app.py

import os
import sys
from huggingface_hub import HfApi
import requests
import pathlib
import subprocess
import shutil
import io
import yaml
import gradio as gr
import gradiologin as gl
from huggingface_hub import whoami
from loguru import logger
from yourbench.pipeline import run_pipeline

UPLOAD_DIRECTORY = pathlib.Path("/app/uploaded_files")
UPLOAD_DIRECTORY.mkdir(parents=True, exist_ok=True)
CONFIG_PATH = pathlib.Path("/app/yourbench_config.yml")

logger.remove()
logger.add(sys.stderr, level="INFO")

class SubprocessManager:
    def __init__(self, command):
        self.command = command
        self.process = None
        self.output_stream = io.StringIO()

    def start_process(self):
        """Start the subprocess."""
        if self.is_running():
            logger.info("Process is already running")
            return

        self.process = subprocess.Popen(
            self.command,
            stdout=subprocess.PIPE,
            stderr=subprocess.STDOUT,  # Combine stderr with stdout
            text=True,
            bufsize=1,  # Line-buffered
            start_new_session=True  # Start the process in a new session
        )
        os.set_blocking(self.process.stdout.fileno(), False)
        logger.info("Started the process")

    def read_and_get_output(self):
        """Read available subprocess output and return the captured output."""
        if self.process and self.process.stdout:
            try:
                while True:
                    line = self.process.stdout.readline()
                    if line:
                        self.output_stream.write(line)  # Capture in StringIO
                    else:
                        break
            except BlockingIOError:
                pass
        return self.output_stream.getvalue()

    def stop_process(self):
        """Terminate the subprocess."""
        if not self.is_running():
            logger.info("Started the process")
            return
        logger.info("Sending SIGTERM to the Process")
        self.process.terminate()
        exit_code = self.process.wait() # Wait for process to terminate
        logger.info(f"Process stopped exit code {exit_code}")
        #return exit_code

    def kill_process(self):
        """Forcefully kill the subprocess."""
        if not self.is_running():
            logger.info("Process is not running")
            return
        logger.info("Sending SIGKILL to the Process")
        self.process.kill()
        exit_code = self.process.wait() # Wait for process to be killed
        logger.info(f"Process killed exit code {exit_code}")
        #return exit_code

    def is_running(self):
        """Check if the subprocess is still running."""
        return self.process and self.process.poll() is None


command = ["uv", "run", "yourbench", f"--config={CONFIG_PATH}"]
manager = SubprocessManager(command)


def generate_config(hf_token: gr.OAuthToken | None, hf_org, model_name, provider, base_url, api_key, max_concurrent_requests):
    config = {
        "hf_configuration": {
            "token": hf_token,
            "private": True,
            "hf_organization": hf_org
        },
        "model_list": [{
            "model_name": model_name,
            "provider": provider,
            "base_url": base_url,
            "api_key": api_key,
            "max_concurrent_requests": max_concurrent_requests
        }],
        "model_roles": {role: [model_name] for role in [
            "ingestion", "summarization", "single_shot_question_generation",
            "multi_hop_question_generation", "answer_generation", "judge_answers"
        ]},
        "inference_config": {"max_concurrent_requests": 16},
        "pipeline": {
            "ingestion": {
                "source_documents_dir": "/app/uploaded_files",
                "output_dir": "/app/ingested",
                "run": True
            },
            "upload_ingest_to_hub": {
                "source_documents_dir": "/app/ingested",
                "hub_dataset_name": "test_ingested_documents",
                "local_dataset_path": "/app/ingested_dataset",
                "run": True
            },
            "summarization": {
                "source_dataset_name": "test_ingested_documents",
                "output_dataset_name": "test_summaries",
                "local_dataset_path": "/results/test_summaries",
                "concat_existing_dataset": False,
                "run": True
            },
            "chunking": {
                "source_dataset_name": "test_summaries",
                "output_dataset_name": "test_chunked_documents",
                "local_dataset_path": "/results/test_chunked_documents",
                "concat_existing_dataset": False,
                "chunking_configuration": {
                    "l_min_tokens": 64,
                    "l_max_tokens": 128,
                    "tau_threshold": 0.3,
                    "h_min": 2,
                    "h_max": 4
                },
                "run": True
            },
            "single_shot_question_generation": {
                "source_dataset_name": "test_chunked_documents",
                "output_dataset_name": "test_single_shot_questions",
                "local_dataset_path": "/results/test_single_shot_questions",
                "diversification_seed": "24 year old adult",
                "concat_existing_dataset": False,
                "run": True
            },
            "multi_hop_question_generation": {
                "source_dataset_name": "test_chunked_documents",
                "output_dataset_name": "test_multi_hop_questions",
                "local_dataset_path": "/results/test_multi_hop_questions",
                "concat_existing_dataset": False,
                "run": True
            },
            "answer_generation": {
                "run": True,
                "question_dataset_name": "test_single_shot_questions",
                "output_dataset_name": "test_answered_questions",
                "local_dataset_path": "/results/test_answered_questions",
                "concat_existing_dataset": False,
                "strategies": [{
                    "name": "zeroshot",
                    "prompt": "ZEROSHOT_QA_USER_PROMPT",
                    "model_name": model_name
                }, {
                    "name": "gold",
                    "prompt": "GOLD_QA_USER_PROMPT",
                    "model_name": model_name
                }]
            },
            "judge_answers": {
                "run": True,
                "source_judge_dataset_name": "test_answered_questions",
                "output_judged_dataset_name": "test_judged_comparisons",
                "local_dataset_path": "/results/test_judged_comparisons",
                "concat_existing_dataset": False,
                "comparing_strategies": [["zeroshot", "gold"]],
                "chunk_column_index": 0,
                "random_seed": 42
            }
        }
    }
    return yaml.dump(config, default_flow_style=False)

