Go modules · #2337 by repository stars

github.com/hybridgroup/yzma

v1.29.1hybridgroup/yzma ↗New releases in your daily digest, changelog included.

Last release today
07 Oct 2026
Ships on a steady schedule
a new release about every 9 days
Some releases are documented
notes for 21 of 49 stable releases
Nothing withdrawn
no release was ever pulled
1 years old
180 releases · first in 2025

Release timeline

180 releases since 2025

20252026

One column per month.

Releases

v1.29.17 Oct 2026
llamawasm: return the named template from ModelChatTemplate by @deadprogram in #450
What's Changed

llamawasm: return the named template from ModelChatTemplate by @deadprogram in #450

Release 1.29.1 by @deadprogram in #451

Full Changelog: v1.29.0...v1.29.1
Open source →
v1.29.1-0.20261007082345-212e061b24007 Oct 2026pre-release
Nothing published for this version

llama, mtmd: mark SetWarmup and Encode as deprecated by @deadprogram in #394

What's Changed

download: use newest nightly build again after the release by @deadprogram in #388
Add Apple Macbook M4 Max benchmarks by @dlsniper in #387
llama: set the generation threads from the model size by @deadprogram in #389
docs: update instructions for coding tools by @deadprogram in #390
llama: report ggml backend libraries that fail to load by @deadprogram in #391
yzma-checker: find bindings made through loader.Lib by @deadprogram in #392
llama: add the extended batch API, Version and SplitModeTensor by @deadprogram in #393
llama, mtmd: mark SetWarmup and Encode as deprecated by @deadprogram in #394
benchmarks: add images to the engine comparison by @deadprogram in #395
ROCm 10.0 benchmarks by @dlsniper in #396
deps: update ffi to v0.8.0 by @deadprogram in #397
exp/decide: add System One decisions for Jev-Style and JevK5 models by @deadprogram in #398
exp/decide: run in the browser with pkg/llamawasm by @deadprogram in #399
wasm: give the Chrome switches that turn on Vulkan WebGPU on Linux by @deadprogram in #400
wasm: let a page select the GPU with a power preference by @deadprogram in #401
wasm: run the WebGPU build on a real GPU with no browser by @deadprogram in #402
wasm: give speed as the reason that auto mode skips Firefox by @deadprogram in #403
benchmarks: add WebGPU results in Chrome, and bring the WebGPU notes up to date by @deadprogram in #404
wasm/serve: send Cache-Control no-cache by @deadprogram in #405
docs: rewrite comments and docs in a simple direct style by @deadprogram in #406
exp/decide: add decider models by @deadprogram in #407
examples/wasm: add a README for each example by @deadprogram in #408
llama: add GetCausalAttn binding for llama_get_causal_attn by @deadprogram in #410
llama: do not accept draft tokens twice in DraftGenerate by @deadprogram in #412
llama: return failure when a string argument contains a NUL byte by @deadprogram in #413
llama: return LoadModeAuto for an unknown string in LoadModeFromStr by @deadprogram in #414
llama: return an error when AdapterLoraInit fails to load by @deadprogram in #415
llama: unload every backend in Close by @deadprogram in #416
llama: use YZMA_LIB in Load and return the error from Init by @deadprogram in #417
llama, mtmd: share one callback for log, abort and progress by @deadprogram in #418
llama, llamawasm: return RoPEType from ModelRopeType by @deadprogram in #419
llama: fix NewSampler params and EOG bias by @deadprogram in #420
llama: derive token capacity in StateLoadFile and StateSeqLoadFile by @deadprogram in #421
mtmd: size BitmapGetData with BitmapGetNBytes by @deadprogram in #422
llama, mtmd: free the old progress closure in SetProgressCallback by @deadprogram in #425
download: keep tar extraction inside the destination by @deadprogram in #427
vlm: grow buffers in ChatTemplate and Results by @deadprogram in #429
message: stop parseGLMToolCalls panicking on out of order tags by @deadprogram in #428
examples/chat: keep the history and decode only the new turn by @deadprogram in #430
llamawasm: return an error when a shim promise is rejected by @deadprogram in #431
yzma-checker: check bindings built from a looked up address by @deadprogram in #432
mtmd: lock HelperEvalChunks per context instead of globally by @deadprogram in #433
mtmd: take a byte slice in BitmapInit by @deadprogram in #434
llama: let Tokenize take text with NUL bytes by @deadprogram in #435
llama: bump session and state seq versions for llama.cpp v0.6.0 by @deadprogram in #443
llama: add BatchExtSetDecisionOrder by @deadprogram in #445
download: use OpenVINO 2026.4.1 asset names from b11374 by @deadprogram in #446
exp/decide: add the TypeSafe /v1/systemone request and answer by @deadprogram in #437
exp/decide: escape special token text and add BothOrders by @deadprogram in #438
exp/decide: add Open for GGUF models that describe their readout by @deadprogram in #441
exp/decide: add Laya, Julia-1 and Kev to Open by @deadprogram in #442
loader: resolve Windows DLL dependencies from the bundle folder by @deadprogram in #448
Release 1.29.0 by @deadprogram in #449

