Create server.py

server.py

@@ -0,0 +1,25 @@
+from fastapi import FastAPI
+import subprocess
+
+app = FastAPI()
+
+MODEL_PATH = "deepseek-coder-v2-lite.Q4_K_M.gguf"
+
+@app.get("/")
+def read_root():
+    return {"status": "DeepSeek server running"}
+
+@app.post("/generate")
+def generate(prompt: str):
+
+    cmd = [
+        "./llama.cpp/main",
+        "-m", MODEL_PATH,
+        "-p", prompt,
+        "-n", "300",
+        "--ctx-size", "8192"
+    ]
+
+    result = subprocess.run(cmd, capture_output=True, text=True)
+
+    return {"response": result.stdout}