Spaces:
Sleeping
Sleeping
| # Docker Spaces > Settings > Secrets | |
| API_TOKEN=replace-with-random-long-secret | |
| # OpenAI-compatible APIs, not local model weights. | |
| # Model IDs below are examples of REAL models on HF Inference Providers; | |
| # check current availability and price in the provider catalog. | |
| FAST_LLM_BASE_URL=https://router.huggingface.co/v1 | |
| FAST_LLM_MODEL=openai/gpt-oss-20b:fastest | |
| FAST_LLM_KEY_ENV=HF_TOKEN | |
| STRONG_LLM_BASE_URL=https://router.huggingface.co/v1 | |
| STRONG_LLM_MODEL=openai/gpt-oss-120b:fastest | |
| STRONG_LLM_KEY_ENV=HF_TOKEN | |
| # Add HF_TOKEN as a Space SECRET; no credentials in files. | |
| # Supply current provider-specific per-1-million-token prices; never assume example pricing. | |
| FAST_INPUT_USD_PER_1M= | |
| FAST_OUTPUT_USD_PER_1M= | |
| STRONG_INPUT_USD_PER_1M= | |
| STRONG_OUTPUT_USD_PER_1M= | |
| # External retrieval service - POST {query,top_k} -> {documents:[{id,title,text,score}]} | |
| RETRIEVAL_URL=https://your-retriever.example/search | |
| RETRIEVAL_API_KEY= | |
| # Enter per-search retrieval charge, or explicitly set 0 if retrieval is free. | |
| RETRIEVAL_USD_PER_CALL= | |
| # A calibrated table for these particular models and retriever, generated by fit_policy.py | |
| # CALIBRATION_FILE=config/policy.json # Enable only after uploading calibrated policy.json | |
| ACTIONS_FILE=config/actions.json | |
| MAX_INFLIGHT_REQUESTS=64 | |
| HTTP_MAX_CONNECTIONS=128 | |
| QUEUE_WAIT_SEC=2 | |
| BACKEND_TIMEOUT_SEC=90 | |