Update app.py

app.py

@@ -1,48 +1,168 @@
-from huggingface_hub import InferenceClient
+import os
+import time
+import spaces
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
 import gradio as gr
+from threading import Thread
 
-client = InferenceClient("meta-llama/Meta-Llama-3.1-8B")
+MODEL_LIST = ["meta-llama/Meta-Llama-3.1-8B-Instruct"]
+HF_TOKEN = os.environ.get("HF_TOKEN", None)
+MODEL = os.environ.get("MODEL_ID")
 
-def format_prompt(message, history):
-    fixed_prompt= """            """
-    prompt = f"<s>{fixed_prompt}"
-    for user_prompt, bot_response in history:
-        prompt += f"\n User:{user_prompt}\n LLM Response:{bot_response}"
-    prompt += f"\nUser: {message}\nLLM Response:"
+TITLE = "<h1><center>Meta-Llama3.1-8B</center></h1>"
 
-    return prompt
+PLACEHOLDER = """
+<center>
+<p>Hi! How can I help you today?</p>
+</center>
+"""
 
-def generate(
-    prompt, history, temperature=0.1, max_new_tokens=2048, top_p=0.8, repetition_penalty=1.0,
+
+CSS = """
+.duplicate-button {
+    margin: auto !important;
+    color: white !important;
+    background: black !important;
+    border-radius: 100vh !important;
+}
+h3 {
+    text-align: center;
+}
+"""
+
+device = "cuda" # for GPU usage or "cpu" for CPU usage
+
+quantization_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_compute_dtype=torch.bfloat16,
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_quant_type= "nf4")
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL)
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL,
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    quantization_config=quantization_config)
+
+@spaces.GPU()
+def stream_chat(
+    message: str, 
+    history: list,
+    system_prompt: str,
+    temperature: float = 0.8, 
+    max_new_tokens: int = 1024, 
+    top_p: float = 1.0, 
+    top_k: int = 20, 
+    penalty: float = 1.2,
 ):
-    temperature = float(temperature)
-    if temperature < 1e-2:
-        temperature = 1e-2
-    top_p = float(top_p)
+    print(f'message: {message}')
+    print(f'history: {history}')
 
+    conversation = [
+        {"role": "system", "content": system_prompt}
+    ]
+    for prompt, answer in history:
+        conversation.extend([
+            {"role": "user", "content": prompt}, 
+            {"role": "assistant", "content": answer},
+        ])
+
+    conversation.append({"role": "user", "content": message})
+
+    input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt").to(model.device)
+    
+    streamer = TextIteratorStreamer(tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
+    
     generate_kwargs = dict(
-        temperature=temperature,
-        max_new_tokens=max_new_tokens,
-        top_p=top_p,
-        repetition_penalty=repetition_penalty,
-        do_sample=True,
-        seed=42,
+        input_ids=input_ids, 
+        max_new_tokens = max_new_tokens,
+        do_sample = False if temperature == 0 else True,
+        top_p = top_p,
+        top_k = top_k,
+        temperature = temperature,
+        repetition_penalty=penalty,
+        eos_token_id=[128001,128008,128009],
+        streamer=streamer,
     )
 
-    formatted_prompt = format_prompt(prompt, history)
-
-    stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
-   
-    yield stream
+    with torch.no_grad():
+        thread = Thread(target=model.generate, kwargs=generate_kwargs)
+        thread.start()
+        
+    buffer = ""
+    for new_text in streamer:
+        buffer += new_text
+        yield buffer
 
+            
+chatbot = gr.Chatbot(height=600, placeholder=PLACEHOLDER)
 
+with gr.Blocks(css=CSS, theme="soft") as demo:
+    gr.HTML(TITLE)
+    gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button")
+    gr.ChatInterface(
+        fn=stream_chat,
+        chatbot=chatbot,
+        fill_height=True,
+        additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
+        additional_inputs=[
+            gr.Textbox(
+                value="You are a helpful assistant",
+                label="System Prompt",
+                render=False,
+            ),
+            gr.Slider(
+                minimum=0,
+                maximum=1,
+                step=0.1,
+                value=0.8,
+                label="Temperature",
+                render=False,
+            ),
+            gr.Slider(
+                minimum=128,
+                maximum=8192,
+                step=1,
+                value=1024,
+                label="Max new tokens",
+                render=False,
+            ),
+            gr.Slider(
+                minimum=0.0,
+                maximum=1.0,
+                step=0.1,
+                value=1.0,
+                label="top_p",
+                render=False,
+            ),
+            gr.Slider(
+                minimum=1,
+                maximum=20,
+                step=1,
+                value=20,
+                label="top_k",
+                render=False,
+            ),
+            gr.Slider(
+                minimum=0.0,
+                maximum=2.0,
+                step=0.1,
+                value=1.2,
+                label="Repetition penalty",
+                render=False,
+            ),
+        ],
+        examples=[
+            ["Help me study vocabulary: write a sentence for me to fill in the blank, and I'll try to pick the correct option."],
+            ["What are 5 creative things I could do with my kids' art? I don't want to throw them away, but it's also so much clutter."],
+            ["Tell me a random fun fact about the Roman Empire."],
+            ["Show me a code snippet of a website's sticky header in CSS and JavaScript."],
+        ],
+        cache_examples=False,
+    )
 
-demo = gr.ChatInterface (fn=generate, 
-                        title="Mood-Based Music Recommender",
-                        retry_btn=None,
-                        undo_btn=None,
-                        clear_btn=None,
-                        description="<span style='font-size: larger; font-weight: bold;'>Hi! I'm your music buddy—tell me about your mood and the type of tunes you're in the mood for today!</span>",
-                       )
 
-demo.queue().launch()
+if __name__ == "__main__":
+    demo.launch()