/opt/chainguard/nemo-gym/venv/bin/nemo_gym_e2e_collect_rollouts

⚠  `nemo_gym_e2e_collect_rollouts` is deprecated and will be removed in a future release; use `gym eval run` instead.
usage: gym eval run [-h] [-v] [--config PATH] [--benchmark NAME] [--environment NAME]
                    [--resources-server NAME] [--model-type NAME] [--search-dir DIR] [--no-serve]
                    [--resume] [--agent AGENT] [--input INPUT] [--output OUTPUT] [--limit LIMIT]
                    [--num-repeats NUM_REPEATS] [--prompt-config PROMPT_CONFIG]
                    [--concurrency CONCURRENCY] [--split SPLIT] [--model MODEL]
                    [--model-url MODEL_URL] [--model-api-key MODEL_API_KEY]
                    [--temperature TEMPERATURE] [--top-p TOP_P]
                    [--max-output-tokens MAX_OUTPUT_TOKENS] [--disable-aggregation]

Collate data, start servers, and collect rollouts.

options:
  -h, --help            show this help message and exit
  -v, --verbose         Set logging level to DEBUG.
  --config PATH         Config file to load; repeatable. Maps to +config_paths=[...].
  --benchmark NAME      Load the named benchmark config.
  --environment NAME    Load the named environment config.
  --resources-server NAME
                        Load the named resources-server config.
  --model-type NAME     Load the named model-type config.
  --search-dir DIR      Extra root directory to search for components; repeatable.
  --no-serve            Collect against already-running servers instead of starting them.
  --resume              Resume from cached rollouts instead of recollecting.
  --agent AGENT, -a AGENT
                        Agent to collect rollouts with.
  --input INPUT, -i INPUT
                        Input tasks JSONL file.
  --output OUTPUT, -o OUTPUT
                        Output rollouts JSONL file.
  --limit LIMIT         Maximum number of tasks to run.
  --num-repeats NUM_REPEATS
                        Number of rollouts per task.
  --prompt-config PROMPT_CONFIG
                        Prompt template YAML to apply.
  --concurrency CONCURRENCY
                        Maximum number of concurrent samples.
  --split SPLIT         Dataset split to use (train, validation, or benchmark).
  --model MODEL, -m MODEL
                        Model name, HF id, or local checkpoint path (interpreted per --model-
                        type).
  --model-url MODEL_URL
                        Model server base URL.
  --model-api-key MODEL_API_KEY
                        Model server API key.
  --temperature TEMPERATURE
                        Sampling temperature.
  --top-p TOP_P         Nucleus sampling top-p.
  --max-output-tokens MAX_OUTPUT_TOKENS
                        Maximum output tokens.
  --disable-aggregation
                        Skip post-run aggregate-metrics computation. Use with gym eval aggregate
                        for sharded jobs.