# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved. # SPDX-License-Identifier: Apache-2.0 video_caption: resources_servers: video_caption: entrypoint: app.py num_workers: 1 judge_model_server: {type: responses_api_models, name: video_caption_judge} judge_chat_completion_create_params: messages: [] max_tokens: 20480 reasoning_effort: low caption_tokenizer_path: ${oc.env:VIDEO_CAPTION_TOKENIZER_PATH,MISSING} judge_endpoint_max_concurrency: 256 judge_max_retries: 2 judge_attempt_timeout_seconds: 1200 min_caption_tokens: 100 max_caption_tokens: 4000 visual_reward_type: recall penalize_degeneration: false domain: other verified: false description: Visual event recall, precision, or F1 with length and optional degeneration reward value: Improve coverage and accuracy of visible events in video captions # Video caption reward with an OpenAI-compatible API judge (no judge GPUs). Same server settings as # video_caption.yaml; only the judge model server differs. The judge must support # chat.completions with a JSON-schema `response_format`. video_caption_judge: responses_api_models: openai_model: entrypoint: app.py openai_base_url: ${oc.env:VIDEO_CAPTION_JUDGE_BASE_URL,MISSING} openai_api_key: ${oc.env:VIDEO_CAPTION_JUDGE_API_KEY,MISSING} openai_model: ${oc.env:VIDEO_CAPTION_JUDGE_MODEL,MISSING} video_caption_simple_agent: responses_api_agents: simple_agent: entrypoint: app.py max_steps: 1 resources_server: {type: resources_servers, name: video_caption} model_server: {type: responses_api_models, name: policy_model} datasets: - name: example type: example jsonl_fpath: resources_servers/video_caption/data/example.jsonl num_repeats: 1 video_caption_environment_server: environment_servers: single_agent_turn_legacy: entrypoint: app.py agent_server: type: responses_api_agents name: video_caption_simple_agent resources_server: type: resources_servers name: video_caption resources_tool_transports: - direct_http