| PROJECT_NAME=KnowledgeHub |
| VERSION=1.0.0 |
| API_V1_STR=/api/v1 |
|
|
| QDRANT_URL=http://localhost:6333 |
| QDRANT_API_KEY= |
| QDRANT_COLLECTION_NAME=knowledge_base |
|
|
| NEMOTRON_PARSE_MODEL=Qwen/Qwen2-VL-2B-Instruct |
| NEMOTRON_EMBED_MODEL=nvidia/llama-nemotron-colembed-vl-3b-v2 |
| EMBEDDING_DEVICE=cpu |
| HF_TOKEN= |
| NVIDIA_API_KEY= |
| NVIDIA_CHAT_MODEL=nvidia/nvidia-nemotron-nano-9b-v2 |
| NVIDIA_API_URL=https://integrate.api.nvidia.com/v1 |
|
|
| MAX_UPLOAD_SIZE=10485760 |
| CHUNK_SIZE=1100 |
| CHUNK_OVERLAP=180 |
| ZEROGPU_DURATION_SECONDS=180 |
| CHAT_TEMPERATURE=0.6 |
| CHAT_TOP_P=0.95 |
| CHAT_MAX_TOKENS=2048 |
| MIN_THINKING_TOKENS=1024 |
| MAX_THINKING_TOKENS=2048 |
|
|
