• Stars
    star
    2,754
  • Rank 16,461 (Top 0.4 %)
  • Language
    Python
  • License
    Apache License 2.0
  • Created about 2 years ago
  • Updated about 1 month ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Simple, safe way to store and distribute tensors

Hugging Face Safetensors Library

Python Pypi Documentation Codecov Downloads

Rust Crates.io Documentation Codecov Dependency status

safetensors

Safetensors

This repository implements a new simple format for storing tensors safely (as opposed to pickle) and that is still fast (zero-copy).

Installation

Pip

You can install safetensors via the pip manager:

pip install safetensors

From source

For the sources, you need Rust

# Install Rust
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
# Make sure it's up to date and using stable channel
rustup update
git clone https://github.com/huggingface/safetensors
cd safetensors/bindings/python
pip install setuptools_rust
pip install -e .

Getting started

import torch
from safetensors import safe_open
from safetensors.torch import save_file

tensors = {
   "weight1": torch.zeros((1024, 1024)),
   "weight2": torch.zeros((1024, 1024))
}
save_file(tensors, "model.safetensors")

tensors = {}
with safe_open("model.safetensors", framework="pt", device="cpu") as f:
   for key in f.keys():
       tensors[key] = f.get_tensor(key)

Python documentation

Format

  • 8 bytes: N, a u64 int, containing the size of the header
  • N bytes: a JSON utf-8 string representing the header.
    • The header is a dict like {"TENSOR_NAME": {"dtype": "F16", "shape": [1, 16, 256], "offsets": [BEGIN, END]}, "NEXT_TENSOR_NAME": {...}, ...}, where offsets point to the tensor data relative to the beginning of the byte buffer, with BEGIN as the starting offset and END as the one-past offset (so total tensor byte size = END - BEGIN).
    • A special key __metadata__ is allowed to contain free form text-to-text map.
  • Rest of the file: byte-buffer.

Notes:

  • Duplicate keys are disallowed. Not all parsers may respect this.
  • In general the subset of JSON is implicitly decided by serde_json for this library. Anything obscure might be modified at a later time, that odd ways to represent integer, newlines and escapes in utf-8 strings. This would only be done for safety concerns
  • Tensor values are not checked against, in particular NaN and +/-Inf could be in the file
  • Empty tensors (tensors with 1 dimension being 0) are allowed. They are not storing any data in the databuffer, yet retaining size in the header. They don't really bring a lot of values but are accepted since they are valid tensors from traditional tensor libraries perspective (torch, tensorflow, numpy, ..).
  • 0-rank Tensors (tensors with shape []) are allowed, they are merely a scalar.
  • The byte buffer needs to be entirely indexed, and cannot contain holes. This prevents the creation of polyglot files.

Yet another format ?

The main rationale for this crate is to remove the need to use pickle on PyTorch which is used by default. There are other formats out there used by machine learning and more general formats.

Let's take a look at alternatives and why this format is deemed interesting. This is my very personal and probably biased view:

Format Safe Zero-copy Lazy loading No file size limit Layout control Flexibility Bfloat16
pickle (PyTorch) 🗸 🗸 🗸
H5 (Tensorflow) 🗸 🗸 🗸 ~ ~
SavedModel (Tensorflow) 🗸 🗸 🗸 🗸
MsgPack (flax) 🗸 🗸 🗸 🗸
Protobuf (ONNX) 🗸 🗸
Cap'n'Proto 🗸 🗸 ~ 🗸 🗸 ~
Arrow ? ? ? ? ? ?
Numpy (npy,npz) 🗸 ? ? 🗸
pdparams (Paddle) 🗸 🗸 🗸
SafeTensors 🗸 🗸 🗸 🗸 🗸 🗸
  • Safe: Can I use a file randomly downloaded and expect not to run arbitrary code ?
  • Zero-copy: Does reading the file require more memory than the original file ?
  • Lazy loading: Can I inspect the file without loading everything ? And loading only some tensors in it without scanning the whole file (distributed setting) ?
  • Layout control: Lazy loading, is not necessarily enough since if the information about tensors is spread out in your file, then even if the information is lazily accessible you might have to access most of your file to read the available tensors (incurring many DISK -> RAM copies). Controlling the layout to keep fast access to single tensors is important.
  • No file size limit: Is there a limit to the file size ?
  • Flexibility: Can I save custom code in the format and be able to use it later with zero extra code ? (~ means we can store more than pure tensors, but no custom code)
  • Bfloat16: Does the format support native bfloat16 (meaning no weird workarounds are necessary)? This is becoming increasingly important in the ML world.

