Update app.py

app.py

@@ -1,48 +1,97 @@
+# save as app.py
+import threading
 import gradio as gr
-from huggingface_hub import InferenceClient
+import torch
+from transformers import (
+    AutoTokenizer,
+    AutoModelForCausalLM,
+    TextIteratorStreamer,
+)
+
+MODEL_ID = "EpistemeAI/VibeCoder-20B-alpha-0.001"
+
+# --------- Model load (do this once at startup) ----------
+# Adjust dtype / device_map to your environment.
+# If you have limited GPU memory, consider: device_map="auto", load_in_8bit=True (requires bitsandbytes)
+print("Loading tokenizer and model (this may take a while)...")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, use_fast=True, trust_remote_code=True)
+
+# Recommended: try device_map="auto" with accelerate installed; fallback to cpu if not available.
+try:
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_ID,
+        device_map="auto",
+        torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
+        trust_remote_code=True,
+    )
+except Exception as e:
+    print("Automatic device_map load failed, falling back to cpu. Error:", e)
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_ID,
+        device_map={"": "cpu"},
+        torch_dtype=torch.float32,
+        trust_remote_code=True,
+    )
 
+model.eval()
+print("Model loaded. Device:", next(model.parameters()).device)
 
+# --------- Helper: build prompt ----------
+def build_prompt(system_message: str, history: list[dict], user_message: str) -> str:
+    # Keep your conversation structure — adapt to model's preferred format if needed.
+    pieces = []
+    if system_message:
+        pieces.append(f"<|system|>\n{system_message}\n")
+    for turn in history:
+        role = turn.get("role", "user")
+        content = turn.get("content", "")
+        pieces.append(f"<|{role}|>\n{content}\n")
+    pieces.append(f"<|user|>\n{user_message}\n<|assistant|>\n")
+    return "\n".join(pieces)
+
+# --------- Gradio respond function (streams tokens) ----------
 def respond(
     message,
-    history: list[dict[str, str]],
+    history: list[dict],
     system_message,
     max_tokens,
     temperature,
     top_p,
-    hf_token: gr.OAuthToken,
+    hf_token=None,  # kept for compatibility with UI; not used for local pipeline
 ):
     """
-    For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
+    Streams tokens as they are generated using TextIteratorStreamer.
+    Gradio will accept a generator yielding partial response strings.
     """
-    client = InferenceClient(token=hf_token.token, model="EpistemeAI/VibeCoder-20B-alpha-0.001")
-
-    messages = [{"role": "system", "content": system_message}]
+    prompt = build_prompt(system_message, history or [], message)
 
-    messages.extend(history)
+    # Prepare inputs
+    inputs = tokenizer(prompt, return_tensors="pt")
+    input_ids = inputs["input_ids"].to(model.device)
 
-    messages.append({"role": "user", "content": message})
+    # Create streamer to yield token-chunks as they are generated
+    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
 
-    response = ""
+    gen_kwargs = dict(
+        input_ids=input_ids,
+        max_new_tokens=int(max_tokens),
+        do_sample=True,
+        temperature=float(temperature),
+        top_p=float(top_p),
+        streamer=streamer,
+    )
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        choices = message.choices
-        token = ""
-        if len(choices) and choices[0].delta.content:
-            token = choices[0].delta.content
+    # Start generation in background thread
+    thread = threading.Thread(target=model.generate, kwargs=gen_kwargs)
+    thread.start()
 
-        response += token
-        yield response
+    partial = ""
+    # Iterate streamer yields token chunks (strings)
+    for token_str in streamer:
+        partial += token_str
+        yield partial
 
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
+# --------- Build Gradio UI ----------
 chatbot = gr.ChatInterface(
     respond,
     type="messages",
@@ -50,21 +99,15 @@ chatbot = gr.ChatInterface(
         gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
         gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
     ],
 )
 
 with gr.Blocks() as demo:
     with gr.Sidebar():
+        gr.Markdown("Model: " + MODEL_ID)
         gr.LoginButton()
     chatbot.render()
 
-
 if __name__ == "__main__":
     demo.launch()