chore(gpu): pin whisper.cpp GPU packs to release 0.0.9 - #1584
Merged
Conversation
0.0.9 is the same whisper.cpp source as 0.0.8 (one commit apart, only the build workflow changed), rebuilt with Pascal (sm_61) added to CMAKE_CUDA_ARCHITECTURES. CUDA zips grow by 17.5 MB; Vulkan zips are unchanged in size but rebuilt, so both managers pin fresh digests (verified independently against downloaded assets). No behavior change for any user: installed packs are never re-checked against the tag, so existing installs keep their working binaries; new downloads fetch 0.0.9, which carries identical kernels for every card the compute-cap gate currently admits (7.5+). The gate itself is unchanged — offering CUDA to Pascal cards (6.1) is a separate, hardware-verified follow-up. The 0.0.8 digests stay pinned so a WHISPER_CPP_VERSION rollback keeps its integrity check.
This was referenced Aug 12, 2026
gabrielste1n
added a commit
that referenced
this pull request
Aug 12, 2026
Roll the unreleased changelog into 1.8.3 — covering the GPU acceleration overhaul (#1575-#1578, #1584), fail-closed LLM routing (#1583), STT stabilization (#1556), Linux launch at login, and the rest of the fixes since 1.8.2 — and bump the version. Tagging v1.8.3 (which triggers the release build) is deliberately left as a separate manual step.
Phenisys-adm
pushed a commit
to Phenisys/openwhispr
that referenced
this pull request
Aug 27, 2026
- privacy: wait for the renderer sync before the first retention sweep (OpenWhispr#1558) - cli: respond with HTTP 400 validation_error on route validation errors (OpenWhispr#1521) - parakeet: normalize non-PCM16 WAV input (OpenWhispr#1376); reject cancelled uploads before ffmpeg conversion using the fork's abortError helper (options.signal pattern, matching whisper.js) (#53ec0751) - gpu: stop offering the CUDA pack to cards the build cannot run on (OpenWhispr#1576); pin whisper.cpp GPU packs to release 0.0.9 (OpenWhispr#1584)
19 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Bumps the GPU pack pin from
0.0.8to0.0.9in bothwhisperCudaManagerandwhisperVulkanManager(they shareWHISPER_CPP_TAG), with fresh sha256 pins for all four assets — verified independently by downloading and hashing, not just trusting the API digest.What 0.0.9 is: the exact same whisper.cpp source as 0.0.8 (
compare 0.0.8...0.0.9= 1 commit, 1 file: the build workflow), rebuilt with Pascal (sm_61) added toCMAKE_CUDA_ARCHITECTURES(OpenWhispr/whisper.cpp#5). CUDA zips: +17.5 MB (Win 755→772.5 MB, Linux 201→218.7 MB); Vulkan zips unchanged in size.Zero behavior change for users:
scripts/download-whisper-cpp.jsis deliberately untouched.WHISPER_CPP_VERSION=0.0.8rollback remains integrity-pinned.Offering CUDA to Pascal cards (lowering
MIN_CUDA_COMPUTE_CAP7.5 → 6.1) is a separate follow-up PR, held until a real Pascal card smoke-tests the new kernels.Full suite: 1922 pass / 0 fail.