inference_llama.py

import os
import sys

import fire
import gradio as gr
import torch
import transformers
from peft import PeftModel
from transformers import GenerationConfig, LlamaForCausalLM, LlamaTokenizer

from alpaca_lora.utils.prompter import Prompter

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

try:
    if torch.backends.mps.is_available():
        device = "mps"
except:  # noqa: E722
    pass


def main(
    load_8bit: bool = False,
    base_model: str = "decapoda-research/llama-7b-hf",
    prompt_template: str = "",  # The prompt template to use, will default to alpaca.
):
    base_model = base_model or os.environ.get("BASE_MODEL", "")
    assert (
        base_model
    ), "Please specify a --base_model, e.g. --base_model='decapoda-research/llama-7b-hf'"

    prompter = Prompter(prompt_template)
    tokenizer = LlamaTokenizer.from_pretrained(base_model)
    if device == "cuda":
        model = LlamaForCausalLM.from_pretrained(
            base_model,
            load_in_8bit=load_8bit,
            torch_dtype=torch.float16,
            device_map="auto",
        )
    elif device == "mps":
        model = LlamaForCausalLM.from_pretrained(
            base_model,
            device_map={"": device},
            torch_dtype=torch.float16,
        )
    else:
        model = LlamaForCausalLM.from_pretrained(
            base_model, device_map={"": device}, low_cpu_mem_usage=True
        )

    # unwind broken decapoda-research config
    model.config.pad_token_id = tokenizer.pad_token_id = 0  # unk
    model.config.bos_token_id = 1
    model.config.eos_token_id = 2

    if not load_8bit:
        model.half()  # seems to fix bugs for some users.

    model.eval()
    if torch.__version__ >= "2" and sys.platform != "win32":
        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=512,
        do_sample=False,
        **kwargs,
    ):
        prompt = instruction
        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,
            do_sample=do_sample,
            **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[len(prompt):]

    while True:
        print()
        instruction = input("Instruction: ")
        if instruction == 'quit':
            break

        result = evaluate(instruction)
        print('response:')
        print(result)

    """
    # testing code for readme
    for instruction in [
        "Tell me about alpacas.",
        "Tell me about the president of Mexico in 2019.",
        "Tell me about the king of France in 2019.",
        "List all Canadian provinces in alphabetical order.",
        "Write a Python program that prints the first 10 Fibonacci numbers.",
        "Write a program that prints the numbers from 1 to 100. But for multiples of three print 'Fizz' instead of the number and for the multiples of five print 'Buzz'. For numbers which are multiples of both three and five print 'FizzBuzz'.",  # noqa: E501
        "Tell me five words that rhyme with 'shock'.",
        "Translate the sentence 'I have no mouth but I must scream' into Spanish.",
        "Count up from 1 to 500.",
    ]:
        print("Instruction:", instruction)
        print("Response:", evaluate(instruction))
        print()
    """


if __name__ == "__main__":
    fire.Fire(main)