app.py

"""
FastAPI server providing OpenAI-compatible endpoints for code generation.
Designed to work with MCP servers and provide unlimited tokens with minimal rate limiting.
"""
import os
import time
import uuid
from typing import Optional, List, Dict, Any
from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel, Field
from huggingface_hub import hf_hub_download
from llama_cpp import Llama
import uvicorn

# ============================================================================
# CONFIGURATION
# ============================================================================
MODEL_REPO = "TheBloke/deepseek-coder-1.3b-instruct-GGUF"
MODEL_FILE = "deepseek-coder-1.3b-instruct.Q4_K_M.gguf"
MODEL_NAME = "deepseek-coder-1.3b-instruct"

# Context and generation settings for "unlimited" tokens
MAX_CONTEXT = 4096      # Larger context window
MAX_TOKENS = 4096       # Allow very long responses
DEFAULT_TEMP = 0.7
DEFAULT_TOP_P = 0.95

# ============================================================================
# PYDANTIC MODELS (OpenAI-compatible)
# ============================================================================
class Message(BaseModel):
    role: str
    content: str

class ChatCompletionRequest(BaseModel):
    model: str = MODEL_NAME
    messages: List[Message]
    temperature: Optional[float] = DEFAULT_TEMP
    top_p: Optional[float] = DEFAULT_TOP_P
    max_tokens: Optional[int] = MAX_TOKENS
    stream: Optional[bool] = False
    stop: Optional[List[str]] = None

class CompletionRequest(BaseModel):
    model: str = MODEL_NAME
    prompt: str
    temperature: Optional[float] = DEFAULT_TEMP
    top_p: Optional[float] = DEFAULT_TOP_P
    max_tokens: Optional[int] = MAX_TOKENS
    stop: Optional[List[str]] = None

class Usage(BaseModel):
    prompt_tokens: int
    completion_tokens: int
    total_tokens: int

class ChatCompletionChoice(BaseModel):
    index: int
    message: Message
    finish_reason: str

class ChatCompletionResponse(BaseModel):
    id: str
    object: str = "chat.completion"
    created: int
    model: str
    choices: List[ChatCompletionChoice]
    usage: Usage

class CompletionChoice(BaseModel):
    index: int
    text: str
    finish_reason: str

class CompletionResponse(BaseModel):
    id: str
    object: str = "text_completion"
    created: int
    model: str
    choices: List[CompletionChoice]
    usage: Usage

# ============================================================================
# FASTAPI APP
# ============================================================================
app = FastAPI(
    title="Code LLM API",
    description="OpenAI-compatible API for code generation with minimal rate limiting",
    version="1.0.0"
)

# Enable CORS for MCP server access
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# Global model instance
llm: Optional[Llama] = None

# ============================================================================
# MODEL LOADING
# ============================================================================
@app.on_event("startup")
async def load_model():
    """Load the LLM model on startup."""
    global llm
    print(f"Downloading model {MODEL_REPO}/{MODEL_FILE}...")
    model_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILE)
    print(f"Model downloaded to: {model_path}")

    print("Loading model into memory...")
    llm = Llama(                                                                                                       
        model_path=model_path,
        n_ctx=MAX_CONTEXT,                                                                                             
        n_threads=8,           # Changed from 4 to 8
        n_batch=1024,          # Changed from 512 to 1024
        verbose=False,
        n_gpu_layers=0
    )
    print("Model loaded successfully!")

# ============================================================================
# HELPER FUNCTIONS
# ============================================================================
def messages_to_prompt(messages: List[Message]) -> str:
    """Convert OpenAI-style messages to a prompt for CodeLlama."""
    prompt_parts = []

    for msg in messages:
        if msg.role == "system":
            prompt_parts.append(f"### System: {msg.content}")
        elif msg.role == "user":
            prompt_parts.append(f"### Instruction: {msg.content}")
        elif msg.role == "assistant":
            prompt_parts.append(f"### Response: {msg.content}")

    prompt_parts.append("### Response:")
    return "\n".join(prompt_parts)

def estimate_tokens(text: str) -> int:
    """Rough token estimation (1 token ≈ 4 chars)."""
    return len(text) // 4

# ============================================================================
# API ENDPOINTS
# ============================================================================
@app.get("/")
async def root():
    """Health check endpoint."""
    return {
        "status": "online",
        "model": MODEL_NAME,
        "max_context": MAX_CONTEXT,
        "max_tokens": MAX_TOKENS,
        "endpoints": {
            "chat": "/v1/chat/completions",
            "completion": "/v1/completions",
            "models": "/v1/models"
        }
    }

