fn.py

import os
import re
import torch
import datetime
import json
import csv
import gc
from transformers import AutoTokenizer, AutoModelForCausalLM
from transformers import TextIteratorStreamer
from transformers import BitsAndBytesConfig, GPTQConfig
from threading import Thread

tokenizer = None
model = None
default_cfg = {
    'model_name': None,
    'qtype': 'bnb',
    'dtype': '4bit',
    'instruction': None,
    'inst_template': None,
    'chat_template': None,
    'max_new_tokens': 1024,
    'temperature': 0.9,
    'top_p': 0.95,
    'top_k': 40,
    'repetition_penalty': 1.2,
}
cfg = default_cfg.copy()

def load_model(model_name, qtype = 'bnb', dtype = '4bit'):
    global tokenizer, model, cfg

    if cfg['model_name'] == model_name and cfg['qtype'] == qtype and cfg['dtype'] == dtype:
        return

    del model
    del tokenizer
    model = None
    tokenizer = None
    gc.collect()
    torch.cuda.empty_cache()

    tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)

    match qtype:
        case 'bnb':
            match dtype:
                case '4bit' | 'int4':
                    kwargs = dict(
                        quantization_config=BitsAndBytesConfig(
                            load_in_4bit=True,
                            bnb_4bit_compute_dtype=torch.bfloat16,
                        ),
                    )
                case '8bit' | 'int8':
                    kwargs = dict(
                        quantization_config=BitsAndBytesConfig(
                            load_in_8bit=True,
                            bnb_4bit_compute_dtype=torch.bfloat16,
                        ),
                    )
                case 'fp16':
                    kwargs = dict(
                        torch_dtype=torch.float16,
                    )
                case 'bf16':
                    kwargs = dict(
                        torch_dtype=torch.bfloat16,
                    )
                case _:
                    kwargs = dict()
        case 'gptq':
            match dtype:
                case '4bit' | 'int4':
                    kwargs = dict(
                        quantization_config=GPTQConfig(
                            bits=4,
                            tokenizer=tokenizer,
                        ),
                    )
                case '8bit' | 'int8':
                    kwargs = dict(
                        quantization_config=GPTQConfig(
                            bits=8,
                            tokenizer=tokenizer,
                        ),
                    )
        case 'gguf':
            kwargs = dict(
                gguf_file=qtype,
            )
        case 'awq':
            match dtype:
                case 'fa2':
                    kwargs = dict(
                        use_flash_attention_2=True,
                    )
                case _:
                    kwargs = dict()
    
    model = AutoModelForCausalLM.from_pretrained(
            model_name,
            device_map="auto",
            trust_remote_code=True,
            **kwargs,
        )

    cfg['model_name'] = model_name
    cfg['qtype'] = qtype
    cfg['dtype'] = dtype

def clear_config():
    global cfg
    cfg = default_cfg.copy()

def set_config(model_name, qtype, dtype, instruction, inst_template, chat_template, max_new_tokens, temperature, top_p, top_k, repetition_penalty):
    global cfg
    load_model(model_name, qtype, dtype)
    cfg.update({
        'instruction': instruction,
        'inst_template': inst_template,
        'chat_template': chat_template,
        'max_new_tokens': int(max_new_tokens),
        'temperature': float(temperature),
        'top_p': float(top_p),
        'top_k': int(top_k),
        'repetition_penalty': float(repetition_penalty),
    })
    return 'done.'

def set_config_args(args):
    global cfg

    load_model(args['model_name'], args['qtype'], args['dtype'])
    cfg.update(args)

    return 'done.'

def chatinterface_to_messages(message, history):
    global cfg

    messages = []
    
    if cfg['instruction']:
        messages.append({'role': 'system', 'content': cfg['instruction']})

    for pair in history:
        [user, assistant] = pair
        if user:
            messages.append({'role': 'user', 'content': user})
        if assistant:
            messages.append({'role': 'assistant', 'content': assistant})

    if message:
        messages.append({'role': 'user', 'content': message})

    return messages

def chat(message, history = [], instruction = None, args = {}):
    global tokenizer, model, cfg

    if instruction:
        cfg['instruction'] = instruction
        prompt = apply_template(message)
    else:
        messages = chatinterface_to_messages(message, history)
        prompt = apply_template(messages)

    model_inputs = tokenizer([prompt], return_tensors="pt").to(model.device)

    streamer = TextIteratorStreamer(
        tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True,
    )

    generate_kwargs = dict(
        model_inputs,
        streamer=streamer,
        do_sample=True,
        num_beams=1,
    )
    for k in [
        'max_new_tokens',
        'temperature',
        'top_p',
        'top_k',
        'repetition_penalty'
        ]:
        if cfg[k]:
            generate_kwargs[k] = cfg[k]

    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    model_output = ""
    for new_text in streamer:
        model_output += new_text
        if 'fastapi' in args:
            # fastapiは差分だけを返して欲しい
            yield new_text
        else:
            # gradioは常に全文を返して欲しい
            yield model_output

    return model_output

def apply_template(messages):
    global tokenizer, cfg

    if cfg['chat_template']:
        tokenizer.chat_template = cfg['chat_template']

    if type(messages) is str:
        if cfg['inst_template']:
            return cfg['inst_template'].format(instruction=cfg['instruction'], input=messages)
        return cfg['instruction'].format(input=messages)
    if type(messages) is list:
        return tokenizer.apply_chat_template(conversation=messages, add_generation_prompt=True, tokenize=False)