Gym/responses_api_agents/swe_agents/configs/swebench_swe_agent.yaml at main · pthombre/Gym · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
# SWE-bench wrapper configuration for SWE-agent
swe_agents:
  responses_api_agents:
    swe_agents:
      entrypoint: app.py
      domain: coding
      description: Software engineering tasks with OpenHands agent harness.
      value: Improve agentic software engineering capabilities.

      # Agent framework configuration
      agent_framework: swe_agent
      agent_config: configs/swe_agent_config.yaml
      agent_max_turns: 100
      agent_tools_file: configs/swe_agent_tools_openai_format.json  # Used for response formatting only
      agent_framework_repo: https://github.com/slikhite-1/nv-SWE-agent.git
      agent_framework_commit: 34755442cb91fe0500a6dcb0485b91e899812f90  # pragma: allowlist secret

      # Container configuration
      container_formatter: "/lustre/fsw/portfolios/llmservice/users/igitman/images/swe-bench/swebench_sweb.eval.x86_64.{instance_id}.sif"
      container_folder_path: null
      swebench_tests_timeout: 900  # 30 minutes
      concurrency: 512

      # Optional model server reference
      model_server:
        name: policy_model # openai_model
        type: responses_api_models

      # Additional NeMo-Skills configuration
      nemo_skills_config:
        max_samples: -1
        skip_filled: false
        dry_run: false
        max_concurrent_requests: 1
        inference.top_logprobs: 10  # Request logprobs to get token IDs for training

      # Dataset configurations
      datasets:
        # Training dataset
        - name: train
          type: train
          jsonl_fpath: responses_api_agents/swe_agents/data/swegym_for_sweagent_and_openhands.jsonl
          gitlab_identifier:
            dataset_name: swegym_for_sweagent_and_openhands
            version: 0.0.1
            artifact_fpath: swegym_for_sweagent_and_openhands.jsonl
          license: MIT

        # Validation dataset
        - name: validation
          type: validation
          jsonl_fpath: responses_api_agents/swe_agents/data/swebench_verified_for_sweagent_and_openhands.jsonl
          gitlab_identifier:
            dataset_name: swebench_verified_for_sweagent_and_openhands
            version: 0.0.1
            artifact_fpath: swebench_verified_for_sweagent_and_openhands.jsonl
          license: TBD

        # Example dataset for quick testing
        - name: example
          type: example
          jsonl_fpath: responses_api_agents/swe_agents/data/example.jsonl