mistral.rs Releases
113 releases of EricLBuehler/mistral.rs
- v0.1.12
Added a new example, fixed local loading authentication issues, and refactored GGUF/GGML loaders.
May 31, 2024
- v0.1.11
Refactored the ModelKind structure using strum and added verbose mode and GGUF tokenizer support.
May 28, 2024
- v0.12.0
This release focuses on expanding integration capabilities by adding support for Disqus, Google Tasks, Vtiger CRM, YNAB, Airtable, and Appwrite. It also includes improvements to Docker configuration, formatter actions, and Single Sign-On (SSO) functionality.
May 28, 2024
- v0.1.10
Refactored layers.rs and removed the candle-layer-norm dependency while fixing KV cache issues.
May 22, 2024
- v0.1.9
Improved chat templates documentation and utilized cuBLASlt for faster attention computations.
May 18, 2024
- v0.1.8
Major update featuring low-precision matmul kernels, a new loading API, and improved CTRL-C handling.
May 16, 2024
- v0.1.7
Updated containers to CUDA 12.4 and fixed missing libraries.
May 14, 2024
- v0.1.6
This release focuses on performance optimizations and new model support, specifically introducing speculative decoding and dynamic LoRA swapping, while refining model selection and fixing critical masking bugs.
May 13, 2024
- v0.1.5
A maintenance-focused release that introduces the quantized Phi 3 model and improves build compatibility and benchmarking capabilities.
May 9, 2024
- v0.1.4
This release features a significant internal refactor of the pipeline to improve organization and stability, alongside user-facing enhancements like EOF support in interactive mode.
May 8, 2024
- v0.1.3
This release streamlines the distribution process by automating PyPI uploads and Docker builds, while updating the PyO3 interface to accept dictionaries.
May 2, 2024
- v0.1.2
This release introduces asynchronous integrations for better concurrency and finally stabilizes the Phi 3 128k model, along with intermediate loading capabilities for quantized models.
Apr 30, 2024
- v0.1.0
Apr 27, 2024