ProjetEngineering/coach-equestre-ai

fix: Increase max_new_tokens to 512 for complete responses

fix: Restore max_new_tokens to 350 to prevent truncated responses

build: Force Docker rebuild to apply CUDA OOM fixes

fix: Resolve CUDA OOM and Gemini model name

feat: Replace keyword classifier with Gemini Flash LLM Router

feat: Add source tracking to RAG responses

feat: Add intelligent RAG classifier to detect regulation questions

fix: Enable PyTorch expandable_segments to prevent GPU OOM

fix: Upgrade qdrant-client to 1.12.1 for Qdrant Cloud compatibility

feat: Integrate Qdrant RAG for FEI documentation (v3.0.0)

Fix: Improve prompt to reduce repetition and prevent truncation

Sync system prompt with proper formatting rules

Revert to float16 without quantization (stable approach)

Add bitsandbytes dependency for 8-bit quantization

Fix: Use BitsAndBytesConfig for proper 8-bit quantization

Optimize GPU memory: Use 8-bit quantization instead of float16

Fix: Add offload_folder=/tmp to prevent offloading error

Fix: Remove RAG references from Dockerfile

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

Chore: Force rebuild for L4 GPU deployment

Fix: Force OMP_NUM_THREADS='' to override Kubernetes millicores value

Fix: Set OMP_NUM_THREADS=1 to prevent OpenMP errors during indexation

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

Add: RAG diagnostic endpoint for debugging imports

Debug: Show full RAG import error in logs

Add: RAG testing endpoints

Fix: Use legacy-resolver to handle transformers 5.x vs sentence-transformers conflict

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

Track PDFs with Git LFS

Add: RAG System (Phase 1) - Règlements FEI

ok indentation

Fix GPU device_map for T4

Fix: Remove bitsandbytes (incompatible with CPU)

Fix: Support CPU mode with 8-bit quantization for free tier

Fix: Support CPU mode with 8-bit quantization

Fix: Use AutoPeftModelForCausalLM for LoRA loading

Use Transformers 5.0.0rc0 + PEFT 0.18.0 for LoRA compatibility

Fix dependencies: Transformers 4.57.6 + PEFT 0.17.1 + Python 3.11

Load LoRA model from Hugging Face Hub

Upload 19 files

Upload adapter_model.safetensors

initial commit