model2vec Releases
74 releases of MinishLab/model2vec
- v2.10.1
A maintenance release involving a version bump.
Nov 22, 2025
- 3.4.0
Fixed an issue where CI was incorrectly closing valid issues and updated the README.
Oct 28, 2025
- v0.7.0
This release focuses on performance improvements including faster model loading, vocabulary quantization, and configurable padding tokens. It also introduces classifier freezing for fine-tuning scenarios and deprecates older parameters like apply_zipf and use_subword.
Oct 5, 2025
- v1.2.1v1.2.1 Release
A bug fix release focused on frontend improvements, introducing new code assistance rules, and enhancing the MCP server configuration.
Aug 26, 2025
- 1.2.0v1.2.0
Aug 3, 2025
- v0.6.00.6.0
This release focuses on reducing tokenizer size and adding supertokenizers for improved performance. It also adds explicit validation set support and makes normalization dependent on spacing.
Jun 3, 2025
- v0.5.0
This release introduces quantization capabilities and faster inference for large vocabularies. It adds dimensionality configuration during loading and token provenance tracking.
Apr 30, 2025
- 0.4.1
This release adds training flexibility with min/max epoch configuration and multilabel classification support. It also introduces an evaluation function for classifiers.
Feb 28, 2025
- v1.1.0
This release introduces group chat functionality, Silero-VAD voice activity detection, and CosyVoice2 text-to-speech support. It also adds a web-based interface for ASR/TTS tools and improves configuration management with a new template system.
Feb 19, 2025
- 0.4.0
This major release introduces the Fittable class for model training and increases training capacity from 8m to 32m parameters. It also includes comprehensive documentation restructuring.
Feb 12, 2025
- v0.3.9
Documentation updates including new model results, alongside the addition of a trust-remote-code option for model loading and a SIF-like coefficient feature.
Feb 6, 2025
- v1.0.0v1.0.1
A major rewrite of the project featuring a completely new frontend (React, ChakraUI, Vite), vision capability, and a desktop pet mode. The backend was rebuilt for asynchronous operations and now supports a wider range of LLM providers including DeepSeek R1.
Feb 4, 2025
- v0.3.8
Focuses on improving distillation for modernBERT, fixing tokenizer issues, and adding optional embedding normalization to StaticModel loading.
Jan 27, 2025
- v0.3.7
Updated the `save_pretrained` method to save models in a format compatible with sentence-transformers.
Jan 21, 2025
- v0.3.6
Added the ability to load models from the sentence-transformers library format (st).
Jan 15, 2025
- v0.3.5
Fixed issues related to local distillation functionality.
Jan 11, 2025
- v0.3.4
Added documentation for integrations, support for multiprocessing during training, and a semantic chunking tutorial.
Jan 5, 2025
- v0.5.2
A minor patch release that reverts the default ASR provider back to FunASR to fix issues for users without CUDA cuDNN.
Dec 21, 2024
- v0.3.3
Added a script to support exporting models as ONNX and tokenizer files.
Nov 20, 2024
- v0.3.2
A minor maintenance release focused on fixing a specific bug in package extras and bumping the version.
Oct 31, 2024