Full Changelog: v1.28.0...v1.29.0

Open source →

v1.28.1-0.20261006144613-08683a0ecd416 Oct 2026pre-release

Nothing published for this version

v1.28.1-0.20261006125350-87450fc2f90b6 Oct 2026pre-release

Nothing published for this version

v1.28.1-0.20261001235646-c25d400d11081 Oct 2026pre-release

Nothing published for this version

v1.28.1-0.20261001182953-daaa4da3a5f31 Oct 2026pre-release

Nothing published for this version

v1.28.1-0.20261001074601-0889e647c9591 Oct 2026pre-release

Nothing published for this version

v1.28.1-0.20260928145547-0af2b85d410228 Sept 2026pre-release

Nothing published for this version

v1.28.1-0.20260927143541-49b68ed4da0427 Sept 2026pre-release

Nothing published for this version

v1.28.1-0.20260927134955-0a1cd493ce6727 Sept 2026pre-release

Nothing published for this version

v1.28.1-0.20260927090251-e4b426e42d4c27 Sept 2026pre-release

Nothing published for this version

v1.28.1-0.20260926062451-22c9a967176026 Sept 2026pre-release

Nothing published for this version

v1.28.1-0.20260925161907-6067070de81325 Sept 2026pre-release

Nothing published for this version

v1.28.1-0.20260925130510-e89046613a6d25 Sept 2026pre-release

Nothing published for this version

v1.28.1-0.20260924123542-ef05851f245024 Sept 2026pre-release

Nothing published for this version

v1.28.024 Sept 2026

download: use the newest nightly build again after the release by @deadprogram in #351

What's Changed

download: use the newest nightly build again after the release by @deadprogram in #351
exp/speculative: add bindings for the NextN hidden state functions by @deadprogram in #352
llamawasm: fail when the model download stops early by @deadprogram in #353
llamawasm: add the calls that read the logits and the embeddings by @deadprogram in #354
llamawasm: add the calls for the sampling of the backend by @deadprogram in #355
llamawasm: add the calls that change a context after it is made by @deadprogram in #356
llamawasm: add the calls that describe a model and the threads of a context by @deadprogram in #357
wasm: use CPU when GPU computes wrong values by @deadprogram in #359
download: install correct version of CUDA for Windows builds by @deadprogram in #361
benchmarks: reorganize and automate benchmark data collection by @deadprogram in #360
benchmarks: one comparable set of numbers by @deadprogram in #362
download: add OpenVINO library downloads by @deadprogram in #364
benchmarks: add comparison suite by @deadprogram in #365
benchmarks: keep the stream test above the clock resolution of Windows by @deadprogram in #366
download: install the CUDA build that matches the machine by @deadprogram in #367
benchmarks: add Jetson Orin on llama.cpp v0.4.1 CPU/CUDA/Vulkan by @deadprogram in #368
benchmarks: add Arduino UNO Q on CPU by @deadprogram in #369
llama: use one thread for each performance core by @deadprogram in #372
llama: pin inference threads to the performance cores by @deadprogram in #375
benchmarks: pin thread pool held to the cores by @deadprogram in #376
benchmarks: fresh numbers of the Intel Core i9-13900HX by @deadprogram in #378
llamawasm: add the metadata, the state, and the performance counters by @deadprogram in #379
benchmarks: use 4 threads for the text benchmark by @deadprogram in #381
benchmarks: fresh numbers of all machines by @deadprogram in #382
Release 1.28.0 by @deadprogram in #383
benchmarks: results for llama.cpp v0.5.0 by @deadprogram in #384
Add Vulkan Benchmarks for Linux by @dlsniper in #386