@app.get("/health")
async def health():
    """Health check for monitoring."""
    return {
        "status": "healthy" if llm is not None else "loading",
        "model_loaded": llm is not None
    }

@app.get("/v1/models")
async def list_models():
    """List available models (OpenAI-compatible)."""
    return {
        "object": "list",
        "data": [
            {
                "id": MODEL_NAME,
                "object": "model",
                "created": int(time.time()),
                "owned_by": "huggingface",
                "permission": [],
                "root": MODEL_NAME,
                "parent": None
            }
        ]
    }

@app.post("/v1/chat/completions", response_model=ChatCompletionResponse)
async def chat_completions(request: ChatCompletionRequest):
    """
    OpenAI-compatible chat completions endpoint.
    No rate limiting - designed for unlimited use.
    """
    if llm is None:
        raise HTTPException(status_code=503, detail="Model still loading")

    if request.stream:
        raise HTTPException(status_code=501, detail="Streaming not yet implemented")

    # Convert messages to prompt
    prompt = messages_to_prompt(request.messages)

    # Generate response
    try:
        output = llm(
            prompt,
            max_tokens=request.max_tokens or MAX_TOKENS,
            temperature=request.temperature or DEFAULT_TEMP,
            top_p=request.top_p or DEFAULT_TOP_P,
            stop=request.stop or ["###", "\n\n\n"],
            echo=False
        )

        generated_text = output['choices'][0]['text'].strip()

        # Estimate token usage
        prompt_tokens = estimate_tokens(prompt)
        completion_tokens = estimate_tokens(generated_text)

        return ChatCompletionResponse(
            id=f"chatcmpl-{uuid.uuid4().hex[:8]}",
            created=int(time.time()),
            model=request.model,
            choices=[
                ChatCompletionChoice(
                    index=0,
                    message=Message(role="assistant", content=generated_text),
                    finish_reason="stop"
                )
            ],
            usage=Usage(
                prompt_tokens=prompt_tokens,
                completion_tokens=completion_tokens,
                total_tokens=prompt_tokens + completion_tokens
            )
        )
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Generation failed: {str(e)}")

@app.post("/v1/completions", response_model=CompletionResponse)
async def completions(request: CompletionRequest):
    """
    OpenAI-compatible completions endpoint.
    No rate limiting - designed for unlimited use.
    """
    if llm is None:
        raise HTTPException(status_code=503, detail="Model still loading")

    try:
        output = llm(
            request.prompt,
            max_tokens=request.max_tokens or MAX_TOKENS,
            temperature=request.temperature or DEFAULT_TEMP,
            top_p=request.top_p or DEFAULT_TOP_P,
            stop=request.stop or [],
            echo=False
        )

        generated_text = output['choices'][0]['text'].strip()

        # Estimate token usage
        prompt_tokens = estimate_tokens(request.prompt)
        completion_tokens = estimate_tokens(generated_text)

        return CompletionResponse(
            id=f"cmpl-{uuid.uuid4().hex[:8]}",
            created=int(time.time()),
            model=request.model,
            choices=[
                CompletionChoice(
                    index=0,
                    text=generated_text,
                    finish_reason="stop"
                )
            ],
            usage=Usage(
                prompt_tokens=prompt_tokens,
                completion_tokens=completion_tokens,
                total_tokens=prompt_tokens + completion_tokens
            )
        )
    except Exception as e:
        raise HTTPException(status_code=500, detail=f"Generation failed: {str(e)}")

# ============================================================================
# SIMPLE ENDPOINTS (for easier testing)
# ============================================================================
@app.post("/generate")
async def generate(prompt: str, max_tokens: int = 512):
    """Simple generation endpoint for quick testing."""
    if llm is None:
        raise HTTPException(status_code=503, detail="Model still loading")

    try:
        output = llm(prompt, max_tokens=max_tokens, temperature=0.7)
        return {
            "prompt": prompt,
            "response": output['choices'][0]['text'].strip(),
            "model": MODEL_NAME
        }
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

# ============================================================================
# MAIN
# ============================================================================
if __name__ == "__main__":
    uvicorn.run(
        app,
        host="0.0.0.0",
        port=int(os.getenv("PORT", "7860")),  # HF Spaces uses port 7860
        log_level="info"
    )