sql_env / docs_draft /architecture-mermaid.md
hjerpe's picture
Upload folder using huggingface_hub
9e64e71 verified
flowchart TD
    subgraph Agent
        A[Your Training Code]
    end

    subgraph SQLEnv_Environment
        B[DOCKER CONTAINER]
        C[FastAPI Server app.py]
        D[SQLEnvironment]
        E[SQLite databases + question sets]
    end

    A -- "WebSocket (persistent session)" --> B

    B --> C
    C --> D
    B -- "Contains" --> E

    subgraph SQLEnvironment_Methods
        I[reset]
        J[step action loop]
        S[state]
    end

    D -- "Calls" --> I
    D -- "Calls" --> J
    D -- "Calls" --> S

    subgraph Episode_Lifecycle_Reset
        I_a[Pick random question]
        I_b[Load corresponding SQLite database read-only]
        I_c[Return initial observation SQLObservation]
    end

    I --> I_a
    I --> I_b
    I --> I_c

    subgraph Episode_Lifecycle_Step
        J_a[DESCRIBE action]
        J_b[SAMPLE action]
        J_c[QUERY action]
        J_d[ANSWER action]
    end

    J --> J_a
    J --> J_b
    J --> J_c
    J --> J_d

    I_c -- "Contains" --> F[SQLObservation typed, IDE-friendly]
    J -- "Returns" --> G[float reward]
    J -- "Returns" --> H[bool done]

    J_d -- "Compares to gold, computes terminal reward" --> G

    classDef entity fill:#f9f,stroke:#333,stroke-width:2px;
    class A,B,C,D,E,F,G,H,I,J,S,I_a,I_b,I_c,J_a,J_b,J_c,J_d entity;