app.py

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

model_name = "numfa/numfa_v2-3b"
model = AutoModelForCausalLM.from_pretrained(model_name,torch_dtype=torch.float16, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_name)
if tokenizer.pad_token_id is None:
    tokenizer.pad_token_id = tokenizer.eos_token_id

streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens = True)

def generate_text(prompt, max_length, top_p, top_k):
    inputs = tokenizer([prompt], return_tensors="pt")

    generate_kwargs = dict(
        inputs,
        max_length=int(max_length),top_p=float(top_p), do_sample=True, top_k=int(top_k), streamer=streamer
    )

    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    generated_text=[]

    for text in streamer:
        generated_text.append(text)
        yield "".join(generated_text)

description = """
# Deploy your first ML app using Gradio
"""
inputs = [
    gr.Textbox(label="Prompt text"),
    gr.Textbox(label="max-lenth generation", value=100),
    gr.Slider(0.0, 1.0, label="top-p value", value=0.95),
    gr.Textbox(label="top-k", value=50,),
]
outputs = [gr.Textbox(label="Generated Text")]

demo = gr.Interface(fn=generate_text, inputs=inputs, outputs=outputs, allow_flagging=False, description=description)

demo.launch()