#llama 3.2 3b IT import os from threading import Thread from typing import Iterator import gradio as gr #import spaces import torch from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer DESCRIPTION = """\ # Llama 3.2 3B Instruct Llama 3.2 3B is Meta's latest iteration of open LLMs. This is a demo of [`meta-llama/Llama-3.2-3B-Instruct`](https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct), fine-tuned for instruction following. For more details, please check [our post](https://huggingface.co/blog/llama32). """ MAX_MAX_NEW_TOKENS = 2048 DEFAULT_MAX_NEW_TOKENS = 1024 MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096")) device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") # Download model from Huggingface Hub # Change this to meta-llama or the correct org name from Huggingface Hub model_id = "meta-llama/Llama-3.2-3B-Instruct" tokenizer = AutoTokenizer.from_pretrained(model_id) model = AutoModelForCausalLM.from_pretrained( model_id, device_map="auto", torch_dtype=torch.bfloat16, ) model.eval() # Main Gradio inference function def generate( message: str, chat_history: list[tuple[str, str]], max_new_tokens: int = 1024, temperature: float = 0.6, top_p: float = 0.9, top_k: int = 50, repetition_penalty: float = 1.2, ) -> Iterator[str]: conversation = [{k: v for k, v in d.items() if k != 'metadata'} for d in chat_history] conversation.append({"role": "user", "content": message}) input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt") if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH: input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:] gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.") input_ids = input_ids.to(model.device) streamer = TextIteratorStreamer(tokenizer, timeout=20.0, 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, top_p=top_p, top_k=top_k, temperature=temperature, num_beams=1, repetition_penalty=repetition_penalty, ) t = Thread(target=model.generate, kwargs=generate_kwargs) t.start() conversation.append({"role": "assistant", "content": ""}) outputs = [] for text in streamer: outputs.append(text) bot_response = "".join(outputs) conversation[-1]['content'] = bot_response yield "", conversation # Implementing Gradio 5 features and building a ChatInterface UI yourself PLACEHOLDER = """

Meta llama3.2

Learn more about Llama 3.2

""" def handle_retry(history, retry_data: gr.RetryData): new_history = history[:retry_data.index] previous_prompt = history[retry_data.index]['content'] yield from generate(previous_prompt, chat_history = new_history, max_new_tokens = 1024, temperature = 0.6, top_p = 0.9, top_k = 50, repetition_penalty = 1.2) def handle_like(data: gr.LikeData): if data.liked: print("You upvoted this response: ", data.value) else: print("You downvoted this response: ", data.value) def handle_undo(history, undo_data: gr.UndoData): chatbot = history[:undo_data.index] prompt = history[undo_data.index]['content'] return chatbot, prompt def chat_examples_fill(data: gr.SelectData): yield from generate(data.value['text'], chat_history = [], max_new_tokens = 1024, temperature = 0.6, top_p = 0.9, top_k = 50, repetition_penalty = 1.2) with gr.Blocks(theme=gr.themes.Soft(), fill_height=True) as demo: with gr.Column(elem_id="container", scale=1): chatbot = gr.Chatbot( label="Llama3.2 3B Instruct Chatbotw using Gradio 5", show_label=False, type="messages", scale=1, suggestions = [ {"text": "How many R are there in a Strawberry?"}, {"text": "What is the meaning of life for an AI?"}, {"text": "Are tomatoes vegetables?"}, {"text": "There's a llama in my garden 😱 What should I do?"}, {"text": "What is the best way to open a can of worms?"}, {"text": "The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1. "}, {"text": 'How to setup a human base on Mars? Give short answer.'}, {"text": 'Explain theory of relativity to me like I’m 8 years old.'}, {"text": 'What is 9,000 * 9,000?'}, {"text": 'Write a pun-filled happy birthday message to my friend Alex.'}, {"text": 'Justify why a penguin might make a good king of the jungle.'} ], placeholder = PLACEHOLDER, ) msg = gr.Textbox(submit_btn=True, show_label=False) with gr.Accordion('Additional inputs', open=False): max_new_tokens = gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS, ) temperature = gr.Slider(label="Temperature",minimum=0.1, maximum=4.0, step=0.1, value=0.6,) top_p = gr.Slider(label="Top-p (nucleus sampling)", minimum=0.05, maximum=1.0, step=0.05, value=0.9, ) top_k = gr.Slider(label="Top-k", minimum=1, maximum=1000, step=1, value=50, ) repetition_penalty = gr.Slider(label="Repetition penalty", minimum=1.0, maximum=2.0, step=0.05, value=1.2, ) msg.submit(generate, [msg, chatbot, max_new_tokens, temperature, top_p, top_k, repetition_penalty], [msg, chatbot]) chatbot.retry(handle_retry, chatbot, [msg, chatbot]) chatbot.like(handle_like, None, None) chatbot.undo(handle_undo, chatbot, [chatbot, msg]) chatbot.suggestion_select(chat_examples_fill, None, [msg, chatbot] ) demo.launch()