Run Effective Large Batch Contrastive Learning on Limited Memory GPU

Overview

Gradient Cache

Gradient Cache is a simple technique for unlimitedly scaling contrastive learning batch far beyond GPU memory constraint. This means training that used to take heavy hardware, e.g. 8 V100 GPU, can be done on a single GPU. In addition, Gradient Cache allow users to replace big RAM GPU with much more cost efficient high FLOP low RAM cards.

This repo holds a generic Pytorch implementation of Gradient Cache described in our paper Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup .

@inproceedings{gao2021scaling,
     title={Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup},
     author={Luyu Gao, Yunyi Zhang, Jiawei Han, Jamie Callan},
     booktitle ={Proceedings of the 6th Workshop on Representation Learning for NLP},
     year={2021},
}

Gradient Cache has also been integrated into dense passage retrieval (DPR). Checkout our GC-DPR toolkit.

Installation

The package depends only on pytorch>=1.6. To install, clone this repo and run pip.

git clone https://github.com/luyug/GradCache
cd GradCache
pip install .

For development,

pip install --editable .

Usage

Gradient caching functionalities are implemented in GradCache class. If you are developing a new project instead of patching an old one, also checkout our functional approach for a effort reduced approach.

Initialization

The class's __init__ method defines the cache and has several functional parameters *_fn for easy adjust of model behaviors. Alternatively you can also sub-class GradCache.

grad_cache.GradCache(  
  models: List[nn.Module],  
  chunk_sizes: Union[int, List[int]],  
  loss_fn: Callable[..., Tensor],  
  split_input_fn: Callable[[Any, int], Any] = None,  
  get_rep_fn: Callable[..., Tensor] = None,  
  fp16: bool = False,  
  scaler: GradScaler = None,  
)

models - A list of encoder models to be updated with with the Gradient Cache.

chunk_sizes - An integer indicating chunk size. Or a list of integers of chunk size for each model. This controls for each model the sub-batch size to run forward-backward pass and should be set based on available GPU memory. A value too small will leave the GPU under utilized.

loss_fn - A loss function that takes representation tensors of number equal to number of models in models and arbitrary numbers of keyword arguments. It should compute loss based on the input tensors, and in no case modify the input tensors' relations in the autograd graph, which are later relied upon to create the gradient cache.

split_input_fn - An optional function that split generic model input into chunks based on defined chunk_sizes. If not provided, this class will try its best to split the inputs of supported types. See split_inputs function.

get_rep_fn - An optional function that takes generic model output and return representation tensors. If not provided, the generic output is assumed to be the representation tensor.

fp16 - If True, run mixed precision training, which requires scaler to also be set.

scaler - A GradScaler object for automatic mixed precision training.

Cache Gradient Step

To run a cached gradient computatoin step, call cache_step function,

cache_step(  
  *model_inputs,  
  no_sync_except_last: bool = False,  
  **loss_kwargs  
)

Run a single gradient cache step. Upon function return, updates are computed for each model in self.models with gradient populated on the weights, as if the model_inputs are run as a huge single batch on sufficiently large hardware. Calling an GradCache object with __call__ will also invoke this function.

model_inputs - List of inputs to each encoder model. Should be in similar order as self.models.

no_sync_except_last - If True, under distributed setup, for each model, only trigger gradient reduction across processes for the last sub-batch's forward-backward pass. This could come in handy when dealing with a) large model, and/or b) non trivial number of sub-batches.

loss_kwargs - Additional keyword arguments to the loss function loss_fn. This is intended to enable flexible loss computation (thanks to dynamic graph in Pytorch) such as reduction, weighting, etc. Potentially, using loss_kwargs you can incorporate outputs from those encoder models not tracked by the cache.

Return - loss, the current steps loss scaler tensor (detached from the graph).

