This website requires JavaScript.
c589f0ed10
metal : add fa-vec tunings for M2 (#27940 )
codemonkey
2026-08-30 07:44:53 +08:00
c841aeeb8b
opencl: use a better matmul path on two Adreno GPU generations (#27640 )
Hongqiang Wang and Li He
2026-08-29 10:46:27 -07:00
3173a56471
metal : assert shared memory padding (#27951 )
Georgi Gerganov
2026-08-29 17:55:15 +03:00
17252c769a
metal : add remaining fa-vec tunings for M4 Pro (#27915 )
Niklas Wenzel
2026-08-29 14:50:13 +02:00
cc83d7b482
sycl: make --fit respect --fit-target better (#27629 )
Nick Farrell
2026-08-29 19:00:09 +10:00
c9ca51c1f6
vulkan: combine duplicated fastdiv functions, rename the one optimizing small divs (#27526 )
Jeff Bolz
2026-08-29 02:59:48 -05:00
5ea1b124e7
metal : add fa-vec tunings for M1 Max (#27932 )
Jhen-Jie Hong
2026-08-29 15:12:23 +08:00
77f132cb1d
vulkan: Change mul_mat_id to pad K rather than N (#27925 )
Jeff Bolz
2026-08-29 02:09:24 -05:00
d7bd3bfcad
snapdragon: python SDK setup (Windows) (#27903 )
kurquhar and Kristopher Urquhart
2026-08-28 14:01:59 -07:00
50f068ffff
bench: add --tensor-read-lazy (#27881 )
Xuan-Son Nguyen
2026-08-28 20:51:05 +02:00
6fe7498016
model: qwen4exp: reduce number of graph splits (#27880 )
Xuan-Son Nguyen
2026-08-28 19:24:46 +02:00
b387ddfd84
vulkan: fix missing view-alias dependencies in ggml_vk_graph_optimize (#27812 )
Eric A Stalee and Jeff Bolz
2026-08-28 12:12:33 -05:00
a43c3986b4
ggml : fix conv_transpose_2d for multiple batches (#26132 )
Tekin Ertekin and Georgi Gerganov
2026-08-28 20:09:08 +03:00
90c26fcd4b
Vulkan: add hoisting support for row IDs and expert count in shaders (#26686 )
ravel7524 and Jeff Bolz
2026-08-28 10:52:49 -04:00
8663224818
context : disable non-fused GDN and LID ops (#27877 )
Georgi Gerganov
2026-08-28 16:34:26 +03:00
f5e85d43a0
metal : add fa-vec tunings for M4 (#27875 )
Strongtut and Strongtut
2026-08-28 05:37:37 -07:00
511f9c1379
OpenVINO: Update OV to 2026.3.1, whisper.cpp support, Qwen3.5 on NPU, and new ops (#27843 )
2026-08-28 19:42:07 +08:00
b19cbe925b
convert: prevent ndarray conversion in LazyChunkedTensor (#27869 )
Xuan-Son Nguyen
2026-08-28 11:46:30 +02:00
d077b4c214
sycl: use TILE for quantized KV decode on BMG (#26689 )
Ozymandias_EBON
2026-08-28 03:58:58 -05:00
be876204aa
sycl: bind the f16 KV cache in place for the oneDNN SDPA path (#27468 )
Titaniumtown
2026-08-28 01:53:31 -07:00
8963a9bdcd
metal : add fa-vec tunings for M3 Max, M5 and M5 Pro (#27863 )
Georgi Gerganov
2026-08-28 11:52:03 +03:00
6d6b697cd5
metal : add fa-vec tunings for M4 Pro (#27824 )
Brad Smith
2026-08-28 04:37:43 -04:00
4e97ac86eb
tests : run test-save-load-state across all architectures (#27755 )
Georgi Gerganov
2026-08-28 09:45:19 +03:00
ca3d5a3e10
model: add DSpark support for Nemotron3.5 (#27804 )
2026-08-28 01:49:27 +02:00
e70802a01f
ggml-hexagon: add HTP unary ops for ABS and LOG (#27786 )
cqderek
2026-08-28 06:05:57 +08:00
