Overview

Maintenance release. Updates to ggml v0.26.0.

Nightly build: b5454
More info: dist : releases and versioning of ggml-org projects

Changelog since v1.9.4

d1be6fd whisper : bump version to 1.9.5 (#4102)
4afec37 talk-llama : update llama.cpp to v0.6.0
3d451ab sync : ggml
47f1c7d ggml : bump version to 0.26.0 (ggml/1652)
e71b784 CUDA: make the alloc_deps check batch independent (llama/29986)
cdc8730 vulkan: fix Flash Attention shmem write out of bounds (llama/29988)
563a9f7 vulkan: revert mul_mat_id tile selection PR #29182 (llama/29936)
ebb8a17 cuda: use the vector lightning indexer kernel on MUSA (llama/29990)
8c6ca4a CUDA: Optimize accumulation in mmq for NVFP4 type (llama/29857)
82fa9df vulkan: fix stale prealloc_y reuse across flash attention and soft_max (llama/29591)
f691542 vulkan: sparse flash attention for quantized K/V (llama/29639)
d477de3 llama : fix unexpected graph reallocation in the k-pool models (llama/29958)
95283ca ggml-cpu : add Q8_0 IME1 matrix kernel for SpacemiT X60 (llama/28479)
155c5c2 vulkan : Fix undeclared identifiers when -DGGML_VULKAN_RUN_TESTS=ON (llama/29912)
7091d4a webgpu: add MMVQ support for Q1_0/Q5_0/Q5_1/Q3_K/Q5_K/Q6_K/MXFP4 (llama/29483)
4cc62b8 cuda: tile the lightning indexer over keys and tokens for 4 heads (llama/29901)
7e576cb metal : few-row MMA mat-mul (llama/29869)
91da470 CUDA: use MMVF for thin f16/bf16 mul_mat at small batch size (llama/29633)
8bb85d9 CUDA: prefer whole-tile FlashAttention scheduling for efficient two-stage kernels (llama/29435)
ce6552e CUDA: refactor swizzling code (llama/29612)
bfe3dc6 ggml-cpu: support BF16/FP16/FP32 K tails in tinyBLAS on x86 (llama/29806)
25453c0 cuda : move neu_padded to where it is used (llama/29940)
7914776 cuda : move blocks_per_col to where it is used (llama/29939)
6421d2c CUDA: fix MMQ memory fault if n_expert >> n_ubatch (llama/29941)
05d4d89 vulkan: fix rdna4 mat_vec tuning (llama/29934)
51bfbd6 webgpu: add f16 support to fill/set_rows (llama/29897)
f2aa80a ggml-openvino: update to 2026.4.1, optimize performance, expand ops, improve device listing. (llama/29852)
7f4a67a qwen4exp : halve the indexer score memory (llama/29825)
cf5d9e3 CUDA: fuse shared experts into MMVQ (llama/29184)
d7682b5 ggml-cuda : fix cpy transposed path corrupting non-contiguous dst (llama/27663)
6b70471 ggml-quants : avoid invalid rounding in qkx3 scale search (llama/29817)
0b35d1f ggml-cpu : fix soft_max_back wrong output when dst aliases src1 (llama/27096)
d0dc436 metal : add tensor API flash attention kernel for F16 KV (llama/29570)
cca8f72 opencl: use sigmoid f16 for bf16 (llama/29787)
cd1bee5 SYCL: Q8_0 DMMV ESIMD and MMVQ wide load (llama/29186)
95a05ed vulkan: disable large matmul tile on Samsung GPUs with 32KB shared memory (llama/28531)
c4051a3 sycl: large register file for D=512 FA vec kernels (llama/29062)
ae92605 sycl : do not use slow oneDNN reference matmul and fattn (llama/28985)
9c968f7 qwen4exp : optimize mask constructions (llama/29824)
4435763 ggml : add alloc_buffer_n to buffer type interface (llama/23671)
8291ab8 vulkan: add logging to pipeline compile issues (llama/29794)
eaadab3 hexagon: install rebuilt HTP skels (llama/29828)
