项目简介:
小李哥今天将继续介绍亚马逊云科技AWS云计算平台上的前沿前沿AI技术解决方案,帮助大家快速了解国际上最热门的云计算平台亚马逊云科技AWS上的AI软甲开发最佳实践,并应用到自己的日常工作里。本次介绍的是如何在Amazon SageMaker上微调(Fine-tune)大语言模型dolly-v2-3b,满足日常生活中不同的场景需求,并将介分享如何在SageMaker上优化模型性能并节省计算资源实现成本控制,最后将部署后的大语言模型URL集成到自己云上的软件应用中。
本方案包括通过Amazon Cloudfront和S3托管前端页面,并通过Amazon API Gateway和AWS Lambda将应用程序与AI模型集成,调用大模型实现推理。本方案的解决方案架构图如下:
利用微调模型创建的对话机器人前端UI
利用本方案小李哥用微调后的模型搭建了一个Q&A对话机器人助手,可以生成代码、文字总结、回答问题。
在开始分享案例之前,我们来了解一下本方案的技术背景,帮助大家更好的理解方案架构。
什么是Amazon SageMaker?
Amazon SageMaker 是一个完全托管的机器学习服务(大家可以理解为Serverless的Jupyter Notebook),专为应用开发和数据科学家设计,帮助他们快速构建、训练和部署机器学习模型。使用 SageMaker,您无需担心底层基础设施的管理,可以专注于模型的开发和优化。它提供了一整套工具和功能,包括数据准备、模型训练、超参数调优、模型部署和监控,简化了整个机器学习工作流程。
本方案将介绍以下内容:
1. 使用 SageMaker Jupyter Notebook进行dolly-v2-3b模型开发和微调
2. 在SageMaker部署微调后的大语言模型LLM并基于数据进行推理
3. 使用多场景的测试案例验证推理结果表现,并将部署的模型API节点集成进云端应用
项目搭建具体步骤:
下面跟着小李哥手把手微调一个亚马逊云科技AWS上的生成式AI模型(dolly-v2-3b)的软件应用,并将AI大模型部署与应用集成。
1. 在控制台进入Amazon SageMaker, 点击Notebook
2. 打开Jupyter Notebook
3. 创建一个新的Notebook:"lab-notebook.ipynb"并打开
4. 接下来我们在单元格内一步一步运行代码,检查CUDA的内存状态
bash
!nvidia-smi
5.接下来,我们安装必要依赖并导入
bash
%%capture
!pip3 install -r requirements.txt --quiet
!pip install sagemaker --quiet --upgrade --force-reinstall
python
%%capture
import os
import numpy as np
import pandas as pd
from typing import Any, Dict, List, Tuple, Union
from datasets import Dataset, load_dataset, disable_caching
disable_caching() ## disable huggingface cache
from transformers import AutoModelForCausalLM
from transformers import AutoTokenizer
from transformers import TextDataset
import torch
from torch.utils.data import Dataset, random_split
from transformers import TrainingArguments, Trainer
import accelerate
import bitsandbytes
from IPython.display import Markdown
6. 导入提前准备好的FAQs数据集
python
sagemaker_faqs_dataset = load_dataset("csv",
data_files='data/amazon_sagemaker_faqs.csv')['train']
sagemaker_faqs_dataset
sagemaker_faqs_dataset[0]
7. 我们定义用于模型推理的提示词格式
python
from utils.helpers import INTRO_BLURB, INSTRUCTION_KEY, RESPONSE_KEY, END_KEY, RESPONSE_KEY_NL, DEFAULT_SEED, PROMPT
'''
PROMPT = """{intro}
{instruction_key}
{instruction}
{response_key}
{response}
{end_key}"""
'''
Markdown(PROMPT)
8. 下面我们进入重头戏,导入一个提前预训练好的LLM大语言模型"databricks/dolly-v2-3b"。
python
tokenizer = AutoTokenizer.from_pretrained("databricks/dolly-v2-3b",
padding_side="left")
tokenizer.pad_token = tokenizer.eos_token
tokenizer.add_special_tokens({"additional_special_tokens":
[END_KEY, INSTRUCTION_KEY, RESPONSE_KEY_NL]})
model = AutoModelForCausalLM.from_pretrained(
"databricks/dolly-v2-3b",
# use_cache=False,
device_map="auto", #"balanced",
load_in_8bit=True,
)
9. 对模型训练进行预准备, 处理数据集、优化模型训练(PEFT)效率
python
model.resize_token_embeddings(len(tokenizer))
from functools import partial
from utils.helpers import mlu_preprocess_batch
MAX_LENGTH = 256
_preprocessing_function = partial(mlu_preprocess_batch, max_length=MAX_LENGTH, tokenizer=tokenizer)
encoded_sagemaker_faqs_dataset = sagemaker_faqs_dataset.map(
_preprocessing_function,
