app.py

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

model_name = "zirui3/gpt_1.4B_oa_instruct"
tokenizer = AutoTokenizer.from_pretrained(model_name)

chip_map= {
    'gpt_neox.embed_in': 0,
    'gpt_neox.layers': 0,
    'gpt_neox.final_layer_norm': 0,
    'embed_out': 0
 }
model = AutoModelForCausalLM.from_pretrained(name, device_map=chip_map, torch_dtype=torch.float16, load_in_8bit=True)
#model = AutoModelForCausalLM.from_pretrained(model_name)


def predict(input, history=[], MAX_NEW_TOKENS = 500):
    text = "User: " + input + "\n\nChip: "
    new_user_input_ids = tokenizer(text, return_tensors="pt").input_ids
    # bot_input_ids = torch.cat([torch.LongTensor(history), new_user_input_ids], dim=-1).to("cuda")
    bot_input_ids = torch.cat([torch.LongTensor(history), new_user_input_ids], dim=-1)

    generated_ids = model.generate(bot_input_ids,
        max_length=MAX_NEW_TOKENS, pad_token_id=tokenizer.eos_token_id,
        do_sample=True,
        top_p=0.95, temperature=0.5, penalty_alpha=0.6, top_k=4, repetition_penalty=1.03,
        num_return_sequences=1)

    response = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
    history = generated_ids.tolist()

    # convert to list of user & bot response
    response = response.split("\n\n")
    response_pairs = [(response[i], response[i+1]) for i in range(0, len(response)-1, 2)]
    return response_pairs, history


with gr.Blocks() as demo:
    chatbot = gr.Chatbot()
    state = gr.State([])

    with gr.Row():
        txt = gr.Textbox(show_label=False, placeholder="Enter text and press enter").style(container=False)

    txt.submit(predict, [txt, state], [chatbot, state])


if __name__ == "__main__":
    # demo.launch(debug=True, server_name="0.0.0.0", server_port=9991)
    demo.launch()