Code Generation & Refactoring: Architecture Blueprint & Production Stack

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
import httpx
import time

app = FastAPI(title="Private Copilot Completion Gateway")

class FIMCompletionRequest(BaseModel):
    prefix_code: str
    suffix_code: str
    file_language: str
    max_tokens: int = 64

class FIMCompletionResponse(BaseModel):
    completion_text: str
    latency_ms: float
    model_name: str

# Standard FIM Tokens for Qwen2.5-Coder / DeepSeek-Coder
FIM_PREFIX = "<file_sep>filename.py\n<fim_prefix>"
FIM_SUFFIX = "<fim_suffix>"
FIM_MIDDLE = "<fim_middle>"

@app.post("/v1/autocomplete", response_model=FIMCompletionResponse)
async def generate_fim_autocomplete(req: FIMCompletionRequest):
    start_time = time.perf_counter()
    
    # Construct Fill-In-The-Middle Prompt Payload
    fim_prompt = (
        f"{FIM_PREFIX}{req.prefix_code}"
        f"{FIM_SUFFIX}{req.suffix_code}"
        f"{FIM_MIDDLE}"
    )
    
    # Dispatch Request to Private VPC vLLM Cluster
    async with httpx.AsyncClient() as client:
        try:
            res = await client.post(
                "http://vllm-copilot-cluster.internal:8000/v1/completions",
                json={
                    "model": "Qwen2.5-Coder-7B-Instruct",
                    "prompt": fim_prompt,
                    "max_tokens": req.max_tokens,
                    "temperature": 0.1,
                    "stop": ["<fim_middle>", "<file_sep>", "\n\nclass ", "\ndef "]
                },
                timeout=2.0  # Strict 2s SLA timeout
            )
            data = res.json()
            generated_text = data["choices"][0]["text"]
            latency_ms = (time.perf_counter() - start_time) * 1000.0
            
            return FIMCompletionResponse(
                completion_text=generated_text,
                latency_ms=round(latency_ms, 2),
                model_name="Qwen2.5-Coder-7B-Instruct"
            )
        except Exception as err:
            raise HTTPException(status_code=500, detail=f"Inference error: {str(err)}")