Spaces:
Sleeping
Sleeping
| # Project: LangGraph HF Agent for GAIA | |
| # config.yaml | |
| data: "data/metadata.jsonl" # Path to the GAIA documents dataset | |
| retrievers: | |
| enable_vector_search: true # Enable vector-based document retrieval | |
| enable_keyword_search: true # Enable keyword-based document retrieval | |
| final_rrf_k: 3 # Number of top documents to consider after reciprocal rank fusion | |
| vector_store: | |
| table: "gaia_documents" # Type of vector store (e.g., faiss, chroma) | |
| query: "match_documents" # Method to query the vector store | |
| k: 10 # Number of top documents to retrieve | |
| threshold: 0.5 # Similarity threshold for document retrieval | |
| bm25: | |
| k: 5 # Number of top documents to retrieve using keyword search | |
| models: | |
| cache_folder: "./models/hf_cache" # Directory to cache Hugging Face models | |
| embeddings: | |
| model_name: "Alibaba-NLP/gte-modernbert-base" # Hugging Face embedding model ID | |
| reranker: | |
| model_name: "Alibaba-NLP/gte-reranker-modernbert-base" # Hugging Face model ID for reranking | |
| llm: | |
| model_name: "Qwen/Qwen3-32B" # Hugging Face model ID | |
| parameters: | |
| temperature: 0.6 | |
| repetition_penalty: 1.3 | |
| provider: "auto" | |
| thinking_enabled: false | |
| timeout: 300 # Read timeout (s) for the HF Inference call. Default 120 is too short under load. | |
| max_new_tokens: 4096 # Output cap. Default 512 truncates long responses and breaks tool calls. | |
| vlm: | |
| model_name: "Qwen/Qwen3-VL-32B-Instruct" # Hugging Face model ID | |
| asr: | |
| model_name: "openai/whisper-large-v3" # Hugging Face model ID — must have a provider on HF Inference Providers | |
| graph: | |
| recursion_limit: 40 # Max graph-node visits before bailing. | |
| api: | |
| base_url: "https://agents-course-unit4-scoring.hf.space" | |
| files_dir: "./data/task_files" # Local directory for downloaded task files | |