Full Changelog: v1.27.0...v1.28.0

Open source →

v1.27.1-0.20260923140914-bbdc74b0a96123 Sept 2026pre-release

Nothing published for this version

v1.27.1-0.20260922191900-c134e34de1a122 Sept 2026pre-release

Nothing published for this version

v1.27.1-0.20260919172741-00744d16f6f619 Sept 2026pre-release

Nothing published for this version

v1.27.1-0.20260918082757-8ddb0055ea9718 Sept 2026pre-release

Nothing published for this version

v1.27.1-0.20260917101309-d48a589de71117 Sept 2026pre-release

Nothing published for this version

v1.27.1-0.20260916134945-b5f55bd9404f16 Sept 2026pre-release

Nothing published for this version

v1.27.1-0.20260915185616-51bc945d799715 Sept 2026pre-release

Nothing published for this version

v1.27.014 Sept 2026

download: use newest nightly llama.cpp build again by @deadprogram in #338

What's Changed

download: use newest nightly llama.cpp build again by @deadprogram in #338
docs: move the long documents to the website by @deadprogram in #339
mtmd: expose mtmd_image_tokens_get_decoder_pos by @deadprogram in #345
Release 1.27.0 by @deadprogram in #350

Full Changelog: v1.26.1...v1.27.0

Open source →

v1.26.2-0.20260914121620-e8562a08d1d014 Sept 2026pre-release

Nothing published for this version

v1.26.2-0.20260906172259-ece48906649f6 Sept 2026pre-release

Nothing published for this version

v1.26.15 Sept 2026

This release makes a check of an installation work without a network.

An offline check. Install keeps the digest manifest beside the install record,
as yzma-manifest.json, and puts the digest of it in the record. yzma verify reads
that copy and checks the bytes against the pin you give, or against the recorded
digest when there is no pin, so the trust is the same as before. The manifest also
names the nightly build of a tagged release, which removes a second fetch. An
installation that an earlier release made needs one check with a network, and none
after that.
A pinned llama.cpp release. This release of yzma installs llama.cpp v0.4.0
when no version is given, the same as v1.26.0. Use --version latest to get the most
recent nightly build.

What's Changed

download: use newest nightly build again after the release by @deadprogram in #333
download: save install manifest by @deadprogram in #335
Release 1.26.1 by @deadprogram in #337

Full Changelog: v1.26.0...v1.26.1

Open source →

v1.26.1-0.20260905103157-8001553069b55 Sept 2026pre-release

Nothing published for this version

v1.26.05 Sept 2026

This release lets yzma run in a browser, and it checks what it installs.

This release lets yzma run in a browser, and it checks what it installs.

WebAssembly. llama.cpp becomes a WebAssembly module, and a Go program compiled
by TinyGo drives it through the new pkg/llamawasm package. There are three builds,
and the JavaScript glue takes the best one the browser can run: WebGPU, more threads,
or one thread. Text, embeddings, and images all work, with chat templates and tool
calling. The names and the order of the calls are the same as in pkg/llama and
pkg/mtmd. See wasm/README.md.
Digest checks. Install checks the SHA-256 of each file against the manifest that
llama-cpp-builder publishes. A version takes the form VERSION@sha256:DIGEST to pin
the manifest as well, and the new yzma verify command checks an installation later.
A pinned llama.cpp release. This release of yzma installs llama.cpp v0.4.0 when
no version is given. Use --version latest to get the most recent nightly build.
mtmd gets TokenizeFromParts and InputPart, and llama gets a wrapper for
llama_sampler_init_temp.
The installer follows the ROCm asset names of llama.cpp from b10356 and b10767.

