Create app.py

app.py

@@ -7,49 +7,30 @@ import config
 
 
 # ============================
-# Environment & Token Setup
+# Download Model
 # ============================
 
 HF_TOKEN = os.environ.get("HF_TOKEN")
 
-if HF_TOKEN is None:
-    print("Warning: HF_TOKEN not found. Download may fail for gated repos.")
-
-
-# ============================
-# Model Download (cached automatically by HF)
-# ============================
-
 print("Downloading model from Hugging Face Hub...")
 
-try:
-    model_path = hf_hub_download(
-        repo_id=config.MODEL_REPO,
-        filename=config.MODEL_FILE,
-        token=HF_TOKEN,
-        cache_dir="/tmp/hf_cache"
-    )
-
-    print(f"Model downloaded successfully: {model_path}")
+model_path = hf_hub_download(
+    repo_id=config.MODEL_REPO,
+    filename=config.MODEL_FILE,
+    token=HF_TOKEN,
+    cache_dir="/tmp/hf_cache"
+)
 
-except Exception as e:
-    print("Model download failed:", str(e))
-    raise e
+print("Model downloaded successfully:", model_path)
 
 
 # ============================
-# CPU Optimization
+# Load Model
 # ============================
 
 CPU_THREADS = multiprocessing.cpu_count()
 
-print(f"CPU Threads available: {CPU_THREADS}")
-
-
-# ============================
-# Load llama.cpp model
-# ============================
-
+print("CPU Threads available:", CPU_THREADS)
 print("Loading model into memory...")
 
 llm = Llama(
@@ -66,118 +47,82 @@ print("Model loaded successfully.")
 
 
 # ============================
-# Prompt System
+# Prompt Formatting
 # ============================
 
 SYSTEM_PROMPT = """You are DeepSeek Coder, an expert programming assistant.
-
-Rules:
-- Write clean, correct, production-ready code
-- Be concise
-- Only explain if asked
-- Prefer efficient solutions
+You write clean, efficient, production-ready code.
+Only explain if user asks.
 """
 
 
-def format_prompt(message, history):
+def build_prompt(messages):
 
     prompt = SYSTEM_PROMPT + "\n\n"
 
-    for user, assistant in history:
-        prompt += f"User: {user}\nAssistant: {assistant}\n"
+    for msg in messages:
+        if msg["role"] == "user":
+            prompt += f"User: {msg['content']}\n"
+        elif msg["role"] == "assistant":
+            prompt += f"Assistant: {msg['content']}\n"
 
-    prompt += f"User: {message}\nAssistant:"
+    prompt += "Assistant:"
 
     return prompt
 
 
 # ============================
-# Streaming Generation
-# ============================
-
-def generate_stream(message, history):
-
-    prompt = format_prompt(message, history)
-
-    output = ""
-
-    try:
-        for token in llm(
-            prompt,
-            max_tokens=config.MAX_TOKENS,
-            temperature=config.TEMPERATURE,
-            top_p=0.95,
-            stream=True
-        ):
-            text = token["choices"][0]["text"]
-            output += text
-            yield output
-
-    except Exception as e:
-        yield f"Error during generation: {str(e)}"
-
-
-# ============================
-# Gradio UI Logic
+# Streaming Generator
 # ============================
 
-def user(user_message, history):
-    return "", history + [[user_message, ""]]
+def generate_response(message, history):
 
+    messages = history + [{"role": "user", "content": message}]
 
-def bot(history):
+    prompt = build_prompt(messages)
 
-    user_message = history[-1][0]
+    output = ""
 
-    for response in generate_stream(user_message, history[:-1]):
-        history[-1][1] = response
-        yield history
+    for token in llm(
+        prompt,
+        max_tokens=config.MAX_TOKENS,
+        temperature=config.TEMPERATURE,
+        top_p=0.95,
+        stream=True
+    ):
+        text = token["choices"][0]["text"]
+        output += text
+        yield output
 
 
 # ============================
-# Gradio Interface
+# Gradio Chat Interface
 # ============================
 
-with gr.Blocks(theme=gr.themes.Soft()) as demo:
-
-    gr.Markdown("# DeepSeek Coder 1.3B (GGUF Production)")
-    gr.Markdown("Fast, efficient coding assistant running on llama.cpp")
-
-    chatbot = gr.Chatbot(height=500)
-
-    msg = gr.Textbox(
-        placeholder="Ask a coding question...",
-        container=False
-    )
+def chat(message, history):
 
-    clear = gr.Button("Clear Chat")
+    history = history or []
 
+    assistant_response = ""
 
-    msg.submit(
-        user,
-        [msg, chatbot],
-        [msg, chatbot],
-        queue=True
-    ).then(
-        bot,
-        chatbot,
-        chatbot
-    )
-
-
-    clear.click(
-        lambda: [],
-        None,
-        chatbot,
-        queue=False
-    )
+    for partial in generate_response(message, history):
+        assistant_response = partial
+        yield history + [
+            {"role": "user", "content": message},
+            {"role": "assistant", "content": assistant_response},
+        ]
 
 
 # ============================
-# Launch Server
+# Launch UI
 # ============================
 
-demo.queue()
+demo = gr.ChatInterface(
+    fn=chat,
+    title="DeepSeek Coder 1.3B",
+    description="Production GGUF model running on llama.cpp",
+    type="messages"
+)
 
 demo.launch(
     server_name="0.0.0.0",