0295ef6 hexagon: add q2_k and q3_k quant type support (llama/29717)
396f68d CUDA: fix 2 broken Volta FA cases (llama/29803)
3a3598c llama: refer to segment documentation [no ci] (llama/29074)
15229ac hexagon: shared strided DMA copy for CPY and CONCAT, any-dim CONCAT via DMA (llama/29685)
aa53800 cuda : route sm70 to the Turing MMVQ nwarps table (llama/29753)
81ca4f8 metal : release temporary private transfer buffers (llama/29777)
88948bd webgpu: add bfloat16 support for MUL_MAT/MUL_MAT_ID/GET_ROWS- #29358 (llama/29358)
10872be CUDA: Handle compute type for NVFP4 on cublass path (llama/29173)
3b68f90 CUDA: Make CCCL configurable + pin it to 3.4.3 for CI jobs (llama/29792)
56500f4 meta: clear inactive AllReduce shards with FILL, not SCALE (llama/29793)
817294a HIP: avoid treating CDNA as dgx spark for gqa_ratio 20 in fattn_mma dqk 576 (llama/29572)
b4a085a hex-workqueue: fix race condition in seqn getting out of sync with idx_read/write (llama/29785)
0ecc317 BLAS : Document AOCL-BLAS build and label the device AOCL-BLAS (llama/29640)
67ad86d opencl: mark vec subgroup bcast as supproted for Adreno E17 compiler (llama/29698)
61ea502 metal : use bf16 math for mxfp4 mul-mat (llama/29770)
d371e37 webgpu: fix SSM_SCAN binding aliasing (llama/29750)
c1c1390 ggml-opencl : replace alloca() with std::vector (llama/29765)
bffb6e4 cuda: guard the iq4_nl dequantize row kernel against short rows (llama/29683)
6773ef8 Hexagon: optimize ALLREDUCE with support for safe scatter mode (llama/29757)
02ea0c2 ggml/gguf : fix integer overflow (llama/29384)
067bb06 ggml-et : remove useless alloca() (llama/29663)
fd66b6e ggml : add BF16 unary, GLU, binary and scale ops (CPU, CUDA) (llama/29675)
c1b3fb1 cpu: accept BF16 in src1 of mul_mat (llama/28937)
d0af734 openvino: serve GET_ROWS on a weight view from the base Constant (llama/28381)
1cbf7e7 musa : define CUDA_ARCH for device passes (llama/29508)
d03bfd7 SYCL: reduce tensor allreduce sync with pinned host buffers (llama/29604)
66e8b95 ggml-zdnn: impl buffer reset, fix memory leaks (llama/29637)
b8051db Hexagon f16 activation ops (llama/29209)
6e48a36 gguf : reject tensor size that wraps after padding (llama/26979)
1442399 ggml : check row bounds in get_rows_back (llama/29575)
8d2a2cb hexagon: optimize concat op (llama/29673)
45093fc CUDA: bitonic argsort handles rows wider than one block (llama/28957)
482baac ggml-cuda: HIP: optimize packed byte subtraction (__vsubss4 -> __vsub4) (llama/29478)
3b86cab ci: add zdnn backend build but not test (llama/29541)
9f12c9f opencl: fix get_tensor for q5_K adreno gemm_nonshuffle kernel (llama/29555)
5439f3d vulkan: Tune GDN kernel, fix Intel performance (llama/29476)
6c07d82 vulkan : Load F32 A matrix 2 at a time when its 2-aligned (llama/29254)
8cc3ca4 vulkan: MOE aware mat_mul_id tile selection (llama/29182)
deae6f3 ggml : fix c++ odr by properly using GGML_COMMON_DECL_CPP (llama/29504)
b97d419 ggml : accumulate f16 dot products in f32 on AVX512-FP16 (llama/29545)
4d4817e ggml : require input tensors to be GGML_OP_NONE (llama/29647)
5e1914a hexagon: add FP32 GELU_ERF and GEGLU_ERF support (llama/29631)