What's Changed

llama, mtmd: follow the upstream llama.cpp b10655 ABI changes by @deadprogram in #298
download: install the llama.cpp version pinned by this yzma release by @deadprogram in #300
llama: follow the upstream llama.cpp b10675 ABI change by @deadprogram in #301
loader: keep the FFI argument types alive by @deadprogram in #304
llama: align lazy-loading API names with llama.cpp b10679 by @deadprogram in #305
Run llama.cpp in a browser with TinyGo by @deadprogram in #306
WebGPU in the browser by @deadprogram in #307
A model with eyes in the browser by @deadprogram in #308
download: follow the llama.cpp ROCm asset rename at b10356 by @deadprogram in #310
wasm: start the threads of llama.cpp correctly by @deadprogram in #311
wasm: add chat templates and tool calling by @deadprogram in #312
docs: show the WebAssembly state of each function in the roadmap by @deadprogram in #313
download: support llama.cpp ROCm 10.0 asset names from b10767 by @deadprogram in #316
llamawasm: do not put the model in the browser cache by @deadprogram in #317
wasm: run the WebGPU build in Firefox by @deadprogram in #318
mtmd: add TokenizeFromParts and InputPart by @deadprogram in #320
wasm: use the CPU in Firefox and document browser needs by @deadprogram in #324
download: check the digest of each asset and add a verify command by @deadprogram in #323
download: accept a SHA-256 digest with a version by @deadprogram in #326
wasm: bind the batch and memory calls of ABI 6 by @deadprogram in #328
download: publish and read the digest of the version manifest by @deadprogram in #330
Release 1.26.0 by @deadprogram in #331
download: do not read a digest mismatch as a file that is not there by @deadprogram in #332

Full Changelog: v1.25.0...v1.26.0

Open source →

v1.25.1-0.20260904102343-1e8d073c20d44 Sept 2026pre-release

Nothing published for this version

v1.25.1-0.20260903130709-6bd0208c33c63 Sept 2026pre-release

Nothing published for this version

v1.25.1-0.20260903095601-133223b779b53 Sept 2026pre-release

Nothing published for this version

v1.25.1-0.20260831183726-385dd04a167131 Aug 2026pre-release

Nothing published for this version

v1.25.1-0.20260831114346-d0abb7d5acda31 Aug 2026pre-release

Nothing published for this version

v1.25.1-0.20260831083411-584e0e19e7ae31 Aug 2026pre-release

Nothing published for this version

v1.25.1-0.20260830192830-a6bdb615293a30 Aug 2026pre-release

Nothing published for this version

v1.25.1-0.20260830132541-3306fd1cf58f30 Aug 2026pre-release

Nothing published for this version

v1.25.1-0.20260829144631-5b0ce8118e3d29 Aug 2026pre-release

Nothing published for this version

v1.25.1-0.20260829130514-db42cc2c8f3929 Aug 2026pre-release

Nothing published for this version

v1.25.1-0.20260829084921-57fa60c4b14c29 Aug 2026pre-release

Nothing published for this version

v1.25.1-0.20260829081752-e345d69312f429 Aug 2026pre-release

Nothing published for this version

v1.25.1-0.20260829080449-ae7ed0ade67829 Aug 2026pre-release

Nothing published for this version

v1.25.1-0.20260827163650-c4fed7865c4c27 Aug 2026pre-release

Nothing published for this version

v1.25.025 Aug 2026

download: support the new llama.cpp tagged releases by @deadprogram in #297

What's Changed

download: support the new llama.cpp tagged releases by @deadprogram in #297

Full Changelog: v1.24.0...v1.25.0

Open source →

v1.24.021 Aug 2026

llama: bind the ggml backend device accessors for device detection by @deadprogram in #294

What's Changed

llama: bind the ggml backend device accessors for device detection by @deadprogram in #294
llama: bind the ggml type sizing functions for GGUF sizing by @deadprogram in #295
mtmd: add the device field to mtmd_context_params by @deadprogram in #296

Full Changelog: v1.23.0...v1.24.0

Open source →

v1.23.013 Aug 2026

yzma-checker: add a standalone FFI type-contract auditor by @dlsniper in #283

What's Changed

yzma-checker: add a standalone FFI type-contract auditor by @dlsniper in #283
llama: fix FFI type-contract violations at the libffi boundary by @dlsniper in #282
llama: address the FFI type-contract review follow-ups by @deadprogram in #288
download: make --upgrade actually replace the version symlinks by @dlsniper in #290
yzma-checker: seven more checks at the FFI boundary by @dlsniper in #291
ci: run the FFI checker in GitHub Actions by @deadprogram in #292