Main oppositions

  • Pickle: Unsafe, runs arbitrary code
  • H5: Apparently now discouraged for TF/Keras. Seems like a great fit otherwise actually. Some classic use after free issues: https://www.cvedetails.com/vulnerability-list/vendor_id-15991/product_id-35054/Hdfgroup-Hdf5.html. On a very different level than pickle security-wise. Also 210k lines of code vs ~400 lines for this lib currently.
  • SavedModel: Tensorflow specific (it contains TF graph information).
  • MsgPack: No layout control to enable lazy loading (important for loading specific parts in distributed setting)
  • Protobuf: Hard 2Go max file size limit
  • Cap'n'proto: Float16 support is not present link so using a manual wrapper over a byte-buffer would be necessary. Layout control seems possible but not trivial as buffers have limitations link.
  • Numpy (npz): No bfloat16 support. Vulnerable to zip bombs (DOS). Not zero-copy.
  • Arrow: No bfloat16 support. Seem to require decoding link

Notes

  • Zero-copy: No format is really zero-copy in ML, it needs to go from disk to RAM/GPU RAM (that takes time). On CPU, if the file is already in cache, then it can truly be zero-copy, whereas on GPU there is not such disk cache, so a copy is always required but you can bypass allocating all the tensors on CPU at any given point. SafeTensors is not zero-copy for the header. The choice of JSON is pretty arbitrary, but since deserialization is <<< of the time required to load the actual tensor data and is readable I went that way, (also space is <<< to the tensor data).

  • Endianness: Little-endian. This can be modified later, but it feels really unnecessary at the moment.

  • Order: 'C' or row-major. This seems to have won. We can add that information later if needed.

  • Stride: No striding, all tensors need to be packed before being serialized. I have yet to see a case where it seems useful to have a strided tensor stored in serialized format.

Benefits

Since we can invent a new format we can propose additional benefits:

  • Prevent DOS attacks: We can craft the format in such a way that it's almost impossible to use malicious files to DOS attack a user. Currently, there's a limit on the size of the header of 100MB to prevent parsing extremely large JSON. Also when reading the file, there's a guarantee that addresses in the file do not overlap in any way, meaning when you're loading a file you should never exceed the size of the file in memory

  • Faster load: PyTorch seems to be the fastest file to load out in the major ML formats. However, it does seem to have an extra copy on CPU, which we can bypass in this lib by using torch.UntypedStorage.from_file. Currently, CPU loading times are extremely fast with this lib compared to pickle. GPU loading times are as fast or faster than PyTorch equivalent. Loading first on CPU with memmapping with torch, and then moving all tensors to GPU seems to be faster too somehow (similar behavior in torch pickle)

  • Lazy loading: in distributed (multi-node or multi-gpu) settings, it's nice to be able to load only part of the tensors on the various models. For BLOOM using this format enabled to load the model on 8 GPUs from 10mn with regular PyTorch weights down to 45s. This really speeds up feedbacks loops when developing on the model. For instance you don't have to have separate copies of the weights when changing the distribution strategy (for instance Pipeline Parallelism vs Tensor Parallelism).

License: Apache-2.0

More Repositories

1

transformers

🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
Python
132,772
star
2

pytorch-image-models

PyTorch image models, scripts, pretrained weights -- ResNet, ResNeXT, EfficientNet, NFNet, Vision Transformer (ViT), MobileNet-V3/V2, RegNet, DPN, CSPNet, Swin Transformer, MaxViT, CoAtNet, ConvNeXt, and more
Python
28,073
star
3

diffusers

🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch and FLAX.
Python
25,010
star
4

datasets

🤗 The largest hub of ready-to-use datasets for ML models with fast, easy-to-use and efficient data manipulation tools
Python
17,530
star
5

peft

🤗 PEFT: State-of-the-art Parameter-Efficient Fine-Tuning.
Python
15,663
star
6

candle

Minimalist ML framework for Rust
Rust
15,011
star
7

trl

Train transformer language models with reinforcement learning.
Python
9,278
star
8

tokenizers

💥 Fast State-of-the-Art Tokenizers optimized for Research and Production
Rust
8,885
star
9

text-generation-inference

Large Language Model Text Generation Inference
Python
8,663
star
10

accelerate

🚀 A simple way to launch, train, and use PyTorch models on almost any device and distributed configuration, automatic mixed precision (including fp8), and easy-to-configure FSDP and DeepSpeed support
Python
7,759
star
11

chat-ui

Open source codebase powering the HuggingChat app
TypeScript
7,113
star
12

lerobot

🤗 LeRobot: Making AI for Robotics more accessible with end-to-end learning
Python
6,522
star
13

