app.py

#
# SPDX-FileCopyrightText: Hadad <hadad@linuxmail.org>
# SPDX-License-Identifier: Apache-2.0
#

import uuid
import time
import json
import asyncio
import logging
from typing import Optional, List, Union, Dict, Any, Literal
from fastapi import FastAPI, HTTPException, Request, status
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import StreamingResponse, JSONResponse
from pydantic import BaseModel, Field, ValidationError
from gradio_client import Client
from contextlib import asynccontextmanager

logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s | %(levelname)s | %(name)s | %(threadName)s | %(message)s",
    handlers=[logging.StreamHandler()]
)
logger = logging.getLogger("api_gateway")

class SessionData:
    def __init__(self):
        self.system: str = ""
        self.history: List[Dict[str, Any]] = []
        self.last_access: float = time.time()
        self.active_tasks: Dict[str, asyncio.Task] = {}

class SessionManager:
    def __init__(self):
        self.sessions: Dict[str, Dict[str, SessionData]] = {}
        self.lock = asyncio.Lock()

    async def cleanup(self):
        while True:
            await asyncio.sleep(60)
            async with self.lock:
                now = time.time()
                expired = []
                for user, sessions in list(self.sessions.items()):
                    for sid, data in list(sessions.items()):
                        if now - data.last_access > 300:
                            expired.append((user, sid))
                            for task_id, task in data.active_tasks.items():
                                if not task.done():
                                    task.cancel()
                                    logger.info(f"Cancelled active task {task_id} for expired session {sid}")
                for user, sid in expired:
                    if user in self.sessions and sid in self.sessions[user]:
                        del self.sessions[user][sid]
                        if not self.sessions[user]:
                            del self.sessions[user]
                        logger.info(f"Session expired: user={user} session={sid}")

    async def get_session(self, user: Optional[str], session_id: Optional[str]) -> (str, str, SessionData):
        async with self.lock:
            if not user:
                user = str(uuid.uuid4())
                logger.debug(f"Generated new user ID: {user}")
            if user not in self.sessions:
                self.sessions[user] = {}
            if not session_id or session_id not in self.sessions[user]:
                session_id = str(uuid.uuid4())
                self.sessions[user][session_id] = SessionData()
                logger.info(f"Created new session: user={user} session={session_id}")
            session = self.sessions[user][session_id]
            session.last_access = time.time()
            logger.debug(f"Session accessed: user={user} session={session_id} history_length={len(session.history)}")
            return user, session_id, session

session_manager = SessionManager()

async def refresh_client(app: FastAPI):
    while True:
        await asyncio.sleep(1)
        async with app.state.client_lock:
            if app.state.client is None:
                await asyncio.sleep(1)
                continue
        while True:
            await asyncio.sleep(15)
            async with app.state.client_lock:
                if app.state.client is not None:
                    try:
                        old_client = app.state.client
                        app.state.client = None
                        del old_client
                        app.state.client = Client("https://hadadrjt-ai.hf.space/")
                        logger.info("Refreshed Gradio client connection")
                    except Exception as e:
                        logger.error(f"Error refreshing Gradio client: {e}", exc_info=True)
                        app.state.client = None
                        await asyncio.sleep(5)
                else:
                    break

@asynccontextmanager
async def lifespan(app: FastAPI):
    app.state.session_manager = session_manager
    app.state.client = None
    app.state.client_lock = asyncio.Lock()
    app.state.refresh_task = asyncio.create_task(refresh_client(app))
    logger.info("App lifespan started, refresh client task running")
    try:
        yield
    finally:
        app.state.refresh_task.cancel()
        await asyncio.sleep(0.1)
        logger.info("App lifespan ended, refresh client task cancelled")

app = FastAPI(
    title="J.A.R.V.I.S. OpenAI-Compatible API",
    version="2.1.3-0625",
    lifespan=lifespan,
)

