This website requires JavaScript.
d77599234e
common : do not pass prompt tokens to reasoning budget sampler (#22488 )
2026-04-29 14:10:58 -05:00
41a63be28e
hexagon: make vmem and buffer-size configurable (#22487 )
2026-04-29 11:51:21 -07:00
098705a29e
CUDA: fuse SSM_CONV + ADD(bias) + SILU (#22478 )
2026-04-29 11:39:56 -07:00
683c5acb90
spec : disacard last drafted token with low prob (#22506 )
2026-04-29 17:00:00 +03:00
b1d5f5b449
sync : ggml
2026-04-29 16:43:08 +03:00
4b221b7f1e
ggml : bump version to 0.10.1 (ggml/1469)
2026-04-29 16:41:45 +03:00
59237bfbbc
webui: fix slow mic stop and WAV encode (#22480 )
2026-04-29 12:58:35 +02:00
1cbc846eba
ggml-cpu : disable tiled matmul on AIX to fix page boundary segfault (#22293 )
2026-04-29 16:02:40 +05:30
3142f1dbb9
ggml-cuda: refactor fusion code (#22468 )
2026-04-29 16:19:33 +08:00
b5c4227dc6
ggml-cpu: cmake: append xsmtvdotii march for SpacemiT IME (#22317 )
2026-04-29 15:59:21 +08:00
d6a5094004
ggml-webgpu: Fix bug in FlashAttention support check (#22492 )
2026-04-29 00:59:00 -07:00
7b95ea5d11
common: Intentionally leak logger instance to fix hanging on Windows (#22273 )
2026-04-29 16:58:43 +09:00
bdc9c743a5
ggml : add sve tuned code for gemm_q8_0_4x8_q8_0() kernel (#21916 )
2026-04-29 13:27:37 +05:30
739393beeb
TP: fix delayed AllReduce + zero-sized slices (#22489 )
2026-04-29 08:55:07 +02:00
fc2b0053ff
ggml-cuda: Repost of 21896: Blackwell native NVFP4 support (#22196 )
2026-04-28 15:47:42 -07:00
7b8443ac78
ggml-cuda: add flash-attn support for DKQ=320/DV=256 with ncols2=32 (… (#22286 )
2026-04-29 01:07:35 +05:30
5d56effdee
convert : add support for Nemotron Nano 3 Omni (#22481 )
2026-04-28 19:17:57 +02:00
52e5f0a5c1
common : re-arm reasoning budget after DONE on new <think> (#22323 )
2026-04-28 19:15:36 +02:00
f9f33654a6
vulkan: Coalesce Q4_K/Q5_K scale loads (#21751 )
2026-04-28 15:31:04 +00:00
98bb57916a
ggml-webgpu: fix buffer aliasing for ssm_scan and refactor aliasing logic (#22456 )
2026-04-28 07:27:17 -07:00
f42e29fdf1
webui: Server tools (#21237 )
2026-04-28 14:35:49 +03:00
19821178be
vulkan: add barrier after writetimestamp (#21865 )
2026-04-28 12:28:12 +02:00
698d19b93c
ggml: improve SPIR-V headers detection with __has_include (#21918 )
2026-04-28 13:19:06 +03:00
50494a2800
ggml : skip already registered backends and devices (#22296 )
2026-04-28 09:02:32 +02:00
d530d6e7a2
ggml : revert to -lm linking instead of find_library (#22355 )
2026-04-28 08:56:02 +02:00
c3e08f4700
CANN: add new ops, optimize existing ops (#21204 )
2026-04-28 14:27:22 +08:00
14e733e36f
spec : refactor params (#22397 )
2026-04-28 09:07:33 +03:00
516e8d7a8a
server: use pos_next instead of n_tokens for m-rope (#22439 )
2026-04-28 13:41:00 +08:00
434b2a1ff6
ggml-webgpu: add Q1_0 support (#22374 )
2026-04-27 15:50:59 -07:00
983ca8992e
server: (router) Forward form-data to model server (Fixes #22044 ) (#22118 )
2026-04-27 23:55:00 +02:00
665abc6097
add fast mat-vec kernels for i-quants (#22344 )
2026-04-27 08:25:45 -07:00
4414c04b9a
Additional test for common/gemma4 : handle parsing edge cases (#22420 )
2026-04-27 17:36:59 +03:00
ceaf47c4b1
fix: rpc-server cache may not work in Windows environments (#22394 )
2026-04-27 23:25:09 +09:00
42401c72b8
