app.py

import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from transformers import StoppingCriteria, StoppingCriteriaList, TextIteratorStreamer
from threading import Thread


bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=True,
)
# Loading the tokenizer and model from Hugging Face's model hub.
tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-v0.1")
model = AutoModelForCausalLM.from_pretrained("mistralai/Mistral-7B-v0.1",
        load_in_4bit=True,
        quantization_config=bnb_config,
        torch_dtype=torch.bfloat16,
        device_map="auto",
        trust_remote_code=True)

# using CUDA for an optimal experience
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device)


# Defining a custom stopping criteria class for the model's text generation.
class StopOnTokens(StoppingCriteria):
    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        stop_ids = [2]  # IDs of tokens where the generation should stop.
        for stop_id in stop_ids:
            if input_ids[0][-1] == stop_id:  # Checking if the last generated token is a stop token.
                return True
        return False


# Function to generate model predictions.
def predict(message, history):
    history_transformer_format = history + [[message, ""]]
    stop = StopOnTokens()

    # Formatting the input for the model.
    messages = "</s>".join(["</s>".join(["\n<|user|>:" + item[0], "\n<|assistant|>:" + item[1]])
                        for item in history_transformer_format])
    model_inputs = tokenizer([messages], return_tensors="pt").to(device)
    streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        model_inputs,
        streamer=streamer,
        max_new_tokens=1024,
        do_sample=True,
        top_p=0.95,
        top_k=50,
        temperature=0.7,
        num_beams=1,
        stopping_criteria=StoppingCriteriaList([stop])
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()  # Starting the generation in a separate thread.
    partial_message = ""
    for new_token in streamer:
        partial_message += new_token
        if '</s>' in partial_message:  # Breaking the loop if the stop token is generated.
            break
        yield partial_message


# Setting up the Gradio chat interface.
gr.ChatInterface(predict,
                 title="Tinyllama_chatBot",
                 description="Ask Tiny llama any questions",
                 examples=['How to cook a fish?', 'Who is the president of US now?']
                 ).launch()  # Launching the web interface.