763b67f ggml : collect all input tensors into graph_inputs (llama/29634)
8478129 ggml-zdnn: fix 0-row tensor crash (llama/29636)
f449de9 ggml : speed up model loading (llama/29598)
b30ef05 metal: FWHT perf optimizations (llama/29602)
eecee67 vulkan : reuse descriptor sets when bindings are constant (llama/29280)
9769010 vulkan: include functional header (llama/29597)
830dc66 ggml-openvino: mark unaligned batch-stride views unsupported (llama/29603)
81b951e webgpu: Handle unaligned writes in ggml_backend_webgpu_buffer_set_tensor (llama/29471)
ee86a15 tests : refactor test-recurrent-state-rollback (llama/29426)
83fd155 ggml-cpu: enable tiled flash attention for non-vector-multiple head dims on x86 (llama/29423)
65449db vulkan: fuse qwen4exp's SCALE -> SIGMOID -> SCALE -> hc_post chain (llama/29520)
0c17f7d HIP: fix template skip for DKQ > 256 mfma kernels (llama/29559)
0e53751 metal: support left and circular padding in GGML_OP_PAD (llama/29561)
b7b8495 Enables Windows ARM64 build with MSVC cl.exe (llama/28362)
85f6926 vulkan: fix wrong results when a mul_mat reads a slice of a larger cache (llama/28956)
51db475 opencl: refine bin kernel loading condition (llama/29503)
55a695d sycl: FWHT kernels for block widths above 512 (llama/29243)
846af52 CUDA: tune fp16 tile FlashAttention configs for head sizes 40-112 (llama/26289)
743f1ad HIP: Enable fattn-mma kernel on cdna for dkq > 256 for large batch sizes (llama/28907)
14d1aa7 vulkan: fix argsort kernel selection for Adreno (llama/29469)
24cf265 RPC: use RDMA completion channel to not spin (llama/29440)
7eea518 hexagon: support tiled Q4_0 and Q8_0 GET_ROWS (llama/29511)
fe06027 hexagon: support for backend sampler (llama/29502)
fc1ebfa cuda: support Nemotron 3 Puzzle state size 96 for ssm scan (llama/28717)
dfe8fbd cuda: add F16 input to the FWHT (llama/29096)
447a775 ggml-cpu: tiled mul_mat for k-quants (llama/27851)
bf1d787 opencl: add A8 Q8_0 non-MoE dp4a binary kernel (llama/29439)
9731f6f hexagon: find software divide calls using binary inspection tool (llama/29449)
2d91497 opencl: add bin kernel kernel_gemm_noshuffle_q5_k_f32_32b_trans_ila_a8_bin, kernel_gemm_noshuffle_q5_k_q8_1_dp4a_ila_a8_bin (llama/29401)
669188e Fixing the vulkan build issue of legacy GLSLC version that has no cooperativeMatrix API support (ggml-org/llama.cpp#29373) (llama/29409)
9a91942 metal: FWHT kernels for block widths above 512 (llama/29095)
331d3c8 sync : ggml
4b0998a metal : split fa kernels into per-dtype libraries (llama/29329)
bad9b54 sync : ggml
afc10c2 llama : add llama_prec_policy + model-driven W4A4 path (llama/24364)
d7e83b1 HIP: bump HIP_VERSION requried for fp8 to avoid missing __hip_fp8_e4m3 support in 6.2 (llama/29231)
26bfdc6 rpc: include nb in the get_alloc_size cache key and floor the result at ggml_nbytes (llama/29283)
93c4308 support sparse FA (llama/28796)
f7d5bf3 musa: fix PH1 (MTT S5000) operator failures and build issues (llama/29193)
40c872b CUDA: fuse RMS_NORM + SCALE into one kernel (llama/29393)
0458c9e hexagon: add q5_k quant type support (llama/29123)
eea9aa6 hexagon: use DMA for contiguous dim1 CONCAT (llama/29404)
4550b9d metal : fix graph capture and handle empty graphs (llama/2939...