alignment-handbook

Robust recipes to align language models with human and AI preferences
Python
4,474
star
14

parler-tts

Inference and training library for high-quality TTS models.
Python
4,027
star
15

deep-rl-class

This repo contains the syllabus of the Hugging Face Deep Reinforcement Learning Course.
MDX
3,680
star
16

autotrain-advanced

🤗 AutoTrain Advanced
Python
3,671
star
17

diffusion-models-class

Materials for the Hugging Face Diffusion Models Course
Jupyter Notebook
3,508
star
18

notebooks

Notebooks using the Hugging Face libraries 🤗
Jupyter Notebook
3,492
star
19

distil-whisper

Distilled variant of Whisper for speech recognition. 6x faster, 50% smaller, within 1% word error rate.
Python
3,455
star
20

neuralcoref

✨Fast Coreference Resolution in spaCy with Neural Networks
C
2,842
star
21

knockknock

🚪✊Knock Knock: Get notified when your training ends with only two additional lines of code
Python
2,682
star
22

text-embeddings-inference

A blazing fast inference solution for text embeddings models
Rust
2,596
star
23

speech-to-speech

Speech To Speech: an effort for an open-sourced and modular GPT4-o
Python
2,540
star
24

swift-coreml-diffusers

Swift app demonstrating Core ML Stable Diffusion
Swift
2,506
star
25

optimum

🚀 Accelerate training and inference of 🤗 Transformers and 🤗 Diffusers with easy to use hardware optimization tools
Python
2,469
star
26

setfit

Efficient few-shot learning with Sentence Transformers
Jupyter Notebook
2,142
star
27

blog

Public repo for HF blog posts
Jupyter Notebook
2,136
star
28

course

The Hugging Face course on Transformers
MDX
2,005
star
29

awesome-papers

Papers & presentation materials from Hugging Face's internal science day
1,996
star
30

datatrove

Freeing data processing from scripting madness by providing a set of platform-agnostic customizable pipeline processing blocks.
Python
1,909
star
31

evaluate

🤗 Evaluate: A library for easily evaluating machine learning models and datasets.
Python
1,825
star
32

transfer-learning-conv-ai

🦄 State-of-the-Art Conversational AI with Transfer Learning
Python
1,654
star
33

cookbook

Open-source AI cookbook
Jupyter Notebook
1,577
star
34

swift-coreml-transformers

Swift Core ML 3 implementations of GPT-2, DistilGPT-2, BERT, and DistilBERT for Question answering. Other Transformers coming soon!
Swift
1,543
star
35

pytorch-openai-transformer-lm

🐥A PyTorch implementation of OpenAI's finetuned transformer language model with a script to import the weights pre-trained by OpenAI
Python
1,464
star
36

huggingface.js

Utilities to use the Hugging Face Hub API
TypeScript
1,368
star
37

huggingface_hub

All the open source things related to the Hugging Face Hub.
Python
1,311
star
38

gsplat.js

JavaScript Gaussian Splatting library.
TypeScript
1,302
star
39

Mongoku

🔥The Web-scale GUI for MongoDB
TypeScript
1,300
star
40

llm-vscode

LLM powered development for VSCode
TypeScript
1,206
star
41

hmtl

🌊HMTL: Hierarchical Multi-Task Learning - A State-of-the-Art neural network model for several NLP tasks based on PyTorch and AllenNLP
Python
1,185
star
42

nanotron

Minimalistic large language model 3D-parallelism training
Python
1,071
star
43

pytorch-pretrained-BigGAN

🦋A PyTorch implementation of BigGAN with pretrained weights and conversion scripts.
Python
986
star
44

torchMoji

😇A pyTorch implementation of the DeepMoji model: state-of-the-art deep learning model for analyzing sentiment, emotion, sarcasm etc
Python
880
star
45

optimum-nvidia

Python
863
star
46

awesome-huggingface

🤗 A list of wonderful open-source projects & applications integrated with Hugging Face libraries.
853
star
47

optimum-quanto

A pytorch quantization backend for optimum
Python
738
star
48

llm.nvim

LLM powered development for Neovim
Lua
728
star
49

naacl_transfer_learning_tutorial

Repository of code for the tutorial on Transfer Learning in NLP held at NAACL 2019 in Minneapolis, MN, USA
Python
718
star
50

dataset-viewer

Backend that powers the dataset viewer on Hugging Face dataset pages through a public API.
Python
689
star
51

