asdwdasd

app.py

@@ -1,34 +1,60 @@
-from fastapi import FastAPI, Request
+from fastapi import FastAPI
 from pydantic import BaseModel
-from fastapi.middleware.cors import CORSMiddleware
+from huggingface_hub import hf_hub_download
 from llama_cpp import Llama
 import os
 
-app = FastAPI()
+REPO_ID = "google/gemma-2b-it-GGUF"
+FILENAME = "gemma-2b-it.gguf"
+HF_TOKEN = os.environ.get("HF_TOKEN")  # must be set in HF Spaces Secrets
+MODEL_DIR = "./models"
+MODEL_PATH = os.path.join(MODEL_DIR, FILENAME)
+
+# Step 1: Auto-download model if not exists
+if not os.path.exists(MODEL_PATH):
+    os.makedirs(MODEL_DIR, exist_ok=True)
+    try:
+        print("📦 Downloading model from Hugging Face Hub...")
+        hf_hub_download(
+            repo_id=REPO_ID,
+            filename=FILENAME,
+            token=HF_TOKEN,
+            local_dir=MODEL_DIR,
+            local_dir_use_symlinks=False
+        )
+        print("✅ Model downloaded.")
+    except Exception as e:
+        print(f"❌ Download failed: {e}")
+        raise
 
-# Load model
-MODEL_PATH = "./models/gemma-2b-it.gguf"
-llm = Llama(model_path=MODEL_PATH, n_ctx=512)
-
-# Allow CORS (so frontend or Swagger can work)
-app.add_middleware(
-    CORSMiddleware,
-    allow_origins=["*"],  # change to frontend origin in production
-    allow_credentials=True,
-    allow_methods=["*"],
-    allow_headers=["*"],
+# Step 2: Load model using llama-cpp-python
+print("🤖 Loading GGUF model...")
+llm = Llama(
+    model_path=MODEL_PATH,
+    n_ctx=512,
+    n_threads=4,
+    n_batch=512,
+    verbose=False
 )
 
-# Input model
-class PromptInput(BaseModel):
+# Step 3: FastAPI app
+app = FastAPI()
+
+class PromptRequest(BaseModel):
     prompt: str
 
 @app.post("/prompt")
-async def generate_response(data: PromptInput):
-    output = llm(data.prompt, max_tokens=512, stop=["</s>", "\n\n"], echo=False)
-    return {"response": output["choices"][0]["text"].strip()}
-
-# Healthcheck
-@app.get("/")
-def read_root():
-    return {"message": "AI Builder Backend running"}
+def generate_prompt(req: PromptRequest):
+    prompt = req.prompt.strip()
+
+    output = llm(
+        prompt,
+        max_tokens=512,
+        temperature=0.6,
+        top_p=0.95,
+        stop=["<|endoftext|>", "</s>", "```"],
+        echo=False
+    )
+
+    result = output["choices"][0]["text"].strip()
+    return {"response": result}

model.py

@@ -1,49 +0,0 @@
-import os
-import requests
-from huggingface_hub import hf_hub_download, HfApi
-from llama_cpp import Llama
-
-HF_TOKEN = os.environ.get("HF_TOKEN")
-REPO_ID = "google/gemma-2b-it-GGUF"
-MODEL_FILENAME = "gemma-2b-it.gguf"
-LOCAL_MODEL_PATH = f"/models/{MODEL_FILENAME}"
-CACHE_DIR = "/cache"
-
-os.makedirs(os.path.dirname(LOCAL_MODEL_PATH), exist_ok=True)
-os.makedirs(CACHE_DIR, exist_ok=True)
-
-def download_model():
-    try:
-        print("🔄 Attempting HF Hub download...")
-        model_path = hf_hub_download(
-            repo_id=REPO_ID,
-            filename=MODEL_FILENAME,
-            token=HF_TOKEN,
-            cache_dir=CACHE_DIR,
-        )
-        print("✅ Downloaded via hf_hub_download:", model_path)
-        return model_path
-    except Exception as e:
-        print("⚠️ hf_hub_download failed:", e)
-        print("🔁 Falling back to manual download...")
-
-        headers = {"Authorization": f"Bearer {HF_TOKEN}"}
-        url = f"https://huggingface.co/{REPO_ID}/resolve/main/{MODEL_FILENAME}"
-        response = requests.get(url, headers=headers, stream=True)
-        response.raise_for_status()
-
-        with open(LOCAL_MODEL_PATH, "wb") as f:
-            for chunk in response.iter_content(chunk_size=8192):
-                if chunk:
-                    f.write(chunk)
-        print("✅ Manual download completed:", LOCAL_MODEL_PATH)
-        return LOCAL_MODEL_PATH
-
-
-print("📦 Loading GGUF model...")
-model_path = download_model()
-llm = Llama(model_path=model_path)
-
-def generate_structure(prompt: str) -> str:
-    output = llm(prompt, max_tokens=512)
-    return output["choices"][0]["text"].strip()