Overview

New version has been released.

Nightly build: b5130
More info: dist : releases and versioning of ggml-org projects

Changelog since v1.9.3

927cfce metal : remove leftover ggml-metal.metal kernels file (#4051)
dd80eb2 scripts : use sed instead of grep for version parsing [no ci] (#4052)
1fa6dfa ci : update WoA CUDA 13.4 to use 13.4.1 GA (#4053)
a2b36eb whisper : bump version to 1.9.4 (#4050)
6fb4cd6 ci : add Windows On ARM support to release job (#4048)
c44b60b whisper : call encoder_begin_callback before language auto-detect (#3936)
6d0ed91 whisper : re-seed decoder 0 between calls (#4025)
cec4dbe ci : update close-issue job to not close issues (#4045)
79f2d92 tests : load backends before init when built with GGML_BACKEND_DL (#4031)
61e6cca server : return language in detect response (#4035)
52a939a sync : ggml
a937f4e ggml : bump version to 0.23.0 (ggml/1618)
11d4eec metal : add remaining fa-vec tunings for M3 Max (llama/28373)
140e57a ggml : replace compile definitions with version.h.in (llama/28364)
e2389eb ggml : rename and make private ggml_op_alloc_size_may_expand() (ggml/0)
1b37bea ggml : don't crash when backend search path can't be read (llama/28271)
f32e6fa ggml : remove GGML_CUDA_PEER_MAX_BATCH_SIZE (llama/28177)
e1bbe40 ggml-cpu(s390x) : fix q5_1 uninitialized v_acc (llama/28332)
d1e0e64 sycl: fuse rms_norm+mul+add and add+add residual chains (llama/27610)
36f170e SYCL: Refactor GGML_SYCL_ENABLE_MKL_FA to global var (llama/26863)
d784add opencl: quant lm_head / decode GEMV and medium-batch GEMM optimizations (speculative decoding/MTP) (llama/26477)
0a4a95c tune MMVQ to MMQ crossover for SM87 (llama/28285)
4dd48dd metal : add sparse FA (llama/28098)
d55d345 metal : fix glu dispatch with ne00 = 1 (llama/28306)
25350b5 CUDA: Allow concurrent streams per split for multi-GPU (llama/28198)
47d348a vulkan: fix FA dequant path engagement (llama/28190)
f24a386 sycl : enhance the api to support peer-to-peer copy (llama/27550)
a704770 sycl: reduce redundant work in Q4_K multi-column MMVQ (llama/27062)
e560569 finetune: fix no KV cache (llama/27199)
37f0f44 ggml-hexagon: add F16 support for unary ops (llama/28228)
1bdda1e metal : add fa-vec tunings for M3 (llama/28236)
3a1c7d6 metal : fix memory query under low-memory conditions (llama/27701)
4d343d7 ggml-cuda : remove unused vars (llama/28235)
519df61 CUDA + ggml: add sparse-fa for DSV4/GLM (llama/27970)
c2b4007 ggml: avoid KleidiAI buffer type init on dispatch (llama/27891)
dc70853 hexagon: MUL_MAT and MUL_MAT_ID fusion and fixes (llama/28202)
1c7d35e vulkan: handle larger batch sizes (>4) efficiently for IQ3_S mat-vec (llama/27449)
a9e5861 vulkan : only request VK_KHR_shader_bfloat16 extension if supported (llama/28155)
d57ae98 ggml-cpu : conditionally add SpacemiT IME kernel sources (llama/27961)
35133c9 opencl: fix out‐of‐bound reads in the Adreno image kernels (#27632)
c94921f hexagon: add missing FARF logs for cpy/get_rows/set_rows/gdn ops (llama/28217)
fcc2fee metal : add metallib build support for xcframework (llama/28163)
2c48678 cuda: fuse MoE weighted expert reduction (llama/25952)
f162a19 Revert "sycl : add Kronecker product FWHT support for sizes 384, 640, 768, 12…" (#28184)
