NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI · #5277 most downloaded on PyPI
A PyTorch library of transformer models and components
Last release 2 years ago
no release in 18 months
Release timing varies
gaps range from 9 days to 4 months
Most releases are documented
notes for 15 of 20 stable releases
1 version withdrawn
withdrawn after publishing
4 years old
26 releases · first in 2022
One column per quarter.
Fix Python 3.12.3 activation lookup error (#375).
Register models using catalogue to support external models in Auto{Decoder,Encoder,CausalLM} (#351, #352).
catalogue to support external models in Auto{Decoder,Encoder,CausalLM} (#351, #352).HFHubRepository (#354).qkv_split argument is now mandatory for AttentionHeads, AttentionHeads.uniform, AttentionHeads.multi_query, and AttentionHeads.key_value_broadcast (#374).FromHFHub mixins are renamed to FromHF (#374).FromHF.convert_hf_state_dict is removed in favor of FromHF.state_dict_from_hf (#374).@danieldk, @honnibal, @ines, @KennethEnevoldsen, @shadeMe
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Fix Python 3.12.3 activation lookup error (#377).
Ensure that parameters are leaf nodes when loading a model (#364).
Note: we have set the Torch upper bound to <2.1.0 because later versions made some incompatible changes. Newer versions of Torch will be supported by Curated Transformers 2.0.0.
Add support for model repositories other than Hugging Face Hub (#331).
fsspec filesystems as a repository type (#327, #331).config property to models to query their configuration (#328).dtype is not set in the Hugging Face configuration (#330).The new (experimental) repository API adds support for loading models from repositories other than Hugging Face Hub. You can also easily add your own repository types by implementing the Repository interface. Using a repository is as easy as calling the new from_repo method that is provided by all models and tokenizers:
from curated_transformers.models import AutoDecoder
decoder = AutoDecoder.from_repo(MyRepository("mpt-7b-my-qa"))
Curated Transformers comes with two repository classes out-of-the-box:
HfHubRepository downloads models from Hugging Face Hub and is now used by the from_hf_hub methods.FsspecRepository supports the wide range of filesystems provided by the fsspec package and third-party implementations.@danieldk, @honnibal, @ines, @shadeMe
Add support for Safetensor checkpoints (#310).
from_hf_hub_to_cache method to FromHFHub mixins. This method downloads a model from Hugging Face hub to the local cache without loading it (#303).no_bias config option in layer norms (#321).MPTGenerator (#317).@danieldk, @honnibal, @ines, @mayankjobanputra, @shadeMe
Add decoder and causal LM for MosaicML MPT (#294).
@danieldk, @honnibal, @ines, @shadeMe
Three weeks on the heels of our tech preview we are excited to announce first stable release of Curated Transformers! 🎉 From this release onwards, we
Three weeks on the heels of our tech preview we are excited to announce first stable release of Curated Transformers! 🎉 From this release onwards, we provide a stable API following semantic versioning guidelines. Of course, this release is also packed with new features.
torch.compile (#257) and TorchScript tracing for all models (#262, #266).@danieldk, @honnibal, @ines, @shadeMe, @svlandeg
Nothing published for this version
Nothing published for this version
Support old and new Falcon model type and configuration (#243).
@danieldk
We are very happy to announce this major new release of Curated Transformers! 🎉
We are very happy to announce this major new release of Curated Transformers! 🎉
Curated Transformers started as a small transformer library for spaCy pipelines. Over the last two months we made it a pure PyTorch library that is completely independent of spaCy and Thinc. We also added support for popular LLM models, generation, 8-bit/4-bit quantization, and many other features:
bitsandbytes.meta devices.tokenizer.json tokenizers.transformers package.Curated Transformers can be used in spaCy using the spacy-curated-transformers package.
@danieldk, @honnibal, @ines, @shadeMe
Remove unused cutlery dependency (#171).
spaCy/Thinc-specific code and entrypoints are moved to the new `spacy-curated-transformers` package.
spaCy/Thinc-specific code and entrypoints are moved to the new spacy-curated-transformers package.
Normalize captitalization in entry point identifiers (#132).
Fix #126: Restore TorchScript functionality (necessary for quantization) (https://github.com/explosion/curated-transformers/pull/129).
Add a workaround for invalid outputs of nn.Linear on MPS on macOS 13.2.x (#124).
nn.Linear on MPS on macOS 13.2.x (#124).mypy fixes for BertTokenizerFast attributes (#118).Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Your coding agent can read these notes before it upgrades. Set up the MCP server →