app.py

from fastapi import FastAPI, Request, HTTPException, Depends
from fastapi.responses import StreamingResponse
from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
from llama_cpp import Llama
import json
import time
import uuid

app = FastAPI()
security = HTTPBearer()

API_KEY = "connectkey"
MODEL_ID = "glm-4.7-flash"

# IQ1_S = 9.25 GB — single file, pasuje na CPU Upgrade (16GB RAM)
print("==> Loading GLM-4.7-Flash IQ1_S (9.25 GB) from HF...")
llm = Llama.from_pretrained(
    repo_id="unsloth/GLM-4.7-Flash-GGUF",
    filename="GLM-4.7-Flash-IQ1_S.gguf",
    n_ctx=8192,
    n_threads=4,
    n_batch=512,
    verbose=False,
)
print("==> Model loaded!")


def verify_key(credentials: HTTPAuthorizationCredentials = Depends(security)):
    if credentials.credentials != API_KEY:
        raise HTTPException(status_code=401, detail="Invalid API key")
    return credentials.credentials


@app.get("/v1/models")
async def list_models(key: str = Depends(verify_key)):
    return {
        "object": "list",
        "data": [{
            "id": MODEL_ID,
            "object": "model",
            "created": int(time.time()),
            "owned_by": "unsloth",
        }]
    }


@app.post("/v1/chat/completions")
async def chat_completions(request: Request, key: str = Depends(verify_key)):
    body = await request.json()

    messages = body.get("messages", [])
    stream = body.get("stream", False)
    max_tokens = body.get("max_tokens", 1024)
    temperature = body.get("temperature", 1.0)
    top_p = body.get("top_p", 0.95)
    min_p = body.get("min_p", 0.01)
    stop = body.get("stop", None)

    completion_id = f"chatcmpl-{uuid.uuid4().hex}"
    created = int(time.time())

    if stream:
        def generate():
            for chunk in llm.create_chat_completion(
                messages=messages,
                max_tokens=max_tokens,
                temperature=temperature,
                top_p=top_p,
                min_p=min_p,
                stop=stop,
                stream=True,
            ):
                delta = chunk["choices"][0].get("delta", {})
                finish_reason = chunk["choices"][0].get("finish_reason")
                data = {
                    "id": completion_id,
                    "object": "chat.completion.chunk",
                    "created": created,
                    "model": MODEL_ID,
                    "choices": [{
                        "index": 0,
                        "delta": delta,
                        "finish_reason": finish_reason,
                    }]
                }
                yield f"data: {json.dumps(data)}\n\n"
            yield "data: [DONE]\n\n"

        return StreamingResponse(generate(), media_type="text/event-stream")

    else:
        result = llm.create_chat_completion(
            messages=messages,
            max_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            min_p=min_p,
            stop=stop,
            stream=False,
        )
        return {
            "id": completion_id,
            "object": "chat.completion",
            "created": created,
            "model": MODEL_ID,
            "choices": result["choices"],
            "usage": result.get("usage", {}),
        }


@app.post("/v1/completions")
async def completions(request: Request, key: str = Depends(verify_key)):
    body = await request.json()

    prompt = body.get("prompt", "")
    stream = body.get("stream", False)
    max_tokens = body.get("max_tokens", 512)
    temperature = body.get("temperature", 1.0)
    top_p = body.get("top_p", 0.95)
    min_p = body.get("min_p", 0.01)
    stop = body.get("stop", None)

    completion_id = f"cmpl-{uuid.uuid4().hex}"
    created = int(time.time())

    if stream:
        def generate():
            for chunk in llm.create_completion(
                prompt=prompt,
                max_tokens=max_tokens,
                temperature=temperature,
                top_p=top_p,
                min_p=min_p,
                stop=stop,
                stream=True,
            ):
                data = {
                    "id": completion_id,
                    "object": "text_completion",
                    "created": created,
                    "model": MODEL_ID,
                    "choices": chunk["choices"],
                }
                yield f"data: {json.dumps(data)}\n\n"
            yield "data: [DONE]\n\n"

        return StreamingResponse(generate(), media_type="text/event-stream")

    else:
        result = llm.create_completion(
            prompt=prompt,
            max_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            min_p=min_p,
            stop=stop,
            stream=False,
        )
        return {
            "id": completion_id,
            "object": "text_completion",
            "created": created,
            "model": MODEL_ID,
            "choices": result["choices"],
            "usage": result.get("usage", {}),
        }


@app.get("/health")
async def health():
    return {"status": "ok", "model": MODEL_ID}