ModularityAI/LLama3Chat

1import gradio as gr2import spaces3import torch4 5import transformers6import torch7from transformers import AutoModelForCausalLM, AutoTokenizer8 9model_name = "meta-llama/Meta-Llama-3-8B-Instruct"10 11pipeline = transformers.pipeline(12    "text-generation",13    model=model_name,14    model_kwargs={"torch_dtype": torch.bfloat16},15    device="cuda",16)17 18@spaces.GPU19def chat_function(message, history, system_prompt,max_new_tokens,temperature):20    messages = [21        {"role": "system", "content": system_prompt},22        {"role": "user", "content": message},23    ]24    prompt = pipeline.tokenizer.apply_chat_template(25        messages,26        tokenize=False,27        add_generation_prompt=True28    )29    terminators = [30        pipeline.tokenizer.eos_token_id,31        pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")32    ]33    temp = temperature + 0.134    outputs = pipeline(35        prompt,36        max_new_tokens=max_new_tokens,37        eos_token_id=terminators,38        do_sample=True,39        temperature=temp,40        top_p=0.9,41    )42    return outputs[0]["generated_text"][len(prompt):]43 44gr.ChatInterface(45    chat_function,46    chatbot=gr.Chatbot(height=400),47    textbox=gr.Textbox(placeholder="Enter message here", container=False, scale=7),48    title="LLAMA 3 8B Chat",49    description="""50    This space is dedicated for chatting with Meta's Latest LLM - Llama 8b Instruct. Find this model here: https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct 51    Feel free to play with customization in the "Additional Inputs".52    """,53    theme="soft",54    additional_inputs=[55        gr.Textbox("You are helpful AI.", label="System Prompt"),56        gr.Slider(512, 4096, label="Max New Tokens"),57        gr.Slider(0, 1, label="Temperature")58    ]59).launch()