bitsandbytes
Bitsandbytes is a lightweight wrapper around CUDA custom functions, in particular 8-bit optimizers and quantization functions.
TL;DR
Installation:
- Note down version:
conda list | grep cudatoolkit - Replace 111 with the version that you see:
pip install bitsandbytes-cuda111
Usage:
- Comment out optimizer:
#torch.optim.Adam(....) - Add 8-bit optimizer of your choice
bnb.optim.Adam8bit(....)(arguments stay the same) - Replace embedding layer if necessary:
torch.nn.Embedding(..) -> bnb.nn.Embedding(..)
Features
- 8-bit Optimizers: Adam, AdamW, RMSProp, LARS, LAMB (saves 75% memory)
- Stable Embedding Layer: Improved stability through better initialization, and normalization
- 8-bit quantization: Quantile, Linear, and Dynamic quantization
- Fast quantile estimation: Up to 100x faster than other algorithms
Requirements & Installation
Requirements: anaconda, cudatoolkit, pytorch Hardware requirements: NVIDIA Maxwell GPU or newer (>=GTX 9XX) Supported CUDA versions: 9.2 - 11.3
The requirements can best be fulfilled by installing pytorch via anaconda. You can install PyTorch by following the "Get Started" instructions on the official website.
bitsandbytes is compatible with all major PyTorch releases and cudatoolkit versions, but for now, you need to select the right version manually. To do this run:
conda list | grep cudatoolkit
and take note of the Cuda version that you have installed. Then you can install bitsandbytes via:
# choices: {cuda92, cuda 100, cuda101, cuda102, cuda110, cuda111, cuda113}
# replace XXX with the respective number
pip install bitsandbytes-cudaXXX
To check if your installation was successful, you can execute the following command, which runs a single bnb Adam update.
wget https://gist.githubusercontent.com/TimDettmers/1f5188c6ee6ed69d211b7fe4e381e713/raw/4d17c3d09ccdb57e9ab7eca0171f2ace6e4d2858/check_bnb_install.py && python check_bnb_install.py
Using bitsandbytes
Using the 8-bit Optimizers
With bitsandbytes 8-bit optimizers can be used by changing a single line of code in your codebase. For NLP models we recommend also to use the StableEmbedding layers (see below) which improves results and helps with stable 8-bit optimization. To get started with 8-bit optimizers, it is sufficient to replace your old optimizer with the 8-bit optimizer in the following way:
import bitsandbytes as bnb
# adam = torch.optim.Adam(model.parameters(), lr=0.001, betas=(0.9, 0.995)) # comment out old optimizer
adam = bnb.optim.Adam8bit(model.parameters(), lr=0.001, betas=(0.9, 0.995)) # add bnb optimizer
adam = bnb.optim.Adam(model.parameters(), lr=0.001, betas=(0.9, 0.995), optim_bits=8) # equivalent
torch.nn.Embedding(...) -> bnb.nn.StableEmbedding(...) # recommended for NLP models
Note that by default all parameter tensors with less than 4096 elements are kept at 32-bit even if you initialize those parameters with 8-bit optimizers. This is done since such small tensors do not save much memory and often contain highly variable parameters (biases) or parameters that require high precision (batch norm, layer norm). You can change this behavior like so:
# parameter tensors with less than 16384 values are optimized in 32-bit
# it is recommended to use multiplies of 4096
adam = bnb.optim.Adam8bit(model.parameters(), min_8bit_size=16384)
Change Bits and other Hyperparameters for Individual Parameters
If you want to optimize some unstable parameters with 32-bit Adam and others with 8-bit Adam, you can use the GlobalOptimManager. With this, we can also configure specific hyperparameters for particular layers, such as embedding layers. To do that, we need two things: (1) register the parameter while they are still on the CPU, (2) override the config with the new desired hyperparameters (anytime, anywhere). See our guide for more details
Fairseq Users
To use the Stable Embedding Layer, override the respective build_embedding(...) function of your model. Make sure to also use the --no-scale-embedding flag to disable scaling of the word embedding layer (nor replaced with layer norm). You can use the optimizers by replacing the optimizer in the respective file (adam.py etc.).
Release and Feature History
For upcoming features and changes and full history see Patch Notes.
Errors
- RuntimeError: CUDA error: no kernel image is available for execution on the device. Solution
License
The majority of bitsandbytes is licensed under MIT, however portions of the project are available under separate license terms: Pytorch is licensed under the BSD license.
We thank Fabio Cannizzo for his work on FastBinarySearch which we use for CPU quantization.
Citation
If you found this library and 8-bit optimizers or quantization routines useful, please consider citing out work.
@misc{dettmers2021optim8bit,
title={8-bit Optimizers via Block-wise Quantization},
author={Tim Dettmers and Mike Lewis and Sam Shleifer and Luke Zettlemoyer},
year={2021},
eprint={2110.02861},
archivePrefix={arXiv},
primaryClass={cs.LG}
}
![[Question] Usage of bnb.nn.Embedding with existing classes from other libraries](https://avatars.githubusercontent.com/u/12072891?v=4)
16 Nov 03, 2022
1 Nov 27, 2021
13 Jun 19, 2022
44 Nov 21, 2022
117 Jan 07, 2023
676 Jan 03, 2023
149 Dec 19, 2022
367 Dec 24, 2022
11 Dec 09, 2022
33 Dec 30, 2022
2.4k Jan 04, 2023
98 Dec 24, 2022
536 Dec 20, 2022
8 Aug 02, 2022
33 Dec 06, 2022
1k Dec 30, 2022
23 Dec 26, 2022
27 Dec 13, 2022
0 Aug 22, 2021
3 Oct 31, 2022