Gym/resources_servers/reasoning_gym/configs/reflection_agent.yaml at main · NousResearch/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
reasoning_gym:
  resources_servers:
    reasoning_gym:
      entrypoint: app.py
      domain: knowledge
      description: LangGraph reflection agent compatible with resource servers that do not use tools; provides iterative reflection for diverse agent training data and test time scaling, extensible to use tools or other agent architectures
      value: Iterative test time scaling for improved performance in reasoning tasks
      verified: false

reasoning_gym_langgraph_agent:
  responses_api_agents:
    langgraph_agent:
      entrypoint: reflection_agent.py
      resources_server:
        type: resources_servers
        name: reasoning_gym
      model_server:
        type: responses_api_models
        name: policy_model
      max_reflections: 2
      datasets:
      - name: train
        type: train
        jsonl_fpath: resources_servers/reasoning_gym/data/train_knights_knaves.jsonl
        gitlab_identifier:
          dataset_name: knights_knaves_reasoning_gym
          version: 1.0.0
          artifact_fpath: train_knights_knaves.jsonl
        license: Apache 2.0