app.py

import gradio as gr
from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer
import torch
import os

model_path = "vinai/PhoGPT-7B5-Instruct"

config = AutoConfig.from_pretrained(model_path, trust_remote_code=True, token=os.environ['HK_TOKEN'])
config.init_device = "cuda"
# config.attn_config['attn_impl'] = 'triton' # Enable if "triton" installed!

model = AutoModelForCausalLM.from_pretrained(
    model_path, config=config, torch_dtype=torch.bfloat16, trust_remote_code=True, token=os.environ['HK_TOKEN']
)
# If your GPU does not support bfloat16:
# model = AutoModelForCausalLM.from_pretrained(model_path, config=config, torch_dtype=torch.float16, trust_remote_code=True)
model.eval()

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True, token=os.environ['HK_TOKEN'])


def answer(input_prompt):
    input_ids = tokenizer(input_prompt, return_tensors="pt")

    outputs = model.generate(
        inputs=input_ids["input_ids"].to("cuda"),
        attention_mask=input_ids["attention_mask"].to("cuda"),
        do_sample=True,
        temperature=1.0,
        top_k=50,
        top_p=0.9,
        max_new_tokens=1024,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.pad_token_id
    )

    response = tokenizer.batch_decode(outputs, skip_special_tokens=True)[0]
    response = response.split("### Trả lời:")[1]
    return response

iface = gr.Interface(fn=greet, inputs="text", outputs="text")
iface.launch()