# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. # SPDX-License-Identifier: Apache-2.0 proof_jury_judge: resources_servers: proof_jury_judge: entrypoint: app.py domain: math verified: false description: Reference-free IMO proof grading (0-7) by a jury of three frontier LLM judges value: Measure natural-language olympiad proof quality without reference solutions judges: - type: responses_api_models name: proof_jury_judge_openai_model - type: responses_api_models name: proof_jury_judge_google_model - type: responses_api_models name: proof_jury_judge_anthropic_model reconcile_spread: 2 # The jury: one flagship model per lab, served by an OpenAI-compatible gateway (by default # NVIDIA's inference API; set JUDGE_API_KEY). `X-Inference-Priority: batch` selects the gateway's # long-request lane: long judgments exceed the default lane's 360 s limit. proof_jury_judge_openai_model: responses_api_models: vllm_model: entrypoint: app.py base_url: ${oc.env:JUDGE_BASE_URL,https://inference-api.nvidia.com/v1} api_key: ${oc.env:JUDGE_API_KEY,dummy_api_key} model: openai/openai/gpt-6.1-sol return_token_id_information: false uses_reasoning_parser: true extra_body: reasoning_effort: high max_completion_tokens: 128000 default_headers: X-Inference-Priority: batch proof_jury_judge_google_model: responses_api_models: vllm_model: entrypoint: app.py base_url: ${oc.env:JUDGE_BASE_URL,https://inference-api.nvidia.com/v1} api_key: ${oc.env:JUDGE_API_KEY,dummy_api_key} model: gcp/google/gemini-3.1-pro-preview return_token_id_information: false uses_reasoning_parser: true extra_body: reasoning_effort: high max_completion_tokens: 65536 default_headers: X-Inference-Priority: batch proof_jury_judge_anthropic_model: responses_api_models: vllm_model: entrypoint: app.py base_url: ${oc.env:JUDGE_BASE_URL,https://inference-api.nvidia.com/v1} api_key: ${oc.env:JUDGE_API_KEY,dummy_api_key} model: azure/anthropic/claude-opus-5-5 return_token_id_information: false uses_reasoning_parser: true extra_body: max_tokens: 128000 thinking: type: enabled budget_tokens: 120000 default_headers: X-Inference-Priority: batch proof_jury_judge_simple_agent: responses_api_agents: simple_agent: entrypoint: app.py resources_server: type: resources_servers name: proof_jury_judge model_server: type: responses_api_models name: policy_model datasets: - name: example type: example jsonl_fpath: resources_servers/proof_jury_judge/data/example.jsonl license: Creative Commons Attribution 4.0 International proof_jury_judge_environment_server: environment_servers: single_agent_turn_legacy: entrypoint: app.py resources_server: type: resources_servers name: proof_jury_judge agent_server: type: responses_api_agents name: proof_jury_judge_simple_agent resources_tool_transports: - direct_http