app.py

"""Main Gradio app for moderation model testing."""

import os
import sys

import gradio as gr

sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))

from utils.helpers import get_hf_token
from utils.model_interface import extract_model_id, run_test
from ui.sidebar import build_sidebar
from ui.tab_config import build_config_tab
from ui.tab_policy import build_policy_tab
from ui.tab_testing import (
    build_testing_tab,
    format_model_info,
    format_reasoning_info,
    format_test_result,
)


# ============================================================================
# Handlers
# ============================================================================

def handle_run_test(test_input, current_policy, model_choice, reasoning_effort, max_tokens, temperature, top_p, system_prompt_val, response_format_val, oauth_token: gr.OAuthToken | None = None):
    """Handle test execution."""

    if not test_input or not test_input.strip():
        model_info = format_model_info(model_choice, reasoning_effort)
        return model_info, "*Please enter test content*", "*No content*", "*No response yet*", gr.update(value="", visible=False), gr.update(value="", visible=False)

    if not current_policy or current_policy == "*No policy loaded*":
        model_info = format_model_info(model_choice, reasoning_effort)
        return model_info, "*Please load a policy first*", "*No policy*", "*No response yet*", gr.update(value="", visible=False), gr.update(value="", visible=False)

    # OAuth token is automatically injected by Gradio - we don't pass login_button as input
    hf_token, _ = get_hf_token(oauth_token)
    if hf_token is None:
        model_info = format_model_info(model_choice, reasoning_effort)
        return model_info, "*Please log in to use Inference Providers*", "*Authentication required*", "*No response yet*", gr.update(value="", visible=False), gr.update(value="", visible=False)

    model_id = extract_model_id(model_choice)

    result = run_test(
        model_id=model_id,
        test_input=test_input,
        policy=current_policy,
        hf_token=hf_token,
        reasoning_effort=reasoning_effort,
        max_tokens=int(max_tokens),
        temperature=float(temperature),
        top_p=float(top_p),
        system_prompt=system_prompt_val,
        response_format=response_format_val,
    )
    label_text, parsed, cat_text, reasoning, raw_response = format_test_result(result)
    reasoning_visible = bool(reasoning and reasoning.strip())
    model_info = format_model_info(model_choice, reasoning_effort)
    reasoning_info_text, reasoning_info_visible = format_reasoning_info(model_choice, reasoning)
    
    return (
        model_info,
        label_text,
        cat_text,
        raw_response,
        gr.update(value=reasoning_info_text, visible=reasoning_info_visible),
        gr.update(value=reasoning or "", visible=reasoning_visible),
    )


# ============================================================================
# UI Components
# ============================================================================

with gr.Blocks(title="Moderation Model Testing") as demo:
    gr.Markdown("# Moderation Model Testing Interface")
    gr.Markdown(
        "Test moderation models with custom content policies. Define your policy, select a model, "
        "and evaluate how different models classify content according to your rules. "
        "Supports reasoning models that provide detailed explanations for their decisions."
    )

    # Sidebar (collapsible)
    sidebar_components = build_sidebar()
    login_button = sidebar_components["login_button"]

    # Main content area with tabs
    with gr.Tabs():
                # Build tabs
                testing_components = build_testing_tab()
                test_input = testing_components["test_input"]
                run_test_btn = testing_components["run_test_btn"]
                model_info_display = testing_components["model_info_display"]
                label_display = testing_components["label_display"]
                categories_display = testing_components["categories_display"]
                model_response_display = testing_components["model_response_display"]
                reasoning_info = testing_components["reasoning_info"]
                reasoning_display = testing_components["reasoning_display"]

                policy_components = build_policy_tab(os.path.dirname(__file__))
                current_policy_state = policy_components["current_policy_state"]

                config_components = build_config_tab()
                model_dropdown = config_components["model_dropdown"]
                reasoning_effort = config_components["reasoning_effort"]
                max_tokens = config_components["max_tokens"]
                temperature = config_components["temperature"]
                top_p = config_components["top_p"]
                system_prompt_textbox = config_components["system_prompt_textbox"]
                response_format_textbox = config_components["response_format_textbox"]

    # ============================================================================
    # Event Handlers
    # ============================================================================

    # Cross-tab handler: Run test (needs components from all tabs)
    run_test_btn.click(
        handle_run_test,
        inputs=[
            test_input,
            current_policy_state,
            model_dropdown,
            reasoning_effort,
            max_tokens,
            temperature,
            top_p,
            system_prompt_textbox,
            response_format_textbox,
        ],
        outputs=[
            model_info_display,
            label_display,
            categories_display,
            model_response_display,
            reasoning_info,
            reasoning_display,
        ],
    )
    
    model_dropdown.change(
        format_model_info,
        inputs=[model_dropdown, reasoning_effort],
        outputs=model_info_display,
    )
    
    reasoning_effort.change(
        format_model_info,
        inputs=[model_dropdown, reasoning_effort],
        outputs=model_info_display,
    )


if __name__ == "__main__":
    demo.launch(ssr_mode=False)