def save_config(yaml_text):
    with open(CONFIG_PATH, "w") as file:
        file.write(yaml_text)
    return "✅ Config saved!"

def save_files(files: list[str]):
    saved_paths = [shutil.move(str(pathlib.Path(file)), str(UPLOAD_DIRECTORY / pathlib.Path(file).name)) for file in files]
    return f"Files saved to: {', '.join(saved_paths)}"

def populate_user_info(oauth_profile: gr.OAuthProfile = None, oauth_token: gr.OAuthToken = None):
    if oauth_profile is None or oauth_token is None:
        return (
            gr.Dropdown.update(choices=["(Please log in to load tokens)"], value=None),
            gr.Dropdown.update(choices=["(Please log in)"], value=None),
            "🔒 Not logged in"
        )

    username = oauth_profile.username
    org_names = []
    token_names = []

    try:
        headers = {"Authorization": f"Bearer {oauth_token.token}"}
        hf_api = HfApi(token=oauth_token.token)

        # Fetch all user tokens
        token_data = hf_api.list_tokens()
        for t in token_data:
            name = t.get("name") or f"{t['token'][:4]}...{t['token'][-4:]}"
            token_names.append(name)

        # Fetch user organizations
        orgs = hf_api.get_user_organizations()
        org_names = [org.organization for org in orgs]

    except Exception as e:
        print("Error fetching user/org info:", e)
        token_names = [f"{oauth_token.token[:4]}...{oauth_token.token[-4:]}"]

    org_options = [username] + org_names
    default_org = username

    return (
        gr.Dropdown.update(choices=token_names, value=token_names[0] if token_names else None),
        gr.Dropdown.update(choices=org_options, value=default_org),
        f"✅ Logged in as {username}"
    )

def hello(profile: gr.OAuthProfile | None) -> str:
    if profile is None:
        return "I don't know you."
    return f"Hello {profile.name}"

def update_hf_org_dropdown(oauth_token: gr.OAuthToken | None) -> str:
    if oauth_token is None:
        print("Please deploy this on Spaces and log in to list organizations.")
        return []
    org_names = [org["name"] for org in whoami(oauth_token.token)["orgs"]]
    return gr.Dropdown(org_names, label="Organization")

with gr.Blocks() as app:
    gr.Markdown("## YourBench Configuration")
    with gr.Row():
        login_btn = gr.LoginButton()
        hello_text = gr.Markdown()
        app.load(hello, inputs=None, outputs=hello_text)

    with gr.Tab("Configuration"):
        model_name = gr.Textbox(label="Model Name")

        hf_org_dropdown = gr.Dropdown(list(), label="Organization")
        app.load(update_hf_org_dropdown, inputs=None, outputs=hf_org_dropdown)

        provider = gr.Dropdown(["openrouter", "openai", "huggingface"], value="huggingface", label="Provider")
        base_url = gr.Textbox(label="Base URL")
        api_key = gr.Textbox(label="API Key")
        max_concurrent_requests = gr.Dropdown([8, 16, 32], value=16, label="Max Concurrent Requests")
        config_output = gr.Code(label="Generated Config", language="yaml")
        preview_button = gr.Button("Generate Config")
        save_button = gr.Button("Save Config")

        preview_button.click(
            generate_config,
            inputs=[hf_org_dropdown, model_name, provider, base_url, api_key, max_concurrent_requests],
            outputs=config_output
        )
        save_button.click(save_config, inputs=[config_output], outputs=[gr.Textbox(label="Save Status")])

    with gr.Tab("Files"):
        file_input = gr.File(label="Upload text files", file_count="multiple", file_types=[".txt", ".md", ".html"])
        output = gr.Textbox(label="Log")
        file_input.upload(save_files, file_input, output)

    with gr.Tab("Run Generation"):
        log_output = gr.Code(label="Log Output", language=None, lines=20, interactive=False)
        start_button = gr.Button("Start Task")
        start_button.click(manager.start_process)
        timer = gr.Timer(0.1, active=True)
        timer.tick(manager.read_and_get_output, outputs=log_output)

        kill_button = gr.Button("Kill Task")
        kill_button.click(manager.kill_process)

app.launch()