Fix type casting for unaccounted memory calculation (#22424 )
2026-04-27 20:31:13 +08:00
e940b3d468
download : prefer q8_0 when q4_k not available (#22428 )
2026-04-27 15:30:29 +03:00
0f1bb602dd
model : remove duplicate wo_s scale after build_attn (Qwen3, LLaMA) (#22421 )
2026-04-27 07:58:48 +00:00
d13540becd
convert : remove input_scale for dequantized fp8 modelopt (#22356 )
2026-04-27 08:45:01 +02:00
f84270ea10
ggml : use 64 bytes aligned tile buffers (#21058 )
2026-04-27 08:30:55 +02:00
5594d13224
common: fix missing exports in llama-common (#22340 )
2026-04-26 22:06:39 -07:00
f535774325
pr2wt : symlink .pi (#22386 )
2026-04-26 19:49:26 +03:00
06a811d085
add performance-portable tuning for register-tile and subgroup matmul (#22241 )
2026-04-26 09:26:28 -07:00
78433f606f
Fix recurrent state serialization for partial reads and writes (#22362 )
2026-04-26 17:04:40 +05:30
7ec36aa861
Github: set meta backend code owner (#22388 )
2026-04-26 13:34:13 +02:00
b1a5bd4e0c
CUDA: better coalesce data-access for contiguous concat (#22330 )
2026-04-26 09:21:45 +02:00
0c6ee1cade
ggml-cpu : re-enable fast gelu_quick_f16 (#22339 )
2026-04-26 08:28:14 +02:00
2dd84169d1
ggml-cpu: optimize avx2 q6_k (#22345 )
2026-04-26 06:27:50 +00:00
f454bd7eb8
opencl: add iq4_nl support (#22272 )
2026-04-25 21:21:58 -07:00
b760272f1a
hexagon: guard HMX clock request for v75+ platforms (#22377 )
2026-04-25 19:58:26 -05:00
dcad77cc3b
chat: fix handling of space in reasoning markers (#22353 )
2026-04-25 21:24:13 +02:00
98dc1418ea
spec : fix vocab compat checks (#22358 )
2026-04-25 20:11:35 +03:00
9725a313be
CUDA: reduce MMQ stream-k overhead (#22298 )
2026-04-25 14:15:03 +02:00
d1649047a3
metal : optimize Metal Tensor API usage for GGML_OP_MUL_MAT (#20962 )
2026-04-25 05:14:28 -07:00
9d34231bb8
llama-quant : default ftype param Q5_1 --> Q8_0 (#20828 )
2026-04-25 01:25:35 -05:00
8ea8fee966
gitignore : add .pi + personal SYSTEM.md (#22316 )
2026-04-25 09:20:45 +03:00
eddd7a13a5
[SYCL] Optimize Q4_0 mul_mat for Arc770, add scripts (#22291 )
2026-04-25 14:20:14 +08:00
dd2914dc81
ggml-webgpu: support for SSM_SCAN and disable set_rows error checking (#22327 )
2026-04-24 23:18:15 -07:00
0adede866d
parser: fix structured output bug (#22302 )
2026-04-24 23:19:55 +02:00
361fe72acb
Hexagon: Bump HMX Frequency to Max Corner (#22334 )
2026-04-24 15:55:17 -05:00
a702f39597
CI Snapdragon: Switch ubuntu-latest to ubuntu-slim runner (#22303 )
2026-04-24 12:21:36 -07:00
13d36cf891
ggml-webgpu: enable FLASH_ATTN_EXT on browser without subgroup matrix (#22199 )
2026-04-24 10:39:09 -07:00
f65bc34c68
hexagon: use DIRID 13 in libggml-htp.inf for modern InfVerif (#22306 )
2026-04-25 00:21:33 +08:00
15fa3c493b
metal : print GPU description (#22318 )
2026-04-24 13:56:03 +03:00
dc80c5252a
common : fix jinja warnings with clang 21 (#22313 )
2026-04-24 12:36:02 +02:00
e583f3b4f5
ggml : minor coding style (#22308 )
2026-04-24 11:02:00 +03:00
017f090442
jinja : remove unused header (#22310 )
2026-04-24 11:01:46 +03:00
ffdd983fb8
server : fix swa-full logic (#22288 )
2026-04-24 10:17:37 +03:00
793d0a7931
server: rename debug tags to match --cache-idle-slots naming (#22292 )
2026-04-24 09:28:44 +03:00
8bc492ebb4
hexagon: add SOLVE_TRI op (#21974 )
