app.py

import requests
import os
import gradio as gr
import json
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = 'facebook/incoder-1B'
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, low_cpu_mem_usage=True)
print('load ok')

def completion(prompt, max_tokens, temperature, top_k, top_p):
    inpt = tokenizer.encode(prompt, return_tensors="pt")
    tok = len(tokenizer(prompt)['input_ids'])
    out = model.generate(inpt, max_length=tok+max_tokens, top_p=top_p, top_k=top_k, temperature=temperature, num_beams=2, repetition_penalty=2.0)
    res = tokenizer.decode(out[0])
    return res

demo = gr.Interface(
       fn=completion, 
       inputs=[
               gr.inputs.Textbox(lines=10,placeholder='Write some code..'),
               gr.inputs.Slider(10,200,10,100,'Max Tokens',False),
               gr.inputs.Slider(0,1.0,0.1,1.0,'temperature',False),
               gr.inputs.Slider(0,50,1,40,'top_k',True),
               gr.inputs.Slider(0,1.0,0.1,0.9,'top_p',True)
       ],
       outputs="text",
       allow_flagging=False,
 
)

demo.launch()