app.py

from typing import Optional

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

print("starting server ...")

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 LlamaForCausalLM  # noqa
from transformers import LlamaTokenizer  # noqa

BASE_MODEL = "decapoda-research/llama-13b-hf"
LORA_WEIGHTS = "izumi-lab/llama-13b-japanese-lora-v0-1ep"

tokenizer = LlamaTokenizer.from_pretrained(BASE_MODEL)

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

try:
    if torch.backends.mps.is_available():
        device = "mps"
except Exception:
    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)
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: str, input: Optional[str] = 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: str,
    input: Optional[str] = None,
    temperature: float = 0.7,
    top_p: float = 1.0,
    top_k: int = 40,
    num_beams: int = 4,
    max_new_tokens: int = 256,
    **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()


g = gr.Interface(
    fn=evaluate,
    inputs=[
        gr.components.Textbox(lines=2, label="Instruction", placeholder="東京から大阪に行くには？"),
        gr.components.Textbox(lines=2, label="Input", placeholder="none"),
        gr.components.Slider(minimum=0, maximum=1, value=0.7, label="Temperature"),
        gr.components.Slider(minimum=0, maximum=1, value=1.0, 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="izumi-lab/calm-7b-lora-v0-1ep",
    description="izumi-lab/calm-7b-lora-v0-1ep is a 7B-parameter Calm model finetuned to follow instructions. It is trained on the [izumi-lab/llm-japanese-dataset](https://huggingface.co/datasets/izumi-lab/llm-japanese-dataset) dataset and makes use of the Huggingface Calm-7b implementation. For more information, please visit [the project's website](https://llm.msuzuki.me).",
)
g.queue(concurrency_count=1)
g.launch()