Commit History

  1. fix: Increase max_new_tokens to 512 for complete responses⁠

    d585da8
  2. fix: Restore max_new_tokens to 350 to prevent truncated responses⁠

    c7fea88

  1. build: Force Docker rebuild to apply CUDA OOM fixes⁠

    af5d8b7
  2. fix: Resolve CUDA OOM and Gemini model name⁠

    f683310
  3. feat: Replace keyword classifier with Gemini Flash LLM Router⁠

    84f99b1
  4. feat: Add source tracking to RAG responses⁠

    NassimAI7 +1: Claude Sonnet 4.5
    9b8892c
  5. feat: Add intelligent RAG classifier to detect regulation questions⁠

    NassimAI7 +1: Claude Sonnet 4.5
    abd25b8
  6. fix: Enable PyTorch expandable_segments to prevent GPU OOM⁠

    NassimAI7 +1: Claude Sonnet 4.5
    fe94c16
  7. fix: Upgrade qdrant-client to 1.12.1 for Qdrant Cloud compatibility⁠

    NassimAI7 +1: Claude Sonnet 4.5
    1d70cc4
  8. feat: Integrate Qdrant RAG for FEI documentation (v3.0.0)⁠

    NassimAI7 +1: Claude Sonnet 4.5
    c2bd22a

  1. Fix: Improve prompt to reduce repetition and prevent truncation⁠

    NassimAI7 +1: Claude Sonnet 4.5
    317134a
  2. Sync system prompt with proper formatting rules⁠

    NassimAI7 +1: Claude Sonnet 4.5
    87d11c6
  3. Revert to float16 without quantization (stable approach)⁠

    NassimAI7 +1: Claude Sonnet 4.5
    f605cd2
  4. Add bitsandbytes dependency for 8-bit quantization⁠

    NassimAI7 +1: Claude Sonnet 4.5
    2180af7
  5. Fix: Use BitsAndBytesConfig for proper 8-bit quantization⁠

    NassimAI7 +1: Claude Sonnet 4.5
    c135940
  6. Optimize GPU memory: Use 8-bit quantization instead of float16⁠

    NassimAI7 +1: Claude Sonnet 4.5
    e3ec1de
  7. Fix: Add offload_folder=/tmp to prevent offloading error⁠

    c18d818
  8. Fix: Remove RAG references from Dockerfile⁠

    a97a385
  9. Remove RAG system: Clean up all RAG-related files and dependencies⁠

    c9ad8b3
  10. Chore: Force rebuild for L4 GPU deployment

    AutomationAuditsEnergies
    078715f
  11. Fix: Force OMP_NUM_THREADS='' to override Kubernetes millicores value

    AutomationAuditsEnergies
    bcd5ecb
  12. Fix: Set OMP_NUM_THREADS=1 to prevent OpenMP errors during indexation

    AutomationAuditsEnergies
    a70cf8b
  13. Fix: Copy RAG modules to Docker image + pin numpy<2.0 for ChromaDB compatibility

    AutomationAuditsEnergies
    3ccde62
  14. Add: RAG diagnostic endpoint for debugging imports

    AutomationAuditsEnergies
    ad42a79
  15. Debug: Show full RAG import error in logs

    AutomationAuditsEnergies
    b763fc1
  16. Add: RAG testing endpoints⁠

    AutomationAuditsEnergies
    983aa0c
  17. Fix: Use legacy-resolver to handle transformers 5.x vs sentence-transformers conflict⁠

    AutomationAuditsEnergies
    bb77699
  18. Fix: Upgrade sentence-transformers to 3.3.1 (compatible with transformers 5.x)

    AutomationAuditsEnergies
    5e191cd
  19. Track PDFs with Git LFS

    AutomationAuditsEnergies
    f3a2b9a
  20. Add: RAG System (Phase 1) - Règlements FEI⁠

    AutomationAuditsEnergies
    1ad484c

  1. ok indentation

    Unverified signature
    dc39c09
  2. Fix GPU device_map for T4

    Unverified signature
    3fc6a7d
  3. Fix: Remove bitsandbytes (incompatible with CPU)⁠

    AutomationAuditsEnergies
    6229042
  4. Fix: Support CPU mode with 8-bit quantization for free tier⁠

    AutomationAuditsEnergies
    73ba329
  5. Fix: Support CPU mode with 8-bit quantization⁠

    AutomationAuditsEnergies +1: Claude Sonnet 4.5
    9f8397c
  6. Fix: Use AutoPeftModelForCausalLM for LoRA loading⁠

    AutomationAuditsEnergies +1: Claude Sonnet 4.5
    09b6051
  7. Use Transformers 5.0.0rc0 + PEFT 0.18.0 for LoRA compatibility⁠

    AutomationAuditsEnergies
    81878b5
  8. Fix dependencies: Transformers 4.57.6 + PEFT 0.17.1 + Python 3.11⁠

    AutomationAuditsEnergies
    0dda0fe
  9. Load LoRA model from Hugging Face Hub⁠

    AutomationAuditsEnergies
    afadeaf
  10. Upload 19 files

    Unverified signature
    6325a29
  11. Upload adapter_model.safetensors

    Unverified signature
    205d804
  12. initial commit

    Unverified signature
    80048eb