首页 > 营销学院 > AI智能

使用GaLore在本地GPU进行高效的LLM调优

训练大型语言模型(llm)是一项计算密集型的任务，即使是那些“只有”70亿个参数的模型也是如此。这种级别的训练需要的资源超出了大多数个人爱好者的能力范围。为了弥补这一差距，出现了低秩适应(lora)等参数高效方法，使得在消费级gpu上可以对大量模型进行微调。

GaLore是一种创新方法，它采用优化参数训练方式来减少VRAM需求，而非简单减少参数数量。这意味着GaLore是一种新的模型训练策略，允许模型充分利用全部参数进行学习，并比LoRA更有效地节省内存。

GaLore通过将这些梯度映射到低维空间，有效减轻了计算负担，同时保留了关键的训练信息。与传统优化器在反向传播时一次性更新所有层不同，GaLore采用逐层更新的方式进行反向传播。这种策略显著减少了训练过程中的内存占用，进一步优化了性能。

就像LoRA一样，GaLore使我们能够在消费级GPU上微调7B模型，该GPU配备了高达24 GB的VRAM。结果显示，模型的性能与全参数微调相当，甚至似乎优于LoRA。

☞☞☞AI 智能聊天, 问答助手, AI 智能搜索, 免费无限量使用 DeepSeek R1 模型☜☜☜

优于目前Hugging Face还没有官方代码，我们就来手动使用论文的代码进行训练，并与LoRA进行对比

安装依赖

首先就要安装GaLore

pip install galore-torch

然后我们还要一下这些库，并且请注意版本

datasets==2.18.0 transformers==4.39.1 trl==0.8.1 accelerate==0.28.0 torch==2.2.1

调度器和优化器的类

Galore分层优化器是通过模型权重挂钩激活的。由于我们使用Hugging Face Trainer，还需要自己实现一个优化器和调度器的抽象类。这些类的结构不执行任何操作。

from typing import Optional import torch  # Approach taken from Hugging Face transformers https://github.com/huggingface/transformers/blob/main/src/transformers/optimization.py class LayerWiseDummyOptimizer(torch.optim.Optimizer):def __init__(self, optimizer_dict=None, *args, **kwargs):dummy_tensor = torch.randn(1, 1)self.optimizer_dict = optimizer_dictsuper().__init__([dummy_tensor], {"lr": 1e-03}) def zero_grad(self, set_to_none: bool = True) -> None: pass def step(self, closure=None) -> Optional[float]: pass  class LayerWiseDummyScheduler(torch.optim.lr_scheduler.LRScheduler):def __init__(self, *args, **kwargs):optimizer = LayerWiseDummyOptimizer()last_epoch = -1verbose = Falsesuper().__init__(optimizer, last_epoch, verbose) def get_lr(self): return [group["lr"] for group in self.optimizer.param_groups] def _get_closed_form_lr(self): return self.base_lrs

加载GaLore优化器

GaLore优化器的目标是特定的参数，主要是那些在线性层中以attn或mlp命名的参数。通过系统地将函数与这些目标参数挂钩，GaLore 8位优化器就会开始工作。

from transformers import get_constant_schedule from functools import partial import torch.nn import bitsandbytes as bnb  from galore_torch import GaLoreAdamW8bit  def load_galore_optimizer(model, lr, galore_config):# function to hook optimizer and scheduler to a given parameter def optimizer_hook(p, optimizer, scheduler):if p.grad is not None: optimizer.step()optimizer.zero_grad()scheduler.step() # Parameters to optimize with Galoregalore_params = [(module.weight, module_name) for module_name, module in model.named_modules() if isinstance(module, nn.Linear) and any(target_key in module_name for target_key in galore_config["target_modules_list"])] id_galore_params = {id(p) for p, _ in galore_params} # Hook Galore optim to all target params, Adam8bit to all othersfor p in model.parameters():if p.requires_grad:if id(p) in id_galore_params:optimizer = GaLoreAdamW8bit([dict(params=[p], **galore_config)], lr=lr)else:optimizer = bnb.optim.Adam8bit([p], lr = lr)scheduler = get_constant_schedule(optimizer) p.register_post_accumulate_grad_hook(partial(optimizer_hook, optimizer=optimizer, scheduler=scheduler)) # return dummies, stepping is done with hooks return LayerWiseDummyOptimizer(), LayerWiseDummyScheduler()

HF Trainer

准备好优化器后，我们开始使用Trainer进行训练。下面是一个简单的例子，使用TRL的SFTTrainer (Trainer的子类)在Open Assistant数据集上微调llama2-7b，并在RTX 3090/4090等24 GB VRAM GPU上运行。

from transformers import AutoTokenizer, AutoModelForCausalLM, TrainingArguments, set_seed, get_constant_schedule from trl import SFTTrainer, setup_chat_format, DataCollatorForCompletionOnlyLM from datasets import load_dataset import torch, torch.nn as nn, uuid, wandb  lr = 1e-5  # GaLore optimizer hyperparameters galore_config = dict(target_modules_list = ["attn", "mlp"], rank = 1024, update_proj_gap = 200, scale = 2, proj_type="std" )  modelpath = "meta-llama/Llama-2-7b" model = AutoModelForCausalLM.from_pretrained(modelpath,torch_dtype=torch.bfloat16,attn_implementation = "flash_attention_2",device_map = "auto",use_cache = False, ) tokenizer = AutoTokenizer.from_pretrained(modelpath, use_fast = False)  # Setup for ChatML model, tokenizer = setup_chat_format(model, tokenizer) if tokenizer.pad_token in [None, tokenizer.eos_token]: tokenizer.pad_token = tokenizer.unk_token  # subset of the Open Assistant 2 dataset, 4000 of the top ranking conversations dataset = load_dataset("g-ronimo/oasst2_top4k_en")  training_arguments = TrainingArguments(output_dir = f"out_{run_id}",evaluation_strategy = "steps",label_names = ["labels"],per_device_train_batch_size = 16,gradient_accumulation_steps = 1,save_steps = 250,eval_steps = 250,logging_steps = 1, learning_rate = lr,num_train_epochs = 3,lr_scheduler_type = "constant",gradient_checkpointing = True,group_by_length = False, )  optimizers = load_galore_optimizer(model, lr, galore_config)  trainer = SFTTrainer(model = model,tokenizer = tokenizer,train_dataset = dataset["train"],eval_dataset = dataset['test'],data_collator = DataCollatorForCompletionOnlyLM(instruction_template = "<|im_start|>user", response_template = "<|im_start|>assistant", tokenizer = tokenizer, mlm = False),max_seq_length = 256,dataset_kwargs = dict(add_special_tokens = False),optimizers = optimizers,args = training_arguments, )  trainer.train()

GaLore优化器带有一些需要设置的超参数如下:

target_modules_list:指定GaLore针对的层

rank:投影矩阵的秩。与LoRA类似，秩越高，微调就越接近全参数微调。GaLore的作者建议7B使用1024

update_proj_gap:更新投影的步骤数。这是一个昂贵的步骤，对于7B来说大约需要15分钟。定义更新投影的间隔，建议范围在50到1000步之间。

scale:类似于LoRA的alpha的比例因子，用于调整更新强度。在尝试了几个值之后，我发现scale=2最接近于经典的全参数微调。