Update app.py

app.py

@@ -1,64 +1,76 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+from transformers import pipeline, AutoTokenizer, BitsAndBytesConfig, AutoModelForCausalLM
+import torch
+import spaces
 
-"""
-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")
+MODEL_PATH = "benhaotang/phi4-qwq-sky-t1"
+MODEL_URL = f"https://huggingface.co/{MODEL_PATH}"
 
+def load_model():
+    bnb_config = BitsAndBytesConfig(
+        load_in_8bit=False,
+        llm_int8_enable_fp32_cpu_offload=True
+    )
+    
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_PATH,
+        device_map="auto",
+        torch_dtype=torch.float16,
+        offload_folder="offload_folder",
+        quantization_config=bnb_config
+    )
+    
+    tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
+    
+    # Create pipeline
+    pipe = pipeline(
+        "text-generation",
+        model=model,
+        tokenizer=tokenizer,
+        device_map="auto",
+    )
+    
+    return pipe
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+pipe = load_model()
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+@spaces.GPU(duration=110)
+def generate_response(prompt, max_length=1024):
+    # Convert prompt into messages format
+    messages = [
+        {"role": "system", "content": "You are a helpful AI asistent. You always think step by step."},
+        {"role": "user", "content": prompt}
+    ]
+    
+    # Generate response using pipeline
+    outputs = pipe(messages, max_new_tokens=max_length)
+    
+    # Extract the generated text
+    response = outputs[0]["generated_text"]
+    
+    # Since pipeline returns the full conversation, we want to extract just the response
+    # Split by the prompt and take the last part
+    response_only = response.split(prompt)[-1].strip()
+    
+    return response_only
 
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
-
-        response += token
-        yield response
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-demo = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        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)",
+demo = gr.Interface(
+    fn=generate_response,
+    inputs=[
+        gr.Textbox(
+            label="Enter your question",
+            placeholder="Ask me anything...",
+            lines=5
         ),
     ],
+    outputs=gr.Textbox(label="Response", lines=10),
+    title="benhaotang/phi4-qwq-sky-t1",
+    description=f""" To achieve CoT and science reasoning on small scale
+    
+Model: [benhaotang/phi4-qwq-sky-t1]({MODEL_URL})""",
+    examples=[
+        ["For a scalar field theory with interaction Lagrangian $\mathcal{L}_{int} = g\phi^3 + \lambda\phi^4$:\n 1.Enumerate all possible 1-loop Feynman diagrams contributing to a 2-to-2 scattering process\n2.For each diagram, write down its corresponding amplitude\n3. Provide Mathematica code to calculate these loop amplitudes\n Please explain your reasoning step by step."]
+    ]
 )
 
-
-if __name__ == "__main__":
-    demo.launch()
+demo.launch()