msj19/mask_gdn_hrr2

🔥 Flame: Flash Linear Attention Made Easy

The flame project has been migrated to a new project built on torchtitan.
Please visit the new repository for details and updates.

The code here is now archived as legacy, and no future updates will be synchronized here.

A minimal framework for training FLA models, whether from scratch or through finetuning.

Built on the robust infrastructure of 🤗, flame enables you to train large language models with just a few lines of code: we use datasets for data processing, transformers for model definitions, and accelerate[^1] for seamless distributed training.

In this README, we will guide you through the process of using flame to train GLA models.

Setup

To get started, you'll need to install the required packages. Both fla and flame have minimal dependencies. Clone the fla repository and install the necessary packages as follows:

git clone https://github.com/sustcsonglin/flash-linear-attention.git
pip install . 
pip install accelerate

The 🤗 tokenizers have some memory leak issues when processing very long documents. To address this, please ensure you install tokenizers>=0.20.4.

Preprocessing

Before training, you need to download and pre-tokenize your dataset. We provide a straightforward script for this. For instance, to tokenize a 10B sample of the fineweb-edu dataset, run:

python preprocess.py \
  --dataset HuggingFaceFW/fineweb-edu \
  --name sample-10BT \
  --split train \
  --context_length 2048

python preprocess.py \
  --dataset /mnt/jfzn/msj/fineweb100B_hf/datasets--HuggingFaceFW--fineweb-edu/sample/100BT \
  --name sample-100BT \
  --split train \
  --context_length 2048

/mnt/jfzn/msj/fineweb100B_hf/datasets--HuggingFaceFW--fineweb-edu/sample/100BT

This will cache the processed dataset at data/HuggingFaceFW/fineweb-edu/sample-10BT/train.

GLA utilizes a subset of Slimpajama for pretraining in the paper. Given the size of the dataset, the fastest way to download it is using git lfs (refer to this issue).

git lfs install
git clone https://huggingface.co/datasets/cerebras/SlimPajama-627B --depth 1
python preprocess.py \
  --dataset SlimPajama-627B \
  --split train \
  --context_length 2048

Training from scratch

To train your 340M model from scratch, execute the following command:

bash train.sh \
  type=gla \
  lr=3e-4 \
  scheduler=cosine_with_min_lr \
  batch=32 \
  update=1 \
  warmup=1024 \
  steps=20480 \
  context=2048 \
  gpus=8 \  
  nodes=1 \
  path=exp/gla-340M-10B \
  project=fla \
  model=configs/gla_340M.json \
  data=HuggingFaceFW/fineweb-edu \
  name=sample-10BT \
  cache=data/HuggingFaceFW/fineweb-edu/sample-10BT/train

Key parameters:

	Description	Default
lr	learning_rate	3e-4
scheduler	lr_scheduler_type	cosine_with_min_lr
batch	batch_size	32
update	gradient_accumulation_steps	1
context	context_length	2048
gpus	num_gpus_per_node	8
nodes	num_nodes	1
warmup	warmup_steps	1024
steps	max_steps	20480

The learning rate is set to 3e-4 by default, equipped with a cosine scheduler. Other scheduler types like WSD (warmup_stable_decay)[^2] are also supported.

The total number of tokens processed per batch, referred to as global_batch_size, is calculated as batch_size × gradient_accumulation_steps × context_length × num_gpus_per_node × num_nodes. For instance, in the 340M model example, the global_batch_size calculates to $32 \times 1 \times 2048 \times 8 \times 1 = 524,288$ (0.5M tokens).

The warmup_steps parameter indicates the number of steps for the learning rate warmup phase, while max_steps represents the maximum number of training steps. Each step processes global_batch_size tokens. Consequently, 512 and 20480 correspond to processing 0.5B and 10B tokens, respectively.

:warning: Monitor the value of global_batch_size, warmup_steps, and max_steps carefully when modifying any of the hyperparameters!!

flame also supports resuming interrupted training by specifying the checkpoint path. Simply use the following command:

bash train.sh \
  type=gla \
  lr=3e-4 \
  steps=20480 \
  batch=32 \
  update=1 \
  warmup=1024 \
  context=2048 \
  gpus=8 \  
  nodes=1 \
  path=exp/gla-340M-10B \
  project=fla \
  model=configs/gla_340M.json \
  data=HuggingFaceFW/fineweb-edu \
  name=sample-10BT \
  cache=data/HuggingFaceFW/fineweb-edu/sample-10BT/train \
  checkpoint=exp/gla-340M-10B/checkpoint-8192

You can also use wandb to monitor your training process effectively.

Continual Pretraining

flame supports continual training from a pretrained checkpoint. Below, we provide an example of how to finetune Mistral-7B to GLA. You can follow similar steps to reproduce the results in the GSA paper:

1. Initialize a brand-new GLA-7B model from the config and copy the mathced pretrained weights from Mistral-7B:

cd ../utils
python convert_from_llama.py \
  --model mistralai/Mistral-7B-v0.1 \
  --config ../training/configs/gla_7B.json \
  --output ../training/converted/gla-7B
cd -

2. Directly launch training from the converted checkpoint:

bash train.sh \
  type=gla \
  lr=3e-5 \
  steps=10240 \
  batch=4 \
  update=8 \
  warmup=512 \
  context=2048 \
  path=exp/gla-7B-20B \
  project=fla \
  model=converted/gla-7B \
  data=SlimPajama-627B \
  cache=data/SlimPajama-627B/train

Please be aware that finetuning on a single node may not be the most efficient approach. If available, consider leveraging multi-node GPUs for optimal performance. You can find guidance on how to launch a multi-node job in the accelerate tutorial.

[^1]: The accelerate library supports various distributed frameworks, like deepspeed and megatron for large-scale training. We use deepspeed in our case. [^2]: https://arxiv.org/abs/2404.06395