mistral.rs Releases
113 releases of EricLBuehler/mistral.rs
- v0.13.0
Introduces a new flow editor, user invitation functionality, and various formatter actions, alongside refactoring of authentication flows to use REST APIs.
Aug 2, 2024
- v0.2.4
Fixes Metal build issues and implements tool calling functionality.
Aug 1, 2024
- v0.2.3
Implements min-p sampling and improves PagedAttention KV cache allocation.
Jul 28, 2024
- v0.2.2
Fixes scheduler v2 Ctrl+C handling, makes Mixtral sliding window optional, and adds Llama 3.1 scaled rope support.
Jul 24, 2024
- v0.2.1
Fixes path normalization, adds Mistral Nemo support, and adds server throughput logging.
Jul 23, 2024
- v0.2.0
Introduces PagedAttention (major performance gain), supports .bin/.pt/.pth extensions, and adds Starcoder 2 GGUF.
Jul 19, 2024
- v0.1.26
This release focuses on improvements to device mapping, memory tracking, and the addition of LLaVA support alongside various documentation and installer updates.
Jul 5, 2024
- v0.1.25
Major feature release introducing AnyMoE (Mixture of Experts), Starcoder 2 model, vision interactive mode, and optimized loading speeds.
Jul 3, 2024
- v0.1.24
A minor patch release primarily focused on bumping the version number.
Jun 30, 2024
- v0.1.23
Introduced the Gemma 2 model, improved decoding performance, and updated documentation and dependencies.
Jun 29, 2024
- v0.1.22
Focuses on CUDA build improvements, CI stability, and a fix for a divide-by-zero error in CUDA kernels.
Jun 24, 2024
- v0.1.21
Focuses on error handling, dtype selection improvements, and cross-GPU device mapping support.
Jun 23, 2024
- v0.1.20
Focuses on Docker support, Dockerfile updates, and adding Idefics 2 support.
Jun 20, 2024
- v0.1.19
Focuses on memory optimization during loading and code cleanup.
Jun 15, 2024
- v0.1.18
Focuses on dependency updates (Candle, Jinja2) and adding bitwise operators.
Jun 12, 2024
- v0.1.17
Focuses on GGUF format support (Mixtral experts, chat templates), Phi-3 vision, and documentation.
Jun 11, 2024
- v0.1.16
This release introduces the Phi 3 vision model and features significant refactoring work by @polarathene, including improvements to Metal loading and logging.
Jun 8, 2024
- v0.1.15
A minor patch release focused on fixing an incorrect unwrap and bumping the version.
Jun 5, 2024
- v0.1.14
Improved documentation for local GGUF loading and fixed various tokenizer and message handling issues.
Jun 5, 2024
- v0.1.13
Preparation for supporting multiple model types and fixes for chat templates and CUDA GEMM bugs.
Jun 2, 2024