swift-transformers

Swift Package to implement a transformers-like API in Swift
Swift
647
star
52

exporters

Export Hugging Face models to Core ML and TensorFlow Lite
Python
587
star
53

llm-ls

LSP server leveraging LLMs for code completion (and more?)
Rust
586
star
54

ratchet

A cross-platform browser ML framework.
Rust
574
star
55

transformers-bloom-inference

Fast Inference Solutions for BLOOM
Python
557
star
56

lighteval

LightEval is a lightweight LLM evaluation suite that Hugging Face has been using internally with the recently released LLM data processing library datatrove and LLM training library nanotron.
Python
554
star
57

pytorch_block_sparse

Fast Block Sparse Matrices for Pytorch
C++
523
star
58

node-question-answering

Fast and production-ready question answering in Node.js
TypeScript
459
star
59

large_language_model_training_playbook

An open collection of implementation tips, tricks and resources for training large language models
Python
452
star
60

swift-chat

Mac app to demonstrate swift-transformers
Swift
444
star
61

llm_training_handbook

An open collection of methodologies to help with successful training of large language models.
Python
437
star
62

text-clustering

Easily embed, cluster and semantically label text datasets
Python
422
star
63

cosmopedia

Python
416
star
64

optimum-intel

🤗 Optimum Intel: Accelerate inference with Intel optimization tools
Jupyter Notebook
379
star
65

tflite-android-transformers

DistilBERT / GPT-2 for on-device inference thanks to TensorFlow Lite with Android demo apps
Java
368
star
66

community-events

Place where folks can contribute to 🤗 community events
Jupyter Notebook
368
star
67

controlnet_aux

Python
365
star
68

nn_pruning

Prune a model while finetuning or training.
Jupyter Notebook
360
star
69

speechbox

Python
341
star
70

100-times-faster-nlp

🚀100 Times Faster Natural Language Processing in Python - iPython notebook
HTML
325
star
71

education-toolkit

Educational materials for universities
Jupyter Notebook
324
star
72

transformers.js-examples

A collection of 🤗 Transformers.js demos and example applications
JavaScript
323
star
73

open-muse

Open reproduction of MUSE for fast text2image generation.
Python
320
star
74

local-gemma

Gemma 2 optimized for your local machine.
Python
317
star
75

unity-api

C#
313
star
76

audio-transformers-course

The Hugging Face Course on Transformers for Audio
MDX
308
star
77

datablations

Scaling Data-Constrained Language Models
Jupyter Notebook
305
star
78

hf_transfer

Rust
287
star
79

dataspeech

Python
262
star
80

huggingface-llama-recipes

Jupyter Notebook
259
star
81

diarizers

Python
238
star
82

optimum-benchmark

🏋️ A unified multi-backend utility for benchmarking Transformers, Timm, PEFT, Diffusers and Sentence-Transformers with full support of Optimum's hardware optimizations & quantization schemes.
Python
226
star
83

hub-docs

Docs of the Hugging Face Hub
221
star
84

llm-swarm

Manage scalable open LLM inference endpoints in Slurm clusters
Python
216
star
85

sam2-studio

Swift
196
star
86

optimum-neuron

Easy, fast and very cheap training and inference on AWS Trainium and Inferentia chips.
Jupyter Notebook
193
star
87

data-is-better-together

Let's build better datasets, together!
Jupyter Notebook
192
star
88

instruction-tuned-sd

Code for instruction-tuning Stable Diffusion.
Python
189
star
89

simulate

🎢 Creating and sharing simulation environments for embodied and synthetic data research
Python
185
star
90

OBELICS

Code used for the creation of OBELICS, an open, massive and curated collection of interleaved image-text web documents, containing 141M documents, 115B text tokens and 353M images.
Python
184
star
91

diffusion-fast

Faster generation with text-to-image diffusion models.
Python
179
star
92

olm-datasets

Pipeline for pulling and processing online language model pretraining data from the web
Python
170
star
93

api-inference-community

Python
161
star
94

jat

General multi-task deep RL Agent
Python
154
star
95

workshops

Materials for workshops on the Hugging Face ecosystem
Jupyter Notebook
148
star
96

coreml-examples

Swift Core ML Examples
Jupyter Notebook
147
star
97

chug

Minimal sharded dataset loaders, decoders, and utils for multi-modal document, image, and text datasets.
Python
140
star
98

optimum-habana

Easy and lightning fast training of 🤗 Transformers on Habana Gaudi processor (HPU)
Python
140
star
99

sharp-transformers

A Unity plugin for using Transformers models in Unity.
C#
139
star
100

hf-hub

Rust client for the huggingface hub aiming for minimal subset of features over `huggingface-hub` python package
Rust
132
star