408faaa sycl : add Kronecker product FWHT support for sizes 384, 640, 768, 1280 (llama/28016)
5f07f85 metal : add fa-vec tuning for M2 Pro (llama/28122)
8cca1a3 metal : add fa-vec tunings for A18 Pro (MacBook Neo) (llama/28152)
a245a8f metal : fix more leaks due to missing autoreleasepools (llama/27883)
870db2a metal : add fa-vec tuning for M2 Max (llama/28015)
4f3a2a4 sycl : support limit max alloc memory within 2GB for host-pinned memory (llama/27559)
5032008 metal: enable Metal 4.0 tensor API on M5+/A19+ (llama/27461)
8e54c65 metal : add fa-vec tunings for M1 Ultra (llama/28088)
f22bb2e CUDA: XOR swizzle flash attn K,V smem fp16 tiles (llama/25635)
dbc40ef metal : add concat support for quantized types (llama/28116)
2f608ab AVX2: Speed up large batch size prompt processing of IQ models (llama/27402)
c6934d0 metal : add top-k radix implementation (llama/28073)
088c603 opencl: tune the quant paths for Intel Xe-LP GPUs to improve its TG and PP performance (llama/26438)
c648b9a webgpu : avoid crash when offset is not multiple of 4 in WebGPU ggml_backend_tensor_get() implementation (llama/28045)
c1be45b ROCm: add radix TOP_K for long rows (llama/27466)
7614a4c metal : add fa-vec tunings for M1 (llama/28078)
b0f4bc0 CUDA: extend MOE fusion to specdec, earlier MOE glu fusion and topk-router fusion were restricted to 1 token (llama/27621)
76a51e8 sycl : Enhance to get the free memory of Intel GPU (llama/27968)
6ce7b89 vulkan: tune mat-vec rows for batched inference on Strix Halo (llama/27909)
96dddd8 ggml : add MUL_MAT to the list of ops that may need additional memory (for WebGPU) (llama/28071)
db00b01 vulkan: top_k radix select for k >= 1024 for Qwen 3.8 Flash Next (llama/28032)
01ebd22 hexagon: fix CPY fence bug (llama/28033)
e5c96ca metal : add remaining Q4_1/Q5_0/Q5_1 fa-vec tunings for M2 (llama/28017)
4089fa6 rpc: avoid serializing buffers from other servers (llama/26500)
749683d ggml : fix ggml_backend_buft_get_alloc_size() guard (llama/28038)
e9583f0 ggml: add SWIGLU_CLAMP (llama/27930)
e900a73 CUDA: use the fast mm_ids_helper path for any n_expert_used (llama/27978)
35d9e22 hip: tune rdna 3 mmq config (llama/26284)
e5c9e3e hip : optimize Q2_0 dot-product path for gfx1201 (llama/26753)
4b2243a ggml : add ggml_backend_op_alloc_size_may_expand, use it in RPC (llama/27960)
43acf3d rpc: fix apple rdma error spew on teardown (llama/27908)
1e0f382 metal: add fa-vec tunings for M3 Ultra (llama/27999)
b66593e metal : Add fa-vec tuning for M3 Pro (llama/27963)
5e49459 rpc : fix pre-rdma macOS versions (llama/27815)
3ad8b9b hexagon: support for device discovery and create sessions on demand (llama/27785)
3d4e0e9 sycl: split long rows in TOP_K instead of one work-group per row (llama/27847)
c68f205 metal : fix null-pipeline crash for F16 src1 mul_mat/mul_mat_id (llama/25648)
c969c68 ggml: allow passing alloc dependencies in graph_optimize (llama/27301)
b33bbc5 metal : add fa-vec tunings for M2 (llama/27940)
2a11026 opencl: use a better matmul path on two Adreno GPU generations (llama/27640)
285f1ff metal : assert shared memory padding (llama/27951)
308fa4f metal : add remaining fa-vec tunings for M4 Pro (llama/27915)
325c8d1 sycl: make --fit respect --fit-target better (llama/27629)