Natively Supported Input Types

  • x: Tensor - will be passed in as model(x)
  • x: List[Tensor] - will be passed in as model(*x)
  • x: Dict[str, Tensor] (or UserDict[str, Tensor]) - will be passed in as model(**x)
  • x: Tuple[List[Tensor], Dict[str, Tensor]] - will be passed in as model(*x[0], **x[1])

Other generic input are not fully supported, we perform model call using the following heuristics,

  • x: List[Any] - will be passed in as model(*x)
  • x: Dict[str, Any] - will be passed in as model(**x)
  • x: Tuple[List[Any], Dict[str, Any]] - will be passed in as model(*x[0], **x[1])

To run with them, split_input_fn should be specified during cache initialization to break these inputs into smaller batches. In some rare cases, you may also need to override get_input_tensors when its heuristic can not grab enough tensors that covers all cuda devices that hold some tensors in the input.

Example Usage with Huggingface Transformers

Learning a Bi-encoder

Say we want to learn a embedding space of labels and text. Consider the following four pairs. (In practice, you will have many more and much longer text entries.)

labels = ['fruit', 'meat', 'school', 'company']
texts = [
  'this is an apple', 
  'steak should be cooked medium rare', 
  'cmu is pittsburgh', 
  'apple sells laptop'
]

Initialize our encoder models,

from transformers import AutoTokenizer, TFAutoModel
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
encoder1 = AutoModel.from_pretrained("bert-base-uncased").cuda()
encoder2 = AutoModel.from_pretrained("bert-base-uncased").cuda()

Initialize the GradCache object,

from grad_cache import GradCache
from grad_cache.loss import SimpleContrastiveLoss

loss_fn = SimpleContrastiveLoss()
gc = GradCache(
  models=[encoder1, encoder2], 
  chunk_sizes=2, 
  loss_fn=loss_fn, 
  get_rep_fn=lambda v: v.pooler_output
)

Here we use the get_rep_fn argument to specify a function that takes generic Huggingface model output and return the actual representation tensor.

Create model input,

xx = tokenizer(tt, return_tensors='pt', padding=True)
yy = tokenizer(tt2, return_tensors='pt', padding=True)

Run a cache step,

gc(xx, yy, reduction='mean')

Here we use reduction='mean' as a loss_kwargs to control loss behavior. With a defined optimizer, the full gradient update can be done as,

optimizer.zero_grad()
gc(xx, yy, reduction='mean')
optimizer.step()

Use Tied Encoder?

This is naturally handled by the (magic of) dynamic graph. You pass shallow copies of the same encoder model to the GradCache init method.

tied_encoder = AutoModel.from_pretrained("bert-base-uncased").cuda()
gc = GradCache(
  models=[tied_encoder , tied_encoder], 
  chunk_sizes=2, 
  loss_fn=loss_fn, 
  get_rep_fn=lambda v: v.pooler_output
)

Under the hood, distinct hooks will be registered to make correct gradient computation.

Distributed Training with Multiple GPUs?

We expect cross process communication of representations to be handled by the loss_fn.

from grad_cache.loss import DistributedContrastiveLoss
loss_fn_dist = DistributedContrastiveLoss()

Properly wrap the the encoder models for gradient reduction,

encoder1_ddp = DistributedDataParallel(
	encoder1, device_ids=[local_rank], output_device=local_rank, find_unused_parameters=True)
encoder2_ddp = DistributedDataParallel(
	encoder2, device_ids=[local_rank], output_device=local_rank, find_unused_parameters=True)

You can initialize the cache use the distributed loss and the DDP models,

gc = GradCache(
  models=[encoder1_ddp, encoder2_ddp], 
  chunk_sizes=2, 
  loss_fn=loss_fn_dist, 
  get_rep_fn=lambda v: v.pooler_output
)

Run a cache step,

gc(xx, yy, no_sync_except_last=True, reduction='mean')

Set no_sync_except_last=True to avoid unnecessary gradient reduction.

Functional Approach

Decorators

If you are developing a new project, we recommend also checking out the decorators we have provided to create higher order functions for cache.

grad_cache.functional.cached(func: Callable[..., Tensor])

A decorator that takes a model call function into a cached compatible version.

func - A function that calls the model and return representation tensor.

