Hugging Face's logo

Spaces:
Pandaisop
/
codesensei-env
Runtime error

App Files Community
codesensei-env / training

Commit History

  1. revert(training): temporarily revert dataset bounds back to 10 for complete trial confirmation before scaling

    vineetshukla.work@gmail.com
    6fd478f

  2. feat(training): scale dataset to 500 prompts and add hub export cell for full hackathon training run

    vineetshukla.work@gmail.com
    8b03007

  3. fix(trl): inject warnings_issued dict into unsloth model to prevent TRL GRPO init crash on PeftModels

    vineetshukla.work@gmail.com
    0413c85

  4. fix(trl): force hardware-specific fp16/bf16 flags to prevent TRL from crashing on T4 GPU

    vineetshukla.work@gmail.com
    6262683

  5. fix(trl): mock weave.trace to complete package bypass

    vineetshukla.work@gmail.com
    6f6568f

  6. fix(trl): add weave to the magic mock bypass list for cell 5

    vineetshukla.work@gmail.com
    7ea8558

  7. fix(trl): inject magicmock bypass into cell 5 to squash TRL 0.24.0 import crash

    vineetshukla.work@gmail.com
    e0923c4

  8. feat: complete rewrite of grpo training script using Unsloth optimized pathway

    vineetshukla.work@gmail.com
    aa86916

  9. chore: test trial run for colab script with 10 dataset size and fixed dependencies

    vineetshukla.work@gmail.com
    a59d79f

  1. fix: downgrade TRL to 0.15.0 and nuke ghost vLLM install to prevent import crashes

    vineetshukla.work@gmail.com
    b119c82

  2. fix: correct trl version pin to 0.17.0

    vineetshukla.work@gmail.com
    a46eca2

  3. fix: vLLM/TRL version conflict in Colab Cell 1

    vineetshukla.work@gmail.com
    0b9cc90

  4. fix: training reward bounds to (0.01, 0.99) — prevents loss=0 from zero variance

    vineetshukla.work@gmail.com
    545acf0

  1. fix: increase num_generations=6, temperature=0.9, lr=2e-5 to fix zero-loss GRPO

    vineetshukla.work@gmail.com
    b3bf487

  2. fix: remove fp16=True, use paged_adamw_8bit for QLoRA BFloat16 compatibility

    vineetshukla.work@gmail.com
    038f1b4

  3. fix: add 4-bit quantization + LoRA to fit GRPO training in T4 15GB VRAM

    vineetshukla.work@gmail.com
    90d6d56

  1. refactor: rewrite training script - remove rollout_func, use inline reward evaluation⁠

    vineetshukla.work@gmail.com
    d81cba7

  2. fix: remove deprecated GRPOConfig params for latest TRL compatibility

    vineetshukla.work@gmail.com
    0589e5e

  3. fix: set live HF Space URL in training script and demo app

    vineetshukla.work@gmail.com
    d12dd6e

  4. feat: CodeSensei - GRPO-trained LLM code debugger on OpenEnv⁠

    vineetshukla.work@gmail.com
    c47c81c