83d855c5a6
hex-unary: fix RMS_NORM_MUL weight-offset bugs for grouped/broadcast norms (#27798 )
Aparna M P
2026-08-28 03:08:02 +05:30
18443257a3
server: add ctx-per-slot (--kv-unified-per-slot) (#24124 )
Bartowski and Xuan Son Nguyen
2026-08-27 16:39:14 -04:00
32176338a6
ci : build only the ggml-hip backend for windows-rocm release (#27753 )
Harkirat Gill
2026-08-27 16:18:34 -04:00
6c84c7d5d8
model: add Qwen3.8-Flash-Next (qwen4exp) (#27742 )
2026-08-27 12:32:31 -07:00
6fdd0ac890
ci : bundle HIP runtime DLLs with Windows ROCm release (#26973 )
Slobodan Josic
2026-08-27 19:27:57 +02:00
b10f9ca58c
spec : add DFlash2 support (local convolution + candidate selector) (#27342 ) (#27816 )
2026-08-27 19:17:07 +02:00
58546250cf
opencl: add bin kernels kernel_gemm_moe_q4_0_q8_1_dp4a_bin, kernel_gemm_moe_mxfp4_q8_1_dp4a_bin (#27768 )
Shawn Gu
2026-08-27 09:44:05 -07:00
732707dff2
quantize: cap working memory size to avoid loading big tensors onto RAM (#27795 )
Xuan-Son Nguyen
2026-08-27 18:31:13 +02:00
cb300598d5
Feature: Added LIGHTNING_INDEXER support for Deepseek V4 ops on Vulkan Backend (#27453 )
Shobhit
2026-08-27 21:34:42 +08:00
1a946ec745
pr2wt : use ssh/https remote in worktree depending on base (#27800 )
Sigbjørn Skjæret
2026-08-27 15:27:17 +02:00
fac889fb38
llama: model_loader: add TENSOR_READ_LAZY (#27794 )
Xuan-Son Nguyen
2026-08-27 15:14:34 +02:00
cae63579b6
ui: Improve Chat Form Actions UI/UX (models selector, add panel) (#27746 )
Aleksander Grygier
2026-08-27 14:47:36 +02:00
bcb6084a4e
convert : fix Nemotron-H LoRA GGUF conversion (#27356 )
Kartik Gulia
2026-08-27 18:11:24 +05:30
fe235f4343
ui: Replace per-conversation MCP overrides with per-conversation tool policy (#27745 )
Aleksander Grygier
2026-08-27 13:08:01 +02:00
2bb9bddafa
spec: Add benchmark-only synthetic speculative acceptance options (#27711 )
Gaurav Garg
2026-08-27 16:23:42 +05:30
deae5ee133
model : simplify MiniMax-01 graph (#27790 )
fairydreaming and Stanisław Szymczyk
2026-08-27 12:27:52 +02:00
f29551215b
args: add --video-* CLI arguments (#24318 )
Xuan-Son Nguyen
2026-08-27 12:11:12 +02:00
915dc6d38c
metal : fix memory leaks due to missing autoreleasepools (#27758 )
Niklas Wenzel
2026-08-27 11:53:08 +02:00
c5fc7e3488
llama : add --n-cpu-ffn option (#26622 )
Jonas J
2026-08-27 12:26:42 +03:00
d7a2074112
models : support nanbeige4.2-3B (#27730 )
zql and admin
2026-08-27 12:55:31 +08:00
192067b72d
hexagon: support for multi-NPU devices (IQ9, IQ10) and fully asynchronous backend (#26501 )
Max Krasnyansky
2026-08-26 18:46:50 -07:00
925e117994
llama: add token ID tracking to KV cell (#27762 )
Xuan-Son Nguyen
2026-08-26 23:34:28 +02:00
539f24529b
ui: Move Settings and MCP Servers routes to dialog-based views (#27744 )
Aleksander Grygier
2026-08-26 21:07:24 +02:00
0379a19f09
ui: Update Dialog component styling (#27743 )
Aleksander Grygier
2026-08-26 20:19:19 +02:00
5e6a37cb11