batched=True,
remove_columns=["instruction", "response", "text"],
)
processed_dataset = encoded_sagemaker_faqs_dataset.filter(lambda rec: len(rec["input_ids"]) < MAX_LENGTH)
split_dataset = processed_dataset.train_test_split(test_size=14, seed=0)
split_dataset
10. 同时我们使用LoRA(Low-Rank Adaptation)模型加速我们的模型微调
python
from peft import LoraConfig, get_peft_model, prepare_model_for_int8_training, TaskType
MICRO_BATCH_SIZE = 8
BATCH_SIZE = 64
GRADIENT_ACCUMULATION_STEPS = BATCH_SIZE // MICRO_BATCH_SIZE
LORA_R = 256 # 512
LORA_ALPHA = 512 # 1024
LORA_DROPOUT = 0.05
# Define LoRA Config
lora_config = LoraConfig(
r=LORA_R,
lora_alpha=LORA_ALPHA,
lora_dropout=LORA_DROPOUT,
bias="none",
task_type="CAUSAL_LM"
)
model = get_peft_model(model, lora_config)
model.print_trainable_parameters()
from utils.helpers import MLUDataCollatorForCompletionOnlyLM
data_collator = MLUDataCollatorForCompletionOnlyLM(
tokenizer=tokenizer, mlm=False, return_tensors="pt", pad_to_multiple_of=8
)
11. 接下来我们定义模型训练参数并开始训练。其中Batch=1,Step=20000,epoch为10.
python
EPOCHS = 10
LEARNING_RATE = 1e-4
MODEL_SAVE_FOLDER_NAME = "dolly-3b-lora"
training_args = TrainingArguments(
output_dir=MODEL_SAVE_FOLDER_NAME,
fp16=True,
per_device_train_batch_size=1,
per_device_eval_batch_size=1,
learning_rate=LEARNING_RATE,
num_train_epochs=EPOCHS,
logging_strategy="steps",
logging_steps=100,
evaluation_strategy="steps",
eval_steps=100,
save_strategy="steps",
save_steps=20000,
save_total_limit=10,
)
trainer = Trainer(
model=model,
tokenizer=tokenizer,
args=training_args,
train_dataset=split_dataset['train'],
eval_dataset=split_dataset["test"],
data_collator=data_collator,
)
model.config.use_cache = False # silence the warnings. Please re-enable for inference!
trainer.train()
12. 接下来我们将微调后的模型保存在本地
python
trainer.model.save_pretrained(MODEL_SAVE_FOLDER_NAME)
trainer.save_model()
trainer.model.config.save_pretrained(MODEL_SAVE_FOLDER_NAME)
tokenizer.save_pretrained(MODEL_SAVE_FOLDER_NAME)
13. 接下来,我们将保存到本地的模型进行部署,生成公开访问的API节点Endpoint
对部署所需要的参数进行定义和初始化
python
import boto3
import json
import sagemaker.djl_inference
from sagemaker.session import Session
from sagemaker import image_uris
from sagemaker import Model
sagemaker_session = Session()
print("sagemaker_session: ", sagemaker_session)
aws_role = sagemaker_session.get_caller_identity_arn()
print("aws_role: ", aws_role)
aws_region = boto3.Session().region_name
print("aws_region: ", aws_region)
image_uri = image_uris.retrieve(framework="djl-deepspeed",
version="0.22.1",
region=sagemaker_session._region_name)
print("image_uri: ", image_uri)
进行模型部署
python
model_data="s3://{}/lora_model.tar.gz".format(mybucket)
model = Model(image_uri=image_uri,
model_data=model_data,
predictor_cls=sagemaker.djl_inference.DJLPredictor,
role=aws_role)
14.最后我们写入提示词,对大语言模型进行测试, 得到推理
python
outputs = predictor.predict({"inputs": "What solutions come pre-built with Amazon SageMaker JumpStart?"})
from IPython.display import Markdown
Markdown(outputs)