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["GET", "POST", "OPTIONS", "HEAD"],
    allow_headers=["*"],
)

class Function(BaseModel):
    name: str
    description: Optional[str]
    parameters: Dict[str, Any]

class Tool(BaseModel):
    type: Literal["function"] = "function"
    function: Function

class ToolCall(BaseModel):
    id: str
    type: Literal["function"]
    function: Dict[str, str]

class Message(BaseModel):
    role: str = Field(..., pattern="^(system|user|assistant|tool|function)$")
    content: Optional[Union[str, List[Dict[str, Any]]]]
    name: Optional[str] = None
    tool_calls: Optional[List[ToolCall]] = None
    tool_call_id: Optional[str] = None

class CommonParams(BaseModel):
    model: str
    stream: bool = False
    user: Optional[str] = None
    session_id: Optional[str] = None
    top_p: Optional[float] = None
    top_k: Optional[int] = None
    temperature: Optional[float] = None
    max_tokens: Optional[int] = None
    max_new_tokens: Optional[int] = None
    presence_penalty: Optional[float] = None
    frequency_penalty: Optional[float] = None
    repetition_penalty: Optional[float] = None
    logit_bias: Optional[Dict[str, float]] = None
    repeat_penalty: Optional[float] = None
    seed: Optional[int] = None
    tools: Optional[List[Tool]] = None
    tool_choice: Optional[Union[str, Dict[str, str]]] = None
    functions: Optional[List[Function]] = None
    function_call: Optional[Union[str, Dict[str, str]]] = None

class ChatCompletionRequest(CommonParams):
    messages: List[Message]

class CompletionRequest(CommonParams):
    prompt: Union[str, List[str]]

class EmbeddingRequest(BaseModel):
    model: str
    input: Union[str, List[str]]
    user: Optional[str] = None

class RouterRequest(CommonParams):
    endpoint: Optional[str] = "chat/completions"
    messages: Optional[List[Message]] = None
    prompt: Optional[Union[str, List[str]]] = None
    input: Optional[Union[str, List[str]]] = None

def sanitize_messages(messages: List[Message]) -> List[Message]:
    cleaned = []
    for m in messages:
        if isinstance(m.content, list):
            texts = [c.get("text", "") for c in m.content if isinstance(c, dict) and c.get("type") == "text"]
            if texts:
                cleaned.append(Message(role=m.role, content=" ".join(texts)))
        elif isinstance(m.content, str):
            cleaned.append(m)
    return cleaned

def map_messages(system: str, history: List[Dict[str, Any]], new_msgs: List[Message]) -> str:
    msgs = []
    if system:
        msgs.append({"role": "system", "content": system})
    msgs.extend(history)
    msgs.extend([m.model_dump() for m in new_msgs if m.role != "system"])
    text = ""
    for m in msgs:
        text += f"{m.get('role','')}:{m.get('content','')}\n"
    return text.strip()

async def get_client(app: FastAPI) -> Client:
    async with app.state.client_lock:
        if app.state.client is None:
            try:
                app.state.client = Client("https://hadadrjt-ai.hf.space/")
                logger.info("Created Gradio client connection on demand")
            except Exception as e:
                logger.error(f"Failed to create Gradio client: {e}", exc_info=True)
                raise HTTPException(status_code=502, detail="Failed to connect to upstream Gradio app")
        return app.state.client

async def call_gradio(client: Client, params: dict):
    for attempt in range(3):
        try:
            logger.debug(f"Calling Gradio attempt {attempt+1}")
            return await asyncio.to_thread(lambda: client.submit(**params))
        except Exception as e:
            logger.warning(f"Gradio call failed attempt {attempt+1}: {e}", exc_info=True)
            await asyncio.sleep(0.2 * (attempt + 1))
    logger.error("Gradio upstream error after 3 attempts")
    raise HTTPException(status_code=502, detail="Upstream Gradio app error")

async def stream_response(job, session_id: str, session_history: List[Dict[str, Any]], new_messages: List[Message], response_type: str):
    partial = ""
    try:
        chunks = await asyncio.to_thread(lambda: list(job))
    except Exception as e:
        logger.error(f"Streaming error: {e}", exc_info=True)
        chunks = []
    for chunk in chunks:
        try:
            if isinstance(chunk, list):
                response = next((item.get('content') for item in chunk if isinstance(item, dict) and 'content' in item), str(chunk))
            else:
                response = str(chunk)
            token = response[len(partial):] if response.startswith(partial) else response
            partial = response
            if response_type == "chat":
                data = {
                    "id": str(uuid.uuid4()),
                    "object": "chat.completion.chunk",
                    "choices": [{"delta": {"content": token}, "index": 0, "finish_reason": None}],
                    "session_id": session_id
                }
            else:
                data = {
                    "id": str(uuid.uuid4()),
                    "object": "text_completion.chunk",
                    "choices": [{"text": token, "index": 0, "finish_reason": None}],
                    "session_id": session_id
                }
            yield f"data: {json.dumps(data)}\n\n"
        except Exception as e:
            logger.error(f"Chunk yield error: {e}", exc_info=True)
            continue
    session_history.extend([m.model_dump() for m in new_messages if m.role != "system"])
    session_history.append({"role": "assistant", "content": partial})
    done_data = {
        "id": str(uuid.uuid4()),
        "object": f"{response_type}.completion.chunk",
        "choices": [{"delta" if response_type=="chat" else "text": {} if response_type=="chat" else "", "index": 0, "finish_reason": "stop"}],
        "session_id": session_id
    }
    yield f"data: {json.dumps(done_data)}\n\n"

