app.py

import gradio as gr
import os
import spaces
from transformers import GemmaTokenizer, AutoModelForCausalLM
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from threading import Thread

# Set an environment variable
HF_TOKEN = os.environ.get("HF_TOKEN", None)


DESCRIPTION = '''
<div>
<h1 style="text-align: center;">Meta Llama3 8B</h1>
<p>This Space demonstrates the instruction-tuned model <a href="https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct"><b>Meta Llama3 8b Chat</b></a>. Meta Llama3 is the new open LLM and comes in two sizes: 8b and 70b. Feel free to play with it, or duplicate to run privately!</p>
<p>🔎 For more details about the Llama3 release and how to use the model with <code>transformers</code>, take a look <a href="https://huggingface.co/blog/llama3">at our blog post</a>.</p>
<p>🦕 Looking for an even more powerful model? Check out the <a href="https://huggingface.co/chat/"><b>Hugging Chat</b></a> integration for Meta Llama 3 70b</p>
</div>
'''

LICENSE = """
<p/>

---
Built with Meta Llama 3
"""

PLACEHOLDER = """
<div style="padding: 30px; text-align: center; display: flex; flex-direction: column; align-items: center;">
   <img src="https://ysharma-dummy-chat-app.hf.space/file=/tmp/gradio/8e75e61cc9bab22b7ce3dec85ab0e6db1da5d107/Meta_lockup_positive%20primary_RGB.jpg" style="width: 80%; max-width: 550px; height: auto; opacity: 0.55;  "> 
   <h1 style="font-size: 28px; margin-bottom: 2px; opacity: 0.55;">Meta llama3</h1>
   <p style="font-size: 18px; margin-bottom: 2px; opacity: 0.65;">Ask me anything...</p>
</div>
"""


css = """
h1 {
  text-align: center;
  display: block;
}

#duplicate-button {
  margin: auto;
  color: white;
  background: #1565c0;
  border-radius: 100vh;
}
"""


import json


import json

def str_to_json(str_obj):
    json_obj = json.loads(str_obj)
    return json_obj



import subprocess

subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)


from accelerate import init_empty_weights, infer_auto_device_map, load_checkpoint_and_dispatch
from accelerate import Accelerator

import torch
subprocess.run(
    "pip install psutil",
   
    shell=True,
)



# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("HumanLLMs/Human-Like-Mistral-Nemo-Instruct-2407")
model = AutoModelForCausalLM.from_pretrained("HumanLLMs/Human-Like-Mistral-Nemo-Instruct-2407", device_map="auto",
                                             low_cpu_mem_usage=True,
    torch_dtype=torch.bfloat16,
    # quantization_config=quantization_config,
    attn_implementation="flash_attention_2",
                                            
                                            )  # to("cuda:0") 
terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

@spaces.GPU(duration=60)
def chat_llama3_8b(message: str, 
              history: list, 
              temperature: float, 
              max_new_tokens: int
             ) -> str:

    x = str_to_json(str(message))
    conversation = x['messages']
    print(conversation)
    
    # for user, assistant in history:
    #     conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
    # conversation.append({"role": "user", "content": message})

    input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt").to(model.device)
    
    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)

    generate_kwargs = dict(
        input_ids= input_ids,
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        temperature=temperature,
        eos_token_id=terminators,
    )
    # This will enforce greedy generation (do_sample=False) when the temperature is passed 0, avoiding the crash.             
    if temperature == 0:
        generate_kwargs['do_sample'] = False
        
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        outputs.append(text)
        #print(outputs)
        yield "".join(outputs)
        

# Gradio block
chatbot=gr.Chatbot(height=450, placeholder=PLACEHOLDER, label='Gradio ChatInterface')

with gr.Blocks(fill_height=True, css=css) as demo:
    
    gr.Markdown(DESCRIPTION)
    gr.DuplicateButton(value="Duplicate Space for private use", elem_id="duplicate-button")
    gr.ChatInterface(
        fn=chat_llama3_8b,
        chatbot=chatbot,
        fill_height=True,
        additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
        additional_inputs=[
            gr.Slider(minimum=0,
                      maximum=1, 
                      step=0.1,
                      value=0.95, 
                      label="Temperature", 
                      render=False),
            gr.Slider(minimum=128, 
                      maximum=4096,
                      step=1,
                      value=512, 
                      label="Max new tokens", 
                      render=False ),
            ],
        examples=[
            ['How to setup a human base on Mars? Give short answer.'],
            ['Explain theory of relativity to me like I’m 8 years old.'],
            ['What is 9,000 * 9,000?'],
            ['Write a pun-filled happy birthday message to my friend Alex.'],
            ['Justify why a penguin might make a good king of the jungle.']
            ],
        cache_examples=False,
                     )
    
    gr.Markdown(LICENSE)
    
if __name__ == "__main__":
    demo.launch()