beam: resources_servers: beam: entrypoint: app.py domain: long_context verified: false description: BEAM (Beyond a Million Tokens) long-term-memory probing questions over long multi-session chats, scored by an LLM judge with the upstream per-question rubrics. value: Improve long-term conversational memory (recall, updates, contradictions, temporal and multi-session reasoning). judge_model_server: type: responses_api_models name: judge_model judge_responses_create_params: input: [] # Default judge is GPT-5.4 at temperature 0 (upstream used gpt-4.1-mini). The # judge returns a short JSON object ({"score", "reason"}); 2048 leaves headroom for a # reasoning-capable judge. max_output_tokens: 2048 temperature: 0.0 judge_endpoint_max_concurrency: 64 # Retries on 429/5xx happen in the judge_model server (openai_model). A judge call # that still fails masks the row and sends it to the failures sidecar # (rerun with `gym eval reverify --judge-failed-only`). # binary: pass / fail per rubric item (default); partial: upstream 0 / 0.5 / 1. judge_scale: binary # Reward per question: mean over rubric items (upstream; event ordering # uses tau_norm) or all_pass (1 only if every item scores 1). Both are # always reported as mean/overall and all_pass/overall. question_scoring: mean judge_model: responses_api_models: openai_model: entrypoint: app.py openai_base_url: ${judge_base_url} openai_api_key: ${judge_api_key} openai_model: ${judge_model_name} beam_simple_agent: responses_api_agents: simple_agent: entrypoint: app.py resources_server: type: resources_servers name: beam model_server: type: responses_api_models name: policy_model datasets: - name: example type: example jsonl_fpath: resources_servers/beam/data/example.jsonl license: Creative Commons Attribution-ShareAlike 4.0 International # Judge defaults: GPT-5.4 on the NVIDIA inference API (upstream used gpt-4.1-mini). # Pass +judge_api_key=$INFERENCE_API_KEY. Must stay at end of file for # update_env_list.py. policy_* are intentionally NOT defaulted here. judge_base_url: https://inference-api.nvidia.com/v1 judge_api_key: null judge_model_name: azure/openai/gpt-5.4 beam_environment_server: environment_servers: single_agent_turn_legacy: entrypoint: app.py resources_server: type: resources_servers name: beam agent_server: type: responses_api_agents name: beam_simple_agent resources_tool_transports: - direct_http