llama3.0-8B-merged-4bit

How to use :

!pip install --no-deps packaging ninja einops peft accelerate bitsandbytes
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
from peft import PeftModel, PeftConfig

# Load model and tokenizer configurations
config = PeftConfig.from_pretrained("Vijayendra/llama3.0-8B-merged-4bit")
base_model = AutoModelForCausalLM.from_pretrained("unsloth/llama-3-8b-bnb-4bit")
model = PeftModel.from_pretrained(base_model, "Vijayendra/llama3.0-8B-merged-4bit")
tokenizer = AutoTokenizer.from_pretrained("Vijayendra/llama3.0-8B-merged-4bit")

# Ensure padding token is set for the tokenizer
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

# Define the inference function with TextStreamer
def generate_answer_with_stream(model, tokenizer, text, max_new_tokens=1024, temperature=0.5, top_k=40, top_p=0.9):
    prompt = f"Answer the following question\n\n{text}\n\nQuestion:"
    
    # Tokenize the input text
    inputs = tokenizer(prompt, return_tensors="pt", padding=True, truncation=True).to(model.device)
    
    # Initialize the TextStreamer
    streamer = TextStreamer(tokenizer)

    # Generate answer using the model with streaming
    with torch.no_grad():
        model.generate(
            inputs.input_ids,
            attention_mask=inputs.attention_mask,
            max_new_tokens=max_new_tokens,
            temperature=temperature,
            do_sample=True,
            top_k=top_k,
            top_p=top_p,
            repetition_penalty=1.2,
            eos_token_id=tokenizer.eos_token_id,
            pad_token_id=tokenizer.pad_token_id,
            streamer=streamer  # Stream output as it's generated
        )

# Input Question
question = "What is quantum mechanics?"

# Generate and print answer
generate_answer_with_stream(model, tokenizer, question)