e644752 vulkan: combine duplicated fastdiv functions, rename the one optimizing small divs (llama/27526)
590fe18 metal : add fa-vec tunings for M1 Max (llama/27932)
d501a0a vulkan: Change mul_mat_id to pad K rather than N (llama/27925)
4c38040 vulkan: fix missing view-alias dependencies in ggml_vk_graph_optimize (llama/27812)
caea96f ggml : fix conv_transpose_2d for multiple batches (llama/26132)
ba99c09 Vulkan: add hoisting support for row IDs and expert count in shaders (llama/26686)
0a15087 metal : add fa-vec tunings for M4 (llama/27875)
7f78e1b OpenVINO: Update OV to 2026.3.1, whisper.cpp support, Qwen3.5 on NPU, and new ops (llama/27843)
fa4d244 sycl: use TILE for quantized KV decode on BMG (llama/26689)
97d0da2 sycl: bind the f16 KV cache in place for the oneDNN SDPA path (llama/27468)
530e3f4 metal : add fa-vec tunings for M3 Max, M5 and M5 Pro (llama/27863)
ff38b98 metal : add fa-vec tunings for M4 Pro (llama/27824)
b6571e4 ggml-hexagon: add HTP unary ops for ABS and LOG (llama/27786)
a0614d9 hex-unary: fix RMS_NORM_MUL weight-offset bugs for grouped/broadcast norms (llama/27798)
8529971 opencl: add bin kernels kernel_gemm_moe_q4_0_q8_1_dp4a_bin, kernel_gemm_moe_mxfp4_q8_1_dp4a_bin (llama/27768)
55ab1e5 Feature: Added LIGHTNING_INDEXER support for Deepseek V4 ops on Vulkan Backend (llama/27453)
a5db1d6 metal : fix memory leaks due to missing autoreleasepools (llama/27758)
3fea10d hexagon: support for multi-NPU devices (IQ9, IQ10) and fully asynchronous backend (llama/26501)
5271734 vulkan: warptiles currently assume warp sizes <= 64, clamp to work around larger warps (llama/27726)
0a02697 Implemented vulkan cross_entropy_loss and cross_entropy_loss_back (llama/27216)
82f5f85 rpc : implement event and async backend APIs (llama/18626)
9d8e6b9 cuda: unblock mmq for MoE on sm_60 (llama/26264)
8c0adb0 ggml-metal: add chunked SSD MMA for Mamba-2 prefill optimization (llama/26647)
8df657a ggml-meta: propagate buffer usage and call init on the new tensors (llama/27586)
482956e kleidiai: Rework KleidiAI Build System/Integration (llama/26077)
e820c28 rpc: support apple RDMA as an RPC transport (llama/26421)
be12d39 metal : null-check buffer alloc to fix OOM crash (llama/25371)
642b5d3 ruby : Add #free method, check MemoryView strictly (#4032)
eacbd82 whisper : default-initialize whisper_mel to avoid uninitialized read (#3981)
c4ac001 parakeet : fix TDT decode by outputting raw logits from the joint graph (#4017)
9781133 talk-llama : sync llama.cpp
3680f66 pi : init
0414519 sync : ggml
d470c9d ggml : bump version to 0.22.0 (ggml/1607)
322a77c sycl : mark tq2_0 as not supported (llama/27660)
17a522a webgpu : fix handling of infinity values during ARGSORT and TOP_K (llama/27538)
fa3b87c metal : per-device tuned (Q, NE) for flash-attn vec (llama/26570)
fe52277 sync : ggml
b15d31d metal: per-op source split + parallel compile (llama/26561)
4257445 scripts : update ggml-am
1d8e052 sync : ggml
aa25d33 ggml : shorten virtual device naming in CUDA and Metal (llama/27608)
103305e webgpu : reorder includes since V that appears in common_decls.tmpl may be defined as K in flash_attn_decls.tmpl if KV_OVERLAP (llama/27545)
32d0f49 ggml : fix ggml_clamp (llama/27644)
20209c2 Deepseek 4: -sm tensor (llama/26490)
3b89b37 Fix meta tensor split st...