vulkan: warptiles currently assume warp sizes <= 64, clamp to work around larger warps (#27726 )
Ruben Ortlam
2026-08-26 18:02:06 +02:00
bf94216469
Implemented vulkan cross_entropy_loss and cross_entropy_loss_back (#27216 )
Pranav Uttarkar
2026-08-26 09:49:32 -05:00
d0132a680a
rpc : implement event and async backend APIs (#18626 )
Radoslav Gerganov
2026-08-26 17:34:46 +03:00
4d19b28769
ci: Clean up UI builds from releases (#27706 )
Aleksander Grygier
2026-08-26 14:12:09 +02:00
fc35562ba4
cuda: unblock mmq for MoE on sm_60 (#26264 )
David Friehs
2026-08-26 12:35:54 +02:00
da9b5d68c3
ci : make cache bucket public (#27728 )
Sigbjørn Skjæret
2026-08-26 12:08:23 +02:00
dac869b0a0
conversion : fix Nemotron 3.5 Lightning layers (#27729 )
Daniel Bevenius
2026-08-26 12:05:31 +02:00
11cd988428
ggml-metal: add chunked SSD MMA for Mamba-2 prefill optimization (#26647 )
2026-08-26 01:57:07 -07:00
5d5cb4c3a4
ggml-meta: propagate buffer usage and call init on the new tensors (#27586 )
Max Krasnyansky
2026-08-25 22:27:51 -07:00
d222767c7a
kleidiai: Rework KleidiAI Build System/Integration (#26077 )
Jonathan Clohessy
2026-08-25 22:07:29 +01:00
eab8ee41f8
ci : update OS used for ROCM to Ubuntu 24.04 (#27681 )
Mario Limonciello
2026-08-25 12:14:16 -05:00
b114b47397
rpc: support apple RDMA as an RPC transport (#26421 )
Ryan C and Ryan Churaman
2026-08-25 17:12:15 +00:00
0a5ac49bce
devops: use GGML_NATIVE=OFF for OpenVINO (#27338 )
krzsztf and Sigbjørn Skjæret
2026-08-25 18:37:11 +02:00
1729ed5371
server : reject prefilled assistant messages with tool calls (#27626 )
Kyozzz
2026-08-25 22:35:22 +08:00
0cc5b14959
chat : scope qwen3-coder workarounds (#27679 )
Aldehir Rojas
2026-08-25 09:33:00 -05:00
790b5713ca
ci : store ccache on HF buckets (test with cuda-ubuntu for now) (#27699 )
Sigbjørn Skjæret
2026-08-25 16:28:25 +02:00
f1357e4998
ui: ESLint config updates (#27700 )
Aleksander Grygier
2026-08-25 14:34:34 +02:00
3737e41370
metal : null-check buffer alloc to fix OOM crash (#25371 )
Yuri Khrustalev and Georgi Gerganov
2026-08-25 07:35:39 -04:00
c1d0e7a004
llama.cpp : bump version to 0.3.0 (#27696 )
Georgi Gerganov
2026-08-25 12:42:21 +03:00
81191affa5
sync : ggml
Georgi Gerganov
2026-08-25 11:46:33 +03:00
93882361f1
ggml : bump version to 0.22.0 (ggml/1607)
Georgi Gerganov
2026-08-25 11:27:17 +03:00
eb25b7263e
grammar : parse \- in char classes as literal hyphen (#27591 )
Saad Ali
2026-08-25 11:05:24 +05:00
814d84bc9d
sycl : mark tq2_0 as not supported (#27660 )
Neo Zhang
2026-08-25 14:04:58 +08:00
5ea87ddad2
webgpu : fix handling of infinity values during ARGSORT and TOP_K (#27538 )
fairydreaming and Stanisław Szymczyk
2026-08-25 07:08:06 +02:00
f280b26983
metal : per-device tuned (Q, NE) for flash-attn vec (#26570 )
YiChen Lv and Georgi Gerganov
2026-08-25 00:22:27 +08:00
b615f5b4bd
metal: per-op source split + parallel compile (#26561 )
2026-08-25 00:16:13 +08:00