Full Changelog: v1.22.0...v1.23.0

Open source →

v1.22.01 Aug 2026

llama: add LoadMTP to ModelParams for llama.cpp b10212+ by @deadprogram in #281

What's Changed

llama: add LoadMTP to ModelParams for llama.cpp b10212+ by @deadprogram in #281

Full Changelog: v1.21.0...v1.22.0

Open source →

v1.21.030 Jul 2026

llama: correct LoadMode values for current llama.cpp by @samyfodil in #278

What's Changed

llama: correct LoadMode values for current llama.cpp by @samyfodil in #278
download: add a Resolver interface for choosing release assets by @samyfodil in #277
vocab: add wrapper for llama_vocab_get_suppress_tokens by @deadprogram in #279

New Contributors

@samyfodil made their first contribution in #278

Full Changelog: v1.20.0...v1.21.0

Open source →

v1.20.024 Jul 2026

switched to use CUDA 13.3 for Windows llama.cpp by @kaleah08 in #273

What's Changed

switched to use CUDA 13.3 for Windows llama.cpp by @kaleah08 in #273
modules: update go-getter for go 1.27 build compat. by @deadprogram in #271
pkg/llama: refactor mmap/mlock/directio into load_mode enum by @deadprogram in #274
modules: update ffi package to 0.7.0 by @deadprogram in #275
Release 1.20 by @deadprogram in #276

New Contributors

@kaleah08 made their first contribution in #273

Full Changelog: v1.19.0...v1.20.0

Open source →

v1.19.1-0.20260722214706-a4da2ece52f822 Jul 2026pre-release

Nothing published for this version

v1.19.014 Jul 2026

mtmd: add new TextLen field to mtmd.InputText by @deadprogram in #268

What's Changed

mtmd: add new TextLen field to mtmd.InputText by @deadprogram in #268
feature: add support for llama.cpp new ftype identifier functions by @deadprogram in #269
Release 1.19.0 by @deadprogram in #270

Full Changelog: v1.18.0...v1.19.0

Open source →

v1.18.024 Jun 2026

test: check for any valid chunk since return order may be different by @deadprogram in #261

What's Changed

test: check for any valid chunk since return order may be different by @deadprogram in #261
pkg/mtmd: add ProgressCallback fields missing since llama.cpp b9750 by @deadprogram in #263
pkg/llama: add support for llama_model_n_layer_nextn by @deadprogram in #264
mtmd: implement SetProgressCallback for ContextParamsType by @deadprogram in #265
mtmd: add video input support (llama.cpp #24269) by @deadprogram in #266
mtmd: add batch API wrappers (llama.cpp PR #24384) by @deadprogram in #267

Full Changelog: v1.17.1...v1.18.0

Open source →

v1.17.113 Jun 2026

mtmd: add BatchMaxTokens for new field added by llama.cpp by @deadprogram in #260

What's Changed

mtmd: add BatchMaxTokens for new field added by llama.cpp by @deadprogram in #260

Full Changelog: v1.17.0...v1.17.1

Open source →

v1.17.011 Jun 2026

mtmd: modify Bitmap file helpers to return new BitmapWrapper by @deadprogram in #259

What's Changed

mtmd: modify Bitmap file helpers to return new BitmapWrapper by @deadprogram in #259

Full Changelog: v1.16.1...v1.17.0

Open source →

v1.16.18 Jun 2026

context: add new parent context for MTP support added in llama.cpp by @deadprogram in #257

What's Changed

context: add new parent context for MTP support added in llama.cpp by @deadprogram in #257

Full Changelog: v1.16.0...v1.16.1

Open source →

v1.16.06 Jun 2026

Nothing published for this version

v1.15.02 Jun 2026

Nothing published for this version

v1.14.118 May 2026

Nothing published for this version

v1.14.017 May 2026

Nothing published for this version

Your coding agent can read these notes before it upgrades. Set up the MCP server →