Return - A function that returns 1) representation leaf tensors for cache construction, 2) a closure function for the 2nd forward and the cached backward. Call 2) with 1) as argument after calling backward on the loss Tensor.

grad_cache.functional.cat_input_tensor(func: Callable[..., Tensor])

A decorator that concatenates positional and keyword arguments of type List[Tensor] into a single Tensor on the 0th dimension. This can come in handy dealing with results of representation tensors from multiple cached forward.

func - A loss function

Return - Decorated loss function for cached results.

Usage

The functional decorators are particular useful if your data loader is emitting small batches, from which you can construct the big batch. Say you also want to do automatic mixed precision, we first define the model call function and loss function,

from grad_cache.functional import cached, cat_input_tensor

import torch
import torch.nn.functional as F
from torch.cuda.amp import autocast

@cached
@autocast()
def  call_model(model, input):
	return model(**input).pooler_output

@cat_input_tensor
@autocast()
def  contrastive_loss(x, y):
	target = torch.arange(0, y.size(0), int(y.size(0) / x.size(0)), device=x.device)
	scores = torch.matmul(x, y.transpose(0, 1))
	return F.cross_entropy(scores, target=target)

Say you have a DataLoader loader emitting small batches of tuple (xx, yy) of size (M * N) and that you want to train by aggregating 16 small batches to get a batch of (16M * 16N),

cache_x = []
cache_y = []
closures_x = []
closures_y = []

for step, sub_batch in enumerate(loader):  
    xx, yy = sub_batch
    rx, cx = call_model(bert, xx)
    ry, cy = call_model(bert, yy)
    
    cache_x.append(rx)
    cache_y.append(ry)
    closuresx.append(cx)
    closuresy.append(cy)
    
    if (step + 1) % 16 == 0:
        loss = contrastive_loss(cache_x, cache_y)
        scaler.scale(loss).backward()
        
	for f, r in zip(closuresx, cache_x):
            f(r)
        for f, r in zip(closuresy, cache_y):
            f(r)

        cache_x = []
        cache_y = []
        closures_x = []
        closures_y = []
	
        scaler.step(optimizer)
        scaler.update()
        optimizer.zero_grad()

Code Structure

grad_cache/grad_cache.py - Define the GradCache class. The code is under 300 lines including comments. For development, we encourage you to read through it.

grad_cache/functional.py - Define decorators to create higher order function for gradient caching from ordinary model call functions and loss functions.

Owner
Luyu Gao
NLP Research [email protected], CMU
Luyu Gao
A collection of metrics for evaluating timbre dissimilarity using the TorchMetrics API

Timbre Dissimilarity Metrics A collection of metrics for evaluating timbre dissimilarity using the TorchMetrics API Installation pip install -e . Usag

Ben Hayes 21 Jan 05, 2022
ScaleNet: A Shallow Architecture for Scale Estimation

ScaleNet: A Shallow Architecture for Scale Estimation Repository for the code of ScaleNet paper: "ScaleNet: A Shallow Architecture for Scale Estimatio

Axel Barroso 34 Nov 09, 2022
"Structure-Augmented Text Representation Learning for Efficient Knowledge Graph Completion"(WWW 2021)

STAR_KGC This repo contains the source code of the paper accepted by WWW'2021. "Structure-Augmented Text Representation Learning for Efficient Knowled

Bo Wang 60 Dec 26, 2022
The dynamics of representation learning in shallow, non-linear autoencoders

The dynamics of representation learning in shallow, non-linear autoencoders The package is written in python and uses the pytorch implementation to ML

Maria Refinetti 4 Jun 08, 2022
This is a clean and robust Pytorch implementation of DQN and Double DQN.

