Skip to content

Releases: royshil/occ-ai-dep-whispercpp

0.0.13

25 Nov 14:43

Choose a tag to compare

What's Changed

  • Enable Vulkan GGML backend for MacOS by @Tabby in #12

Full Changelog: 0.0.12...0.0.13

0.0.12

24 Nov 20:36
3535261

Choose a tag to compare

What's Changed

  • Update Windows Vulkan SDK to 1.4.328.1 by @Tabby in #11
  • Build ROCm for all AMD GPU targets by @Tabby in #9 and #10

Full Changelog: 0.0.11...0.0.12

0.0.11-2

20 Nov 21:53

Choose a tag to compare

  • Fix Accelerate framework for MacOS 12
  • Add back OpenMP on Mac
    Full Changelog: 0.0.11-1...0.0.11-2

0.0.11-1

17 Nov 15:48

Choose a tag to compare

  • Make MacOS variants targetting different Metal versions, and a CoreML disabled variant
  • Remove OpenMP support on MacOS until crashes and signing issues are resolved
  • Try to fix missing debug info for AMD variants
  • Build AMD variants for more GPU targets
    Full Changelog: 0.0.10-2...0.0.11-1

0.0.10-2

06 Nov 09:20

Choose a tag to compare

  • Build with OpenMP on MacOS
  • Use newer XCode (16.2, same as we use to build the plugin) to fix the incorrect handling of nosve on ARM so we can generate backends for all Apple ARM CPUs
  • Fix missing CoreML shaders from release package

Full Changelog: 0.0.10...0.0.10-2

0.0.10

05 Nov 19:21

Choose a tag to compare

  • Use dynamic Whisper backends
    • Allows the best backends to be loaded and used at runtime without the risk of crashes from backends that can't run on a user's machine, as they simply won't be loaded
    • Allows more advanced CPU features to be used on machines that can do so, while also making it possible to run the plugin on simpler, older hardware without penalising all users by limiting everyone to the worst CPU supported
  • New/updated Linux variants:
    • generic - built with OpenBLAS, Vulkan and OpenCL support
    • nvidia - same as generic but also with CUDA support
    • amd - same as generic but also with hipblas/ROCm support
  • New/updated Windows variants:
    • generic - built with OpenBLAS and Vulkan support (OpenCL to be added later if possible)
    • nvidia - same as generic but also with CUDA support
    • amd - same as generic but also with hipblas/ROCm support
  • MacOS changes:
    • x86_64 - no significant change aside from dynamic backends allowing for newer CPUs to be targeted
    • arm64 - not possible to build for all Apple ARM CPU variants yet due to a bug in whisper so only M1 cpu variant is available, but is at least better than being built for generic ARM (ARMv8.2 with dotprod, instead of ARMv8.0)

Full Changelog: 0.0.9...0.0.10

0.0.9-1

02 Nov 09:40

Choose a tag to compare

  • Fix hipblas build
  • Generate debug symbols for all configurations except hipblas because it uses a different compiler setup
  • Add dynamic cpu configuration
    Full Changelog: 0.0.9...0.0.9-1

0.0.9

01 Nov 09:26

Choose a tag to compare

  • Update Whisper.cpp from v1.7.4 to v1.8.2
  • Build with OpenBLAS for all configurations on Linux and Windows

Full Changelog: 0.0.8...0.0.9

0.0.8-1

04 Feb 13:54

Choose a tag to compare

0.0.8

03 Feb 13:32
290cfff

Choose a tag to compare