app.py

from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM, AutoTokenizer
import gradio as gr

config = PeftConfig.from_pretrained("pranjalpandey/gemma-open-instruct")
model = AutoModelForCausalLM.from_pretrained("google/gemma-2b")
model = PeftModel.from_pretrained(model, "pranjalpandey/gemma-open-instruct")
# model = AutoPeftModelForCausalLM.from_pretrained("pranjalpandey/llama-7b-finetuned-dialogue-summarizer")
tokenizer = AutoTokenizer.from_pretrained("pranjalpandey/gemma-open-instruct")
model = model.to("cuda")

def response(prompt):
  inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
  outputs = model.generate(input_ids=inputs["input_ids"], max_new_tokens=100)
  return tokenizer.batch_decode(outputs.detach().cpu().numpy(), skip_special_tokens=True)[0].split("# Response:")[1].strip()

ir = gr.Interface(
    fn=response,
    inputs=["text"],
    outputs=["text"],
)

ir.launch()