app.py

import gradio as gr
from unsloth import FastLanguageModel
from peft import PeftModel
import torch
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Sumit404/Llama-3.2-3B-Instruct-bnb-4bit-finetuned", torch_dtype="auto")
# Load the base model and tokenizer
max_seq_length = 4096
dtype = None
load_in_4bit = True

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="unsloth/Llama-3.2-3B-Instruct-bnb-4bit",
    max_seq_length=max_seq_length,
    dtype=dtype,
    load_in_4bit=load_in_4bit
)

# Load the LoRA adapters
LORA_ADAPTER_PATH = "Sumit404/Llama-3.2-3B-Instruct-bnb-4bit-finetuned" # Replace with your repo ID
model = PeftModel.from_pretrained(model, LORA_ADAPTER_PATH)

# Set tokenizer and model for inference
from unsloth.chat_templates import get_chat_template
tokenizer = get_chat_template(
   tokenizer,
   chat_template = "llama-3.2",
)
tokenizer.pad_token = tokenizer.eos_token
FastLanguageModel.for_inference(model)

def generate_text(prompt):
    messages = [{"role": "user", "content": prompt}]
    inputs = tokenizer.apply_chat_template(
       messages,
       tokenize=True,
       add_generation_prompt=True,
       return_tensors="pt",
       padding=True,
    ).to("cuda")

    attention_mask = inputs != tokenizer.pad_token_id

    outputs = model.generate(
       input_ids=inputs,
       attention_mask=attention_mask,
       max_new_tokens=128, # Increased output length for potentially longer answers
       use_cache=True,
       temperature=0.6,
       min_p=0.1,
    )

    text = tokenizer.decode(outputs[0], skip_special_tokens=True)
    # Extract only the assistant's response
    assistant_response_start = text.find("<|start_header_id|>assistant<|end_header_id|>\n\n")
    if assistant_response_start != -1:
        text = text[assistant_response_start + len("<|start_header_id|>assistant<|end_header_id|>\n\n"):]

    return text

# Create the Gradio interface
interface = gr.Interface(
    fn=generate_text,
    inputs=gr.Textbox(lines=2, placeholder="Enter your prompt here..."),
    outputs="text",
    title="Fine-tuned Llama-3.2 Instruct Model",
    description="Ask a question to the fine-tuned model."
)

# To run this in Colab, set share=True
interface.launch(share=True)