@app.post("/v1/chat/completions")
async def chat_completions(req: ChatCompletionRequest):
    user, session_id, session = await session_manager.get_session(req.user, req.session_id)
    req.messages = sanitize_messages(req.messages)
    for m in req.messages:
        if m.role == "system":
            session.system = m.content
            break
    text = map_messages(session.system, session.history, req.messages)
    params = {
        "message": text,
        "model_label": req.model,
        "api_name": "/api",
        "top_p": req.top_p,
        "top_k": req.top_k,
        "temperature": req.temperature,
        "max_tokens": req.max_tokens,
        "max_new_tokens": req.max_new_tokens,
        "presence_penalty": req.presence_penalty,
        "frequency_penalty": req.frequency_penalty,
        "repetition_penalty": req.repetition_penalty,
        "repeat_penalty": req.repeat_penalty,
        "logit_bias": req.logit_bias,
        "seed": req.seed,
        "functions": req.functions or req.tools,
        "function_call": req.function_call or req.tool_choice,
    }
    params = {k: v for k, v in params.items() if v is not None}
    logger.info(f"Chat completion request user={user} session={session_id} model={req.model} stream={req.stream}")
    client = await get_client(app)
    if req.stream:
        job = await call_gradio(client, params)
        generator = stream_response(job, session_id, session.history, req.messages, "chat")
        return StreamingResponse(generator, media_type="text/event-stream")
    else:
        loop = asyncio.get_running_loop()
        try:
            result = await loop.run_in_executor(None, lambda: client.predict(**params))
        except Exception as e:
            logger.error(f"Gradio predict error: {e}", exc_info=True)
            raise HTTPException(status_code=502, detail="Upstream Gradio app error")
        session.history.extend([m.model_dump() for m in req.messages if m.role != "system"])
        session.history.append({"role": "assistant", "content": result})
        logger.info(f"Chat completion response sent user={user} session={session_id}")
        return {
            "id": str(uuid.uuid4()),
            "object": "chat.completion",
            "choices": [{"message": {"role": "assistant", "content": result}}],
            "session_id": session_id
        }

@app.post("/v1/completions")
async def completions(req: CompletionRequest):
    user, session_id, _ = await session_manager.get_session(req.user, req.session_id)
    prompt = req.prompt if isinstance(req.prompt, str) else "\n".join(req.prompt)
    params = {
        "message": prompt,
        "model_label": req.model,
        "api_name": "/api",
        "top_p": req.top_p,
        "top_k": req.top_k,
        "temperature": req.temperature,
        "max_tokens": req.max_tokens,
        "max_new_tokens": req.max_new_tokens,
        "presence_penalty": req.presence_penalty,
        "frequency_penalty": req.frequency_penalty,
        "repetition_penalty": req.repetition_penalty,
        "repeat_penalty": req.repeat_penalty,
        "logit_bias": req.logit_bias,
        "seed": req.seed,
    }
    params = {k: v for k, v in params.items() if v is not None}
    logger.info(f"Completion request user={user} session={session_id} model={req.model} stream={req.stream}")
    client = await get_client(app)
    if req.stream:
        job = await call_gradio(client, params)
        generator = stream_response(job, session_id, [], [], "text")
        return StreamingResponse(generator, media_type="text/event-stream")
    else:
        loop = asyncio.get_running_loop()
        try:
            result = await loop.run_in_executor(None, lambda: client.predict(**params))
        except Exception as e:
            logger.error(f"Gradio predict error: {e}", exc_info=True)
            raise HTTPException(status_code=502, detail="Upstream Gradio app error")
        logger.info(f"Completion response sent user={user} session={session_id}")
        return {"id": str(uuid.uuid4()), "object": "text_completion", "choices": [{"text": result}]}