Note: the Windows arm64 (CUDA) build (whisper-bin-win-cuda-*-arm64.zip) uses a preview edition of the CUDA Toolkit for Windows on Arm and should be considered experimental.

Nightly build: b4938

Change log since v1.9.2

371b5a7 release : v1.9.3 (#4000)
81c1905 cmake : update semver and release process [no ci] (#3996)
4834a23 talk-llama : sync llama.cpp
6b014cf sync : ggml
8189458 ggml : bump version to 0.20.2 (ggml/1589)
51319a2 CUDA: MMVQ nwarps=8 for bs=1 for dense models on DGX Spark (llama/26843)
71759b7 cuda : skip UMA override for HIP builds (llama/27083)
964bb1b ggml : bump version to 0.20.1 (ggml/1587)
4257f47 sycl: fix thread/block count in quantized cpy kernel launches (llama/27160)
0f38613 support OP OPT_STEP_ADAMW, OPT_STEP_SGD (llama/25268)
9a0d190 vulkan: add SHMEM_STRIDE_PAD/APPLY_SLM_A_RESHAPE for coopmat1 on Intel Xe (llama/25380)
ab71410 fixed indent
9b98e57 Fixed gating logic for problematic Intel driver version
1fe009c talk-llama : fix build (#0)
733f281 sync : ggml
667da04 ggml : bump version to 0.20.0 (ggml/1584)
b7ea8b1 ggml : recurrent state rollback for ggml_ssm_scan (llama/26623)
2aef2a0 sycl: fuse mul_mat(gate) + mul_mat(up) + GLU for q4_K dense FFN (llama/26779)
5bc0c85 ggml: force single thread on wasi (llama/25686)
2d05b6e sycl: fuse the gated-delta-net state writeback cpy (llama/26643)
43cbe41 OpenVINO: Qwen3.5, memory optimization, and test-recurrent-state-rollback (llama/26952)
94d2d35 Support host pinned mem to improve SYCL Host-to-Device Memory Access (llama/26789)
34009e8 metal : add TQ2_0 support (llama/26980)
62031fe ggml-cpu/ops: vectorize flash-attention V-cache F16 to F32 conversion (llama/26947)
ac9a74f sycl: remove separate fp32 type promotion in gemm non-oneDNN path (llama/26372)
3425d13 sycl: fuse UNARY(silu|sigmoid|softplus) + MUL (llama/26411)
90d4ed1 sycl : Add DMMV ESIMD Q3_K kernel (llama/26251)
f701160 sycl : enhance concat to support Q4_0, Q4_1, Q5_0, Q5_1, Q8_0 (llama/26800)
c7be8b3 ggml-hip : remove -funsafe-math-optimizations (llama/26696)
406b116 ggml : fix arm builds, unused var (llama/26991)
b0e3297 gguf : harden loader against malformed tensor dims and metadata types (llama/25596)
92423af kleidiai: Add runtime feature detection mechanism for aarch64/kleidiai (llama/26076)
1b93067 opencl: default FA c8 cluster width to 16 on X1E (llama/26433)
6329040 vulkan: add TQ2_0 (ternary) support (llama/25850)
db3687b opencl: use flat mv q5_k when weight exceeds image1d_buffer_t limit (llama/26880)
cdbe455 CUDA: only disable CUDA graphs when mul_mat_id actually needs a stream sync (llama/26802)
030656f cuda : add warp-per-row wkv7 kernel for single-token decode (llama/26111)
d958968 llama: add default load-mode auto, which avoids mmap on iGPUs (llama/26081)
0f60f8e ggml-webgpu: fix CI errors from #25025 and #25262 (llama/26566)
a6e2630 opencl: transpose the K tile in local memory for FA prefill kernels (llama/26428)
830ec22 ggml-cpu : fix CPU affinity mask being ignored on Android (llama/26838)
b3bc904 ggml : require contiguous src for ROLL on CUDA and Metal (llama/25928)
f09a97c ggml-webgpu : refactor several wgsl files and simplify flash_attn wgsl. (llama/26134)
877761c ggml-cpu : fix missing Q5_0 dispatch in SpaceMiT backend (llama/26792)
10791af CUDA: fuse rms_norm + mul + rope (+ view + set_rows) (llama/26767)
068d3b0 CUDA: fix thread/block count in quantized cpy kernel launches (llama/26731)
eb3296f sycl: coalesce the ssm_conv window loads (llama/26612)
8a5ba01 metal : fix NORM/RMS_NORM for row lengths that leave a partial simdgroup (llama/26708)
06248ca cmake : add config version support (ggml/1582)
592feef talk-llama : sync llama.cpp
8770492 sync : ggml
84cdcad ggml : bump version to 0.19.0 (ggml/1581)
8587ad3 ggml : add aarch64 HWCAP fallbacks and fix fp16 variant detection (llama/25554)
56cb154 sycl: fix UE4M3 parsing (llama/25608)
077c5d4 sycl: *glu flat path (llama/26354)
9faa9ee sycl : Support DSv4 OPs: LIGHTNING_INDEXER,DSV4_HC_COMB,DSV4_HC_POST,DSV4_HC_PRE (llama/26568)
fb9e8ca sycl : fix error Error OP FLASH_ATTN_EXT on arc770 (llama/26441)
89d45af sycl : enhance OP set_rows to support all missed data types (llama/26515)
79ab70c cuda: fix warnings for unused variable/function (llama/26688)
5693378 metal : avoid threadgroup matrix array instantiation in kernel_lightning_indexer (llama/26646)
69bd0a9 ci : onboard AMD ROCm CI with gfx1151 fixes (llama/26544)
5a80d0a vulkan: fix submission batching size, add debug tools for diagnosing causes of DeviceLost drivers errors (llama/26371)
60894f1 mtmd/ggml: add ggml_build_forward_order (llama/26649)
8731021 vulkan backend ops: implemented GATED_LINEAR_ATTN (llama/25601)
8631825 whisper : heap out-of-bounds read in log_mel_spectrogram on very short audio (#3956)
df1547b whisper,parakeet : reject invalid n_dims in tensor header to prevent stack-buffer-overflow on malformed model files (#3957)

