| # Required for inference.py (baseline agent) | |
| # OPENROUTER_API_KEY is passed as api_key to the OpenAI client — use the token for | |
| # whatever provider matches API_BASE_URL (Hugging Face, OpenRouter, etc.). | |
| # Option A — Hugging Face router (defaults in code if env unset) | |
| # API_BASE_URL=https://router.huggingface.co/v1 | |
| # MODEL_NAME=meta-llama/Llama-3.3-70B-Instruct | |
| # OPENROUTER_API_KEY=your-huggingface-token-here | |
| # Option B — OpenRouter (e.g. Meta Llama 3.2 3B Instruct free) | |
| # API_BASE_URL=https://openrouter.ai/api/v1 | |
| # MODEL_NAME=meta-llama/llama-3.2-3b-instruct:free | |
| # OPENROUTER_API_KEY=your-openrouter-key-here | |
| # Optional: point inference at the local env | |
| ENV_URL=http://localhost:8000 | |
| # Optional: append every raw LLM response to a file (debug / audit) | |
| # INFERENCE_LOG_LLM=outputs/llm_raw.log | |
| # Optional: append API errors (e.g. 429) when the client falls back to list_tools | |
| # INFERENCE_LOG_API=outputs/api_errors.log | |
| # Optional: wait for Enter between steps to avoid rate limits (or use: python inference.py -i) | |
| # INFERENCE_INTERACTIVE=1 | |
| # INFERENCE_PAUSE=step # step = after each env step; scenario = only between easy/medium/hard | |
| # INFERENCE_SUMMARY_FILE=outputs/reward_grader_summary.txt | |
| # Optional: sampling temperature for the LLM (default 0.5; raise for more exploration on small models) | |
| # INFERENCE_TEMPERATURE=0.5 | |