, add to trainer

    eb9e91f
  11. changes default model repo name + adds non default option

    c23e2f5
  12. adds pythonpath based on chatgpt5 suggestion

    c3ab72f

  1. adds harmony format , configurable gpt-oss parameters, launch.sh logic , improved templates for legml gpt-oss training, dynamic results directory and improve model pushing

    59e57ff
  2. adds single token logic read/write , adds gpt-oss demo space , adds spaces refactor , adds new version of track tonic , adds logic in launch.sh

    75bcdb3

  1. adds quantization configuration correctly

    c7cffbb
  2. adds quantization configuration correctly

    fa7de39
  3. adds memory optimized configuration

    7181190
  4. fixes custom trackio implementation

    dfcb060
  5. adds gpt-oss support

    fcf2981

  1. adds readme links at the top

    ce0d824
  2. adds small readme improvements

    a8275b3
  3. Update README.md⁠

    Unverified signature
    26641fd
  4. fixes typo in mermaid diagram links

    a552387
  5. adds readme, removes quantization, adds readtoken logic, updates trackio , spaces

    3c37508

  1. cleanup a bit the files

    Unverified signature
    ad3b15d
  2. workaround for quantization and push

    Tonic
    Unverified signature
    2432208
  3. workaround for quantization and push

    Tonic
    Unverified signature
    b79fab9
  4. adds quantize and push script

    Tonic
    Unverified signature
    e6ad96a
  5. solves model card formatting bug

    Tonic
    Unverified signature
    41e9e02
  6. solves model card formatting bug

    Tonic
    Unverified signature
    2f866e6
  7. Fix model recovery and deployment scripts - add safetensors support and Windows compatibility

    Tonic
    Unverified signature
    22bb04c
  8. Fix model recovery and deployment scripts - add safetensors support and Windows compatibility

    Tonic
    Unverified signature
    fff73fc