@app.post("/v1/embeddings")
async def embeddings(req: EmbeddingRequest):
    inputs = req.input if isinstance(req.input, list) else [req.input]
    embeddings = [[0.0] * 768 for _ in inputs]
    logger.info(f"Embedding request model={req.model} inputs_count={len(inputs)}")
    return {"object": "list", "data": [{"embedding": emb, "index": i} for i, emb in enumerate(embeddings)]}

@app.get("/v1/models")
async def get_models():
    logger.info("Models list requested")
    return {"object": "list", "data": [{"id": "Q8_K_XL", "object": "model", "owned_by": "J.A.R.V.I.S."}]}

@app.get("/v1/history")
async def get_history(user: Optional[str] = None, session_id: Optional[str] = None):
    user = user or "anonymous"
    sessions = session_manager.sessions
    logger.info(f"History requested user={user} session={session_id}")
    if user in sessions and session_id and session_id in sessions[user]:
        return {"user": user, "session_id": session_id, "history": sessions[user][session_id].history}
    return {"user": user, "session_id": session_id, "history": []}

@app.post("/v1/responses/cancel")
async def cancel_response(user: Optional[str], session_id: Optional[str], task_id: Optional[str]):
    user = user or "anonymous"
    if not task_id:
        logger.warning(f"Cancel response missing task_id user={user} session={session_id}")
        raise HTTPException(status_code=400, detail="Missing task_id for cancellation")
    async with session_manager.lock:
        if user in session_manager.sessions and session_id in session_manager.sessions[user]:
            session = session_manager.sessions[user][session_id]
            task = session.active_tasks.get(task_id)
            if task and not task.done():
                task.cancel()
                logger.info(f"Cancelled task {task_id} for user={user} session={session_id}")
                return {"message": f"Cancelled task {task_id}"}
    logger.warning(f"Task not found or already completed task_id={task_id} user={user} session={session_id}")
    raise HTTPException(status_code=404, detail="Task not found or already completed")

@app.api_route("/v1", methods=["POST", "GET", "OPTIONS", "HEAD"])
async def router(request: Request):
    if request.method == "POST":
        try:
            body_json = await request.json()
        except Exception:
            logger.error("Invalid JSON body in router POST")
            raise HTTPException(status_code=400, detail="Invalid JSON body")
        try:
            body = RouterRequest(**body_json)
        except ValidationError as e:
            logger.error(f"Validation error in router POST: {e.errors()}")
            raise HTTPException(status_code=422, detail=e.errors())
        endpoint = body.endpoint or "chat/completions"
        logger.info(f"Router POST to endpoint={endpoint}")
        if endpoint == "chat/completions":
            if not body.model or not body.messages:
                raise HTTPException(status_code=422, detail="Missing 'model' or 'messages'")
            req_obj = ChatCompletionRequest(**body.dict())
            return await chat_completions(req_obj)
        elif endpoint == "completions":
            if not body.model or not body.prompt:
                raise HTTPException(status_code=422, detail="Missing 'model' or 'prompt'")
            req_obj = CompletionRequest(**body.dict())
            return await completions(req_obj)
        elif endpoint == "embeddings":
            if not body.model or body.input is None:
                raise HTTPException(status_code=422, detail="Missing 'model' or 'input'")
            req_obj = EmbeddingRequest(**body.dict())
            return await embeddings(req_obj)
        elif endpoint == "models":
            return await get_models()
        elif endpoint == "history":
            return await get_history(body.user, body.session_id)
        elif endpoint == "responses/cancel":
            return await cancel_response(body.user, body.session_id, body.session_id)
        else:
            logger.warning(f"Router POST unknown endpoint: {endpoint}")
            raise HTTPException(status_code=404, detail="Endpoint not found")
    else:
        logger.info(f"Router {request.method} called - only POST supported with JSON body")
        return JSONResponse({"message": "Send POST request with JSON body"}, status_code=status.HTTP_405_METHOD_NOT_ALLOWED)

@app.get("/")
async def root():
    return {
        "endpoints": [
            "/v1/chat/completions",
            "/v1/completions",
            "/v1/embeddings",
            "/v1/models",
            "/v1/history",
            "/v1/responses/cancel",
            "/v1"
        ]
    }

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)