app.py

import torch
from peft import PeftModel
import transformers
import gradio as gr

assert (
    "LlamaTokenizer" in transformers._import_structure["models.llama"]
), "LLaMA is now in HuggingFace's main branch.\nPlease reinstall it: pip uninstall transformers && pip install git+https://github.com/huggingface/transformers.git"
from transformers import LlamaTokenizer, LlamaForCausalLM, GenerationConfig

tokenizer = LlamaTokenizer.from_pretrained("daryl149/llama-2-13b-chat-hf")

BASE_MODEL = "daryl149/llama-2-13b-chat-hf"
LORA_WEIGHTS = "Sparticle/llama-2-13b-chat-japanese-lora"

if torch.cuda.is_available():
    device = "cuda"
else:
    device = "cpu"

try:
    if torch.backends.mps.is_available():
        device = "mps"
except:
    pass

if device == "cuda":
    model = LlamaForCausalLM.from_pretrained(
        BASE_MODEL,
        load_in_8bit=False,
        torch_dtype=torch.float16,
        device_map="auto",
    )
    model = PeftModel.from_pretrained(
        model, LORA_WEIGHTS, torch_dtype=torch.float16, force_download=True
    )
elif device == "mps":
    model = LlamaForCausalLM.from_pretrained(
        BASE_MODEL,
        device_map={"": device},
        torch_dtype=torch.float16,
    )
    model = PeftModel.from_pretrained(
        model,
        LORA_WEIGHTS,
        device_map={"": device},
        torch_dtype=torch.float16,
    )
else:
    model = LlamaForCausalLM.from_pretrained(
        BASE_MODEL, device_map={"": device}, low_cpu_mem_usage=True
    )
    model = PeftModel.from_pretrained(
        model,
        LORA_WEIGHTS,
        device_map={"": device},
    )


def generate_prompt(instruction, input=None):
    if input:
        return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
### Instruction:
{instruction}
### Input:
{input}
### Response:"""
    else:
        return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction:
{instruction}
### Response:"""

if device != "cpu":
    model.half()
model.eval()
if torch.__version__ >= "2":
    model = torch.compile(model)


def evaluate(
    instruction,
    input=None,
    temperature=0.1,
    top_p=0.75,
    top_k=40,
    num_beams=4,
    max_new_tokens=128,
    **kwargs,
):
    prompt = generate_prompt(instruction, input)
    inputs = tokenizer(prompt, return_tensors="pt")
    input_ids = inputs["input_ids"].to(device)
    generation_config = GenerationConfig(
        temperature=temperature,
        top_p=top_p,
        top_k=top_k,
        num_beams=num_beams,
        **kwargs,
    )
    with torch.no_grad():
        generation_output = model.generate(
            input_ids=input_ids,
            generation_config=generation_config,
            return_dict_in_generate=True,
            output_scores=True,
            max_new_tokens=max_new_tokens,
        )
    s = generation_output.sequences[0]
    output = tokenizer.decode(s)
    return output.split("### Response:")[1].strip().replace('</s>', '')


g = gr.Interface(
    fn=evaluate,
    inputs=[
        gr.components.Textbox(
            lines=2, label="Instruction", placeholder="例：日本語から英語に翻訳してください。"
        ),
        gr.components.Textbox(lines=2, label="Input", placeholder="天気がいいから、散歩しましょう。"),
        gr.components.Slider(minimum=0, maximum=1, value=0.1, label="Temperature"),
        gr.components.Slider(minimum=0, maximum=1, value=0.75, label="Top p"),
        gr.components.Slider(minimum=0, maximum=100, step=1, value=40, label="Top k"),
        gr.components.Slider(minimum=1, maximum=4, step=1, value=4, label="Beams"),
        gr.components.Slider(
            minimum=1, maximum=512, step=1, value=128, label="Max tokens"
        ),
    ],
    outputs=[
        gr.inputs.Textbox(
            lines=5,
            label="Output",
        )
    ],
    title="Llama2_13b_chat_Japanese_Lora",
    description="Llama-2-13b-chat-Japanese-LoRA is a multi-purpose large language model for Japanese text.\n\
        This model is presented by the joint effort of Sparticle Inc. and A. I. Hakusan Inc.\n\n\
        Llama-2-13b-chat-Japanese-LoRAは日本語テキストのための多目的大規模言語モデルです。\n\
        このモデルは、Sparticle株式会社と株式会社白山人工知能の共同開発により発表されました。",
)
g.queue(concurrency_count=1)
g.launch()