2026-04-24 09:39:13 +08:00
e5f070a1dc
fix(shader): handle the buffer aliasing for rms fuse (#22266 )
2026-04-23 19:32:59 -04:00
fa0b8a70a8
cli: Remove redundant local sampling variables (#20429 ) (#22264 )
2026-04-23 15:53:23 -07:00
5d2b52d80d
hexagon: add support for basic and extended Op profiling (#22269 )
2026-04-23 14:17:21 -07:00
187a456370
Enable testing on Snapdragon devices (#21051 )
2026-04-23 13:08:10 -07:00
185cbff6f1
server : convert_anthropic_to_oai: also copy chat_template_kwargs (#22154 )
2026-04-24 03:32:46 +09:00
c78fb909b2
server: fix heap-buffer-overflow from negative n_discard (CVE-2026-21869) (#22267 )
2026-04-23 12:39:07 -04:00
12568ca8c8
vendor : update LibreSSL to 4.3.1 (#22285 )
2026-04-23 17:45:56 +02:00
c807c6e3b0
server: (anthropic API) fix prefix caching (#21793 )
2026-04-23 08:45:02 -07:00
0949beb5a3
fix build number for sycl release (#22283 )
2026-04-23 15:38:58 +02:00
9012c50fc8
model-conversion : fix mmproj output file name [no ci] (#22274 )
2026-04-23 15:07:38 +02:00
0dd7f915fd
cli : cleanup auto-completion code (#21745 )
2026-04-23 15:03:28 +02:00
550d684bd1
server: Enable transcriptions API for LFM2-Audio (#22000 )
2026-04-23 10:47:26 +02:00
8635e221c8
metal : fix event synchronization (#22260 )
2026-04-23 08:22:49 +03:00
930e0210d1
gitignore: add AGENTS.local.md (#22246 )
2026-04-23 08:22:24 +03:00
96c1db26c4
ggml-base: use MATH_LIBRARY variable instead of hardcoded 'm' (#22239 )
2026-04-23 08:22:08 +03:00
4ead6fd957
[SYCL] Update oneapi 2025.3.3, Seperate SYCL build, release Ubuntu 24 package. (#22078 )
2026-04-23 13:21:36 +08:00
5eaee65384
convert : Handle ModelOpt produced mixed precision model during convert to GGUF (#22247 )
2026-04-23 05:19:51 +00:00
60b68a6279
sycl : fused MoE mul_mat_vec_q for TG (#21920 )
2026-04-22 23:18:56 -06:00
b76429a69c
ggml-webgpu: add support for im2col (#22259 )
2026-04-22 23:17:41 -04:00
86db42e97f
CUDA: fuse relu + sqr (#22249 )
2026-04-23 02:28:56 +00:00
6217b49583
HIP: flip GGML_HIP_GRAPHS to default on (#22254 )
2026-04-23 02:34:31 +02:00
0d0764dfd2
[WebGPU] Implement async tensor api and event api (#22099 )
2026-04-22 10:52:01 -07:00
6da7168312
ggml-webgpu: Add fused RMS_NORM + MUL (#21983 )
2026-04-23 02:51:40 +09:00
8bccdbbff9
chat: fix parallel_tool_calls default setting based on model capabilities, add tests for parallel tool calls and structured outputs (#22217 )
2026-04-22 18:10:56 +02:00
bcb5eeb645
speculative-simple : add checkpoint support (#22227 )
2026-04-22 15:44:45 +03:00
225088ea76
sycl: Improve mul_mat_id memory efficiency and add BF16 fast path (#22119 )
2026-04-22 18:02:56 +05:30
82d3f4d3b2
mtmd: also support LLAMA_ROPE_TYPE_NONE (#22242 )
2026-04-22 12:16:29 +02:00
17f6245168
server: ignore reasoning content from transcription api (#21905 )
2026-04-22 12:10:50 +02:00
7bfe60fdf9
mtmd, llama : Update HunyuanVL vision-language model support (#22037 )
2026-04-22 17:58:43 +08:00
750579ff14
common: Refactoring sampler parameters (#20429 ) (#22233 )
2026-04-22 01:40:19 -07:00
134d6e54d4
common/chat, server: refactor, move all conversion functions to common, add tests (#20690 )
2026-04-22 10:28:45 +02:00
ca7f7b7b94
ggml-webgpu(shader): support conv2d kernels. (#21964 )
2026-04-21 23:18:57 -04:00
Aldehir Rojas