b3c3b96a13
misc : read repetition_penalty from generation_config.json (#27659 )
Tarek Dakhran
2026-08-24 16:01:35 +02:00
7584430716
tests : disable DOTS3NOTE arch test for WebGPU (#27654 )
fairydreaming and Stanisław Szymczyk
2026-08-24 12:39:31 +02:00
71cc86fa41
convert: fix GLM regression in index_tensors (#27655 )
jacekpoplawski
2026-08-24 12:21:00 +02:00
a14dba686a
ggml : shorten virtual device naming in CUDA and Metal (#27608 )
Georgi Gerganov
2026-08-24 12:35:08 +03:00
c1c766da59
webgpu : reorder includes since V that appears in common_decls.tmpl may be defined as K in flash_attn_decls.tmpl if KV_OVERLAP (#27545 )
fairydreaming and Stanisław Szymczyk
2026-08-24 10:07:12 +02:00
160c6b0bdd
mtmd: video: fix moov atom at the end of file (#27596 )
Xuan-Son Nguyen and rkfg
2026-08-24 09:59:04 +02:00
985b14912b
ci : apply ccache-clear with older/min/dry-run to all ccache jobs (#27602 )
Georgi Gerganov
2026-08-24 10:49:20 +03:00
6036c635e2
ggml : fix ggml_clamp (#27644 )
Georgi Gerganov
2026-08-24 10:43:04 +03:00
a130532ae1
mamba2 : Flatten in/out projections to dispatch GEMM instead of GEMV (#27513 )
Prabhsimran Singh
2026-08-24 11:55:11 +05:30
bf0a29cc16
Deepseek 4: -sm tensor (#26490 )
Aman Gupta
2026-08-24 11:50:25 +05:30
c060ca974c
model : support MTP in GLM-4.5-Air (#26534 )
jacekpoplawski
2026-08-23 20:20:44 +02:00
ccc8fd2baa
readme : update links (#27617 )
Georgi Gerganov
2026-08-23 20:55:56 +03:00
d05f89562d
fix: Change chat tabs nav shortcuts (#27609 )
Aleksander Grygier
2026-08-23 19:37:19 +02:00
8d9af25633
test : fix multi-GPU server tests (#27614 )
Georgi Gerganov
2026-08-23 19:59:42 +03:00
4a08fa2970
test: move tools/parser to tests (#27548 )
Xuan-Son Nguyen
2026-08-23 18:38:51 +02:00
56db501e73
mtmd: use pillow-accurate algo, correct resize_algo for all models (#27594 )
Xuan-Son Nguyen
2026-08-23 18:35:41 +02:00
95b8e33e16
ci : add test-llama-archs tensor split for Metal (#27598 )
Georgi Gerganov
2026-08-23 15:57:07 +03:00
a278dcef04
contrib : recommend waiting for CI before merging (#27603 )
Niklas Wenzel
2026-08-23 14:56:47 +02:00
e8eed4525a
server : add LLAMA_SERVER_SLOTS_N_DIFF (#27600 )
Georgi Gerganov
2026-08-23 15:55:51 +03:00
ba8e0eddfb
common : skip device_info loop if it's not going to be printed (#26692 )
Bartosz Taudul
2026-08-23 14:39:16 +02:00
b0539c43ed
DeepseekV4: fix rollback with multi-seq (#26756 )
Aman Gupta and Georgi Gerganov
2026-08-23 16:27:49 +05:30
d3371929bb
[Tensor parallel] Fix meta tensor split state propagation (#27574 )
Gaurav Garg
2026-08-23 16:19:12 +05:30
8144f3192e
ui: Chat Conversation Tabbed navigation (#27263 )
Aleksander Grygier and Pascal
2026-08-23 10:46:49 +02:00
6657ded4fa
vendor : update subprocess.h (#27409 )
Alessandro de Oliveira Faria (A.K.A.CABELO)
2026-08-23 04:38:29 -03:00
29ea9412a6
cuda : add POOL_1D support (#27573 )
Aman Karki
2026-08-23 13:07:32 +05:30
codemonkey