What's Changed

• sync : ggml by @ggerganov in #3896
• include parakeet in build-xcframework.sh by @daaku in #3899
• Fix pkgconfig configuration (Nix build failure) by @nmouha in #3894
• examples : update model names in parakeet-cli README.md [no ci] by @danbev in #3906
• examples: Fix argument flag for min speech duration in VAD by @QuantiusBenignus in #3907
• sync : ggml by @ggerganov in #3912
• examples : Improved inference performance of Android example project by @yymin1022 in #3913
• ci : set GGML_NATIVE=OFF for sanitize builds by @danbev in #3920
• whisper : make voice_length() utf-8 aware for CJK by @buxuku in #3915
• whisper : map token timestamps to original time when VAD is enabled by @buxuku in #3910
• ci, examples : add package-lock.json to examples addon node by @danbev in #3919
• whisper : expose internal VAD speech segments by @buxuku in #3916
• sync : ggml by @ggerganov in #3940
• Fix #587: Remove leading space from txt output by @JingliangGao in #3921
• ruby : add VAD speech segments API by @KitaitiMakoto in #3931
• Adding ruby bindings change log by @taf2 in #3959
• verify hparams loaded from parakeet model bin file by @bhargav191098 in #3950
• sync : ggml by @ggerganov in #3962
• Fix VAD min silence argument parsing by @chupakobra6 in #3963
• ci : set GGML_NATIVE=OFF for build-vad workflow by @danbev in #3966
• sync : ggml by @ggerganov in #3969
• release : v1.9.2 by @danbev in #3970

New Contributors

• @daaku made their first contribution in #3899
• @nmouha made their first contribution in #3894
• @QuantiusBenignus made their first contribution in #3907
• @yymin1022 made their first contribution in #3913
• @JingliangGao made their first contribution in #3921
• @bhargav191098 made their first contribution in #3950
• @chupakobra6 made their first contribution in #3963

Full Changelog: v1.9.1...v1.9.2

What's Changed

• ci : add GGML_NATIVE=OFF and GGML_BMI2=OFF to windows-blas by @danbev in #3891
• release : v1.9.1 by @danbev in #3892

Full Changelog: v1.9.0...v1.9.1

What's Changed

• ci : only trigger release jobs for tags by @danbev in #3883
• parakeet : add support for NVIDIA Parakeet by @danbev in #3735
• ruby : add support for Parakeet by @KitaitiMakoto in #3885
• release : v1.9.0 by @danbev in #3886

Full Changelog: v1.8.7...v1.9.0