DQN/DDQN-Pytorch This is a clean and robust Pytorch implementation of DQN and Double DQN. Here is the training curve: All the experiments are trained

XinJingHao 15 Dec 27, 2022
Official implementation of the ICCV 2021 paper "Joint Inductive and Transductive Learning for Video Object Segmentation"

JOINT This is the official implementation of Joint Inductive and Transductive learning for Video Object Segmentation, to appear in ICCV 2021. @inproce

Yunyao 35 Oct 16, 2022
NeROIC: Neural Object Capture and Rendering from Online Image Collections

NeROIC: Neural Object Capture and Rendering from Online Image Collections This repository is for the source code for the paper NeROIC: Neural Object C

Snap Research 647 Dec 27, 2022
Band-Adaptive Spectral-Spatial Feature Learning Neural Network for Hyperspectral Image Classification

Band-Adaptive Spectral-Spatial Feature Learning Neural Network for Hyperspectral Image Classification

258 Dec 29, 2022
Code for Max-Margin Contrastive Learning - AAAI 2022

Max-Margin Contrastive Learning This is a pytorch implementation for the paper Max-Margin Contrastive Learning accepted to AAAI 2022. This repository

Anshul Shah 12 Oct 22, 2022
2021 credit card consuming recommendation

2021 credit card consuming recommendation

Wang, Chung-Che 7 Mar 08, 2022
Code for DeepCurrents: Learning Implicit Representations of Shapes with Boundaries

DeepCurrents | Webpage | Paper DeepCurrents: Learning Implicit Representations of Shapes with Boundaries David Palmer*, Dmitriy Smirnov*, Stephanie Wa

Dima Smirnov 36 Dec 08, 2022
Perturbed Self-Distillation: Weakly Supervised Large-Scale Point Cloud Semantic Segmentation (ICCV2021)

Perturbed Self-Distillation: Weakly Supervised Large-Scale Point Cloud Semantic Segmentation (ICCV2021) This is the implementation of PSD (ICCV 2021),

12 Dec 12, 2022
Code release for NeuS

NeuS We present a novel neural surface reconstruction method, called NeuS, for reconstructing objects and scenes with high fidelity from 2D image inpu

Peng Wang 813 Jan 04, 2023
Deep deconfounded recommender (Deep-Deconf) for paper "Deep causal reasoning for recommendations"

Deep Causal Reasoning for Recommender Systems The codes are associated with the following paper: Deep Causal Reasoning for Recommendations, Yaochen Zh

Yaochen Zhu 22 Oct 15, 2022
Complementary Patch for Weakly Supervised Semantic Segmentation, ICCV21 (poster)

CPN (ICCV2021) This is an implementation of Complementary Patch for Weakly Supervised Semantic Segmentation, which is accepted by ICCV2021 poster. Thi

Ferenas 20 Dec 12, 2022
DeepMReye: magnetic resonance-based eye tracking using deep neural networks

DeepMReye: magnetic resonance-based eye tracking using deep neural networks

73 Dec 21, 2022
Code for Parameter Prediction for Unseen Deep Architectures (NeurIPS 2021)

Parameter Prediction for Unseen Deep Architectures (NeurIPS 2021) authors: Boris Knyazev, Michal Drozdzal, Graham Taylor, Adriana Romero-Soriano Overv

Facebook Research 462 Jan 03, 2023
Publication describing 3 ML examples at NSLS-II and interfacing into Bluesky

Machine learning enabling high-throughput and remote operations at large-scale user facilities. Overview This repository contains the source code and

BNL 4 Sep 24, 2022
Code for the paper: Fighting Fake News: Image Splice Detection via Learned Self-Consistency

Fighting Fake News: Image Splice Detection via Learned Self-Consistency [paper] [website] Minyoung Huh *12, Andrew Liu *1, Andrew Owens1, Alexei A. Ef

minyoung huh (jacob) 174 Dec 09, 2022
Plover-tapey-tape: an alternative to Plover’s built-in paper tape

plover-tapey-tape plover-tapey-tape is an alternative to Plover’s built-in paper

7 May 29, 2022