Benchmark for Tuning Accuracy and Efficiency

Overview

The benchmark includes our efforts in using Colossal-AI to train different tasks to achieve SOTA results. We are interested in both validataion accuracy and training speed, and prefer larger batch size to take advantage of more GPU devices. For example, we trained vision transformer with batch size 512 on CIFAR10 and 4096 on ImageNet1k, which are basically not used in existing works. Some of the results in the benchmark trained with 8x A100 are shown below.

Task	Model	Training Time	Top-1 Accuracy
CIFAR10	ViT-Lite-7/4	~ 16 min	~ 90.5%
ImageNet1k	ViT-S/16	~ 16.5 h	~ 74.5%

The train.py script in each task runs training with the specific configuration script in configs/ for different parallelisms. Supported parallelisms include data parallel only (ends with vanilla), 1D (ends with 1d), 2D (ends with 2d), 2.5D (ends with 2p5d), 3D (ends with 3d).

Each configuration scripts basically includes the following elements, taking ImageNet1k task as example:

TOTAL_BATCH_SIZE = 4096
LEARNING_RATE = 3e-3
WEIGHT_DECAY = 0.3

NUM_EPOCHS = 300
WARMUP_EPOCHS = 32

# data parallel only
TENSOR_PARALLEL_SIZE = 1    
TENSOR_PARALLEL_MODE = None

# parallelism setting
parallel = dict(
    pipeline=1,
    tensor=dict(mode=TENSOR_PARALLEL_MODE, size=TENSOR_PARALLEL_SIZE),
)

fp16 = dict(mode=AMP_TYPE.TORCH, ) # amp setting

gradient_accumulation = 2 # accumulate 2 steps for gradient update

BATCH_SIZE = TOTAL_BATCH_SIZE // gradient_accumulation # actual batch size for dataloader

clip_grad_norm = 1.0 # clip gradient with norm 1.0

Upper case elements are basically what train.py needs, and lower case elements are what Colossal-AI needs to initialize the training.

Usage

To start training, use the following command to run each worker:

$ DATA=/path/to/dataset python train.py --world_size=WORLD_SIZE \
                                        --rank=RANK \
                                        --local_rank=LOCAL_RANK \
                                        --host=MASTER_IP_ADDRESS \
                                        --port=MASTER_PORT \
                                        --config=CONFIG_FILE

It is also recommended to start training with torchrun as:

$ DATA=/path/to/dataset torchrun --nproc_per_node=NUM_GPUS_PER_NODE \
                                 --nnodes=NUM_NODES \
                                 --node_rank=NODE_RANK \
                                 --master_addr=MASTER_IP_ADDRESS \
                                 --master_port=MASTER_PORT \
                                 train.py --config=CONFIG_FILE

ColossalAI-Benchmark - Performance benchmarking with ColossalAI

Related tags

Overview

Benchmark for Tuning Accuracy and Efficiency

Overview

Usage

Owner

HPC-AI Tech

Parametric Contrastive Learning (ICCV2021)

[ICCV 2021] Encoder-decoder with Multi-level Attention for 3D Human Shape and Pose Estimation

BRepNet: A topological message passing system for solid models

Plug and play transformer you can find network structure and official complete code by clicking List

A simple, unofficial implementation of MAE using pytorch-lightning

Dashboard for the COVID19 spread

KoCLIP: Korean port of OpenAI CLIP, in Flax

The official implementation of NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation [ICLR-2021]. https://arxiv.org/pdf/2101.12378.pdf

Leveraging Two Types of Global Graph for Sequential Fashion Recommendation, ICMR 2021

Official implementation of Sparse Transformer-based Action Recognition

SSD: A Unified Framework for Self-Supervised Outlier Detection [ICLR 2021]

Code for paper PairRE: Knowledge Graph Embeddings via Paired Relation Vectors.

Retrieve and analysis data from SDSS (Sloan Digital Sky Survey)

Experimental Python implementation of OpenVINO Inference Engine (very slow, limited functionality). All codes are written in Python. Easy to read and modify.

A web application that provides real time temperature and humidity readings of a house.

Educational API for 3D Vision using pose to control carton.

Code for IntraQ, PyTorch implementation of our paper under review

Learning with Noisy Labels via Sparse Regularization, ICCV2021

CLOOB training (JAX) and inference (JAX and PyTorch)

本步态识别系统主要基于GaitSet模型进行实现