Upload 4 files

Dockerfile

@@ -0,0 +1,15 @@
+FROM python:3.11-slim
+
+WORKDIR /app
+
+COPY requirements.txt .
+RUN pip install --no-cache-dir --prefer-binary -r requirements.txt
+
+COPY app.py .
+
+# Download model at build time so startup is instant
+RUN python -c "from huggingface_hub import hf_hub_download; hf_hub_download('Qwen/Qwen2.5-Coder-0.5B-Instruct-GGUF', 'qwen2.5-coder-0.5b-instruct-q4_k_m.gguf')"
+
+EXPOSE 7860
+
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "7860"]

README.md

@@ -1,11 +1,13 @@
----
-title: Code Collab
-emoji: 🏆
-colorFrom: yellow
-colorTo: yellow
-sdk: docker
-pinned: false
-license: mit
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+---
+title: Code Collab AI Backend
+emoji: 🚀
+colorFrom: purple
+colorTo: indigo
+sdk: docker
+app_port: 7860
+---
+
+# Code Collab AI Backend
+
+Fast code generation API using Qwen 2.5 Coder 0.5B (GGUF).
+Generates HTML, CSS, and JavaScript from natural language prompts.

app.py

@@ -0,0 +1,155 @@
+"""
+Code Collab AI Backend — Fast code generation API
+Uses Qwen 2.5 Coder 0.5B GGUF for low-latency HTML/CSS/JS generation.
+"""
+
+import os
+import re
+import time
+from contextlib import asynccontextmanager
+
+from fastapi import FastAPI, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from pydantic import BaseModel
+from huggingface_hub import hf_hub_download
+from llama_cpp import Llama
+
+# ─── Config ────────────────────────────────────────────────────────────
+MODEL_REPO = "Qwen/Qwen2.5-Coder-0.5B-Instruct-GGUF"
+MODEL_FILE = "qwen2.5-coder-0.5b-instruct-q4_k_m.gguf"
+N_CTX = 1536        # Smaller context = faster (free tier)
+N_THREADS = 2       # HF free tier has 2 vCPU
+MAX_TOKENS = 512    # Keep output short for speed
+TEMPERATURE = 0.5   # Lower = faster + more deterministic
+
+# ─── System prompt (optimized for structured output) ───────────────────
+SYSTEM_PROMPT = """You are a web code generator. Given a user request, output ONLY three fenced code blocks:
+
+```html
+(body content only, no html/head/body tags)
+```
+
+```css
+(complete styles)
+```
+
+```js
+(complete JavaScript)
+```
+
+Rules:
+- No explanations, no markdown text outside code blocks
+- If a section is not needed, output an empty code block for it
+- Write clean, modern code"""
+
+# ─── Global model reference ───────────────────────────────────────────
+llm = None
+
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    """Load model once at startup, keep in memory for fast inference."""
+    global llm
+    print("⬇️  Downloading model...")
+    model_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILE)
+    print(f"✅ Model downloaded: {model_path}")
+
+    print("🔄 Loading model into memory...")
+    llm = Llama(
+        model_path=model_path,
+        n_ctx=N_CTX,
+        n_threads=N_THREADS,
+        n_gpu_layers=0,  # CPU only (free HF Spaces)
+        verbose=False,
+    )
+    print("🚀 Model loaded and ready!")
+    yield
+    llm = None
+
+
+# ─── FastAPI App ───────────────────────────────────────────────────────
+app = FastAPI(title="Code Collab AI", lifespan=lifespan)
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+
+class GenerateRequest(BaseModel):
+    prompt: str
+    max_tokens: int = MAX_TOKENS
+    temperature: float = TEMPERATURE
+
+
+class GenerateResponse(BaseModel):
+    html: str
+    css: str
+    js: str
+    raw: str
+    time_ms: int
+
+
+def parse_code_blocks(text: str) -> dict:
+    """Extract HTML, CSS, JS from fenced code blocks."""
+    result = {"html": "", "css": "", "js": ""}
+
+    for match in re.finditer(r"```(\w+)\s*\n([\s\S]*?)```", text):
+        lang = match.group(1).lower()
+        code = match.group(2).strip()
+
+        if lang in ("html", "xml"):
+            result["html"] = code
+        elif lang == "css":
+            result["css"] = code
+        elif lang in ("js", "javascript"):
+            result["js"] = code
+
+    # Fallback: if no blocks found, treat as HTML
+    if not any(result.values()):
+        result["html"] = text.strip()
+
+    return result
+
+
+# ─── API Endpoints ─────────────────────────────────────────────────────
+
+@app.get("/")
+def health():
+    return {"status": "ok", "model": MODEL_REPO}
+
+
+@app.post("/generate", response_model=GenerateResponse)
+def generate(req: GenerateRequest):
+    if llm is None:
+        raise HTTPException(503, "Model not loaded yet")
+
+    if not req.prompt.strip():
+        raise HTTPException(400, "Prompt cannot be empty")
+
+    start = time.time()
+
+    output = llm.create_chat_completion(
+        messages=[
+            {"role": "system", "content": SYSTEM_PROMPT},
+            {"role": "user", "content": req.prompt},
+        ],
+        max_tokens=req.max_tokens,
+        temperature=req.temperature,
+        stop=["```\n\n", "---"],  # Stop early if model rambles
+    )
+
+    raw_text = output["choices"][0]["message"]["content"]
+    elapsed_ms = int((time.time() - start) * 1000)
+
+    parsed = parse_code_blocks(raw_text)
+
+    return GenerateResponse(
+        html=parsed["html"],
+        css=parsed["css"],
+        js=parsed["js"],
+        raw=raw_text,
+        time_ms=elapsed_ms,
+    )

requirements.txt

@@ -0,0 +1,4 @@
+fastapi==0.115.0
+uvicorn[standard]==0.30.0
+llama-cpp-python==0.2.82
+huggingface-hub==0.27.0