Commit History

  1. update: results table, 0.5B model links, citation year 2026

    Jayant-Kernel
    293f2e4
  2. docs: detailed README with curriculum, reward table, results, usage

    Jayant-Kernel
    a7c6973
  3. rollback: revert to last working Dockerfile and train.py

    Jayant-Kernel
    Unverified signature
    e30d685
  4. fix: proper GRPO with trl 0.12.2 no-deps + force hub downgrade

    Jayant-Kernel
    Unverified signature
    0efac4a
  5. fix: custom training loop without TRL dependency

    Jayant-Kernel
    Unverified signature
    5232a98
  6. fix: force reinstall huggingface_hub 0.24.7 after deceit_env

    Jayant-Kernel
    Unverified signature
    54fc539
  7. fix: pin huggingface_hub 0.24.7, install trl with --no-deps

    Jayant-Kernel
    Unverified signature
    a0058bb
  8. fix: trl 0.12.2 has GRPOTrainer, pin all deps before trl install

    Jayant-Kernel
    Unverified signature
    430098b
  9. fix: try multiple import paths for GRPOConfig

    Jayant-Kernel
    Unverified signature
    2cdce1f
  10. fix: install transformers 4.46.0 BEFORE trl so trl doesnt upgrade it

    Jayant-Kernel
    Unverified signature
    9264b56
  11. fix: bust docker cache force reinstall trl 0.11.4

    Jayant-Kernel
    Unverified signature
    e9971fb
  12. fix: trl 0.11.4 + transformers 4.46.0 + processing_class

    Jayant-Kernel
    Unverified signature
    e8f541c
  13. fix: trl 0.9.4 + transformers 4.41.2 compatible versions

    Jayant-Kernel
    Unverified signature
    e48f580
  14. fix: add torch version check in Dockerfile

    Jayant-Kernel
    Unverified signature
    391a47a
  15. fix: remove tokenizer arg from GRPOTrainer

    Jayant-Kernel
    Unverified signature
    f3d865a
  16. fix: tokenizer not processing_class, torch cu121 for GPU

    Jayant-Kernel
    Unverified signature
    56567fd
  17. fix: correct trl version with GRPOConfig

    Jayant-Kernel
    Unverified signature
    83f6afa
  18. fix: trl 0.13.0, remove verify steps

    Jayant-Kernel
    Unverified signature
    29f2767
  19. fix: cu124 not cu118 for A100 CUDA 12.9 driver

    Jayant-Kernel
    Unverified signature
    74138e3
  20. fix: trl 0.12.2 + torch 2.4.0

    Jayant-Kernel
    Unverified signature
    bc4c6b4
  21. fix: CPU fallback when no GPU detected

    Jayant-Kernel
    Unverified signature
    4c4c68a
  22. fix: trl 0.15.0 definitely has GRPOConfig

    Jayant-Kernel
    Unverified signature
    1c058a2
  23. fix: trl 0.9.6 + bitsandbytes 0.43.1 cu118

    Jayant-Kernel
    Unverified signature
    787e377
  24. fix: trl 0.8.6 has GRPOConfig, compatible with torch 2.1.2

    Jayant-Kernel
    4f33e83
  25. fix: remove multiline python heredoc from Dockerfile

    Jayant-Kernel
    6452e7e
  26. fix: find deceit_env package location and copy data correctly

    Jayant-Kernel
    11baf5d
  27. fix: back to python:3.10-slim for GPU, fix deceit_env path

    Jayant-Kernel
    1058c6b
  28. fix: deceit_env module path and PYTHONPATH

    Jayant-Kernel
    845f95d
  29. fix: nvidia cuda base with python3.10 installed

    Jayant-Kernel
    cbaf9f7
  30. fix: use huggingface transformers-pytorch-gpu base image

    Jayant-Kernel
    73c82af
  31. fix: revert to torch 2.1.0 cu121 with trl 0.7.4 - versions that worked before

    Jayant-Kernel
    10648d1
  32. fix: simplify dockerfile no version pinning

    Jayant-Kernel
    bcc84d6
  33. fix: accelerate 0.34.2 exists, 0.35.0 does not

    Jayant-Kernel
    09ab990
  34. fix: trl 0.12.0 has GRPOTrainer, compatible with torch 2.4.0

    Jayant-Kernel
    84d05af
  35. fix: pin trl 0.11.0 compatible with torch 2.4.0

    Jayant-Kernel
    3bced27
  36. fix: upgrade torch to 2.4.0 with CUDA 12.4 support

    Jayant-Kernel
    0862a5f
  37. fix: run train.py instead of evaluate.py

    Jayant-Kernel
    32e8cc3
  38. improve: abstention penalty, better prompt, mixed curriculum, more steps

    Jayant-Kernel
    253d1ff
  39. evaluate: switch to 0.5B model comparison, 200 episodes

    Jayant-Kernel
    6b64fd2
  40. fix: set N_EPISODES=200 constant (was still 30)

    Jayant-Kernel
    e662a77

  1. update: increase evaluation to 200 episodes per model

    Jayant-Kernel
    a5be204
  2. fix: parse_action confidence bug, numeric answers bug, missing reasoning field bug

    Jayant-Kernel
    66bdd16
  3. fix: debug model output parsing in evaluation

    Jayant-Kernel
    3d9195a
  4. add: evaluate 1.5B base vs trained, upload chart to HF Hub

    Jayant-Kernel
    77e0352
  5. update: 500 steps L1 + 300 steps L2, higher lr for 1.5B

    Jayant-Kernel
    f788873
  6. fix: add matplotlib, split COPY into separate lines

    Jayant-Kernel
    354d3fd
  7. fix: copy evaluate.py into Docker image

    Jayant-Kernel
    88fb03e
  8. add: evaluate 1.5B base vs trained, upload charts

    Jayant-Kernel
    68e5af2
  9. fix: remove misplaced import inside GRPOConfig args

    Jayant-Kernel
    e4aea5d
  10. fix: auto-detect bf16 support

    Jayant-Kernel
    d34e286