Update app.py

app.py

@@ -1,12 +1,17 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
-
-"""
-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
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
 
+# Load the model and tokenizer locally in bfloat16 precision
+model_name = "vietdata/llama32_1b_pub"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype=torch.bfloat16,  # Load model in bfloat16 precision
+    device_map="auto" if torch.cuda.is_available() else None,  # Automatically map to available devices
+)
 
+# Define the respond function
 def respond(
     message,
     history: list[tuple[str, str]],
@@ -15,34 +20,39 @@ def respond(
     temperature,
     top_p,
 ):
-    messages = [{"role": "system", "content": system_message}]
+    from transformers import TextGenerationPipeline
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+    # Build the conversation context
+    prompt = system_message + "\n"
+    for user_msg, bot_msg in history:
+        if user_msg:
+            prompt += f"User: {user_msg}\n"
+        if bot_msg:
+            prompt += f"Bot: {bot_msg}\n"
+    prompt += f"User: {message}\nBot:"
 
-    messages.append({"role": "user", "content": message})
+    # Set up a text generation pipeline
+    pipe = TextGenerationPipeline(
+        model=model, 
+        tokenizer=tokenizer, 
+        device=torch.cuda.current_device() if torch.cuda.is_available() else -1
+    )
 
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
+    # Generate the response
+    response = pipe(
+        prompt,
+        max_length=len(prompt) + max_tokens,
         temperature=temperature,
         top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+        pad_token_id=tokenizer.eos_token_id
+    )[0]["generated_text"]
 
-        response += token
-        yield response
+    # Extract the generated part only
+    generated_response = response[len(prompt):]
+    yield generated_response
 
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
+# Gradio app definition
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
@@ -59,6 +69,5 @@ demo = gr.ChatInterface(
     ],
 )
 
-
 if __name__ == "__main__":
     demo.launch()