This website requires JavaScript.
d80e878501
ui: restore Ctrl+B sidebar toggle shortcut (#25307 )
Pascal
2026-07-06 10:30:07 +02:00
48719618e8
scripts : use HF_TOKEN when downloading UI assets (#25280 )
Adrien Gallouët
2026-07-06 09:53:35 +02:00
d06ddd3589
ggml-hip: enable -ffast-math for HIP builds (#23862 )
a-huk
2026-07-06 09:02:26 +02:00
898b08854d
ui: fake 200 for proxy DELETE req (#25298 )
Xuan-Son Nguyen
2026-07-06 08:41:39 +02:00
72874f559c
ggml-cuda: optimize conv_transpose_1d indexing (#25310 )
adavyas
2026-07-05 20:49:06 -07:00
2da6686176
Fix stale tensor-split params for draft models (#24814 )
Al G
2026-07-05 19:39:36 +01:00
3e5036fbfb
abort if we see a multi buffer (#25276 )
Eve
2026-07-05 18:38:47 +00:00
4b2a0cdee1
ggml : fix tensor-parallel + -ncmoe crash on MoE models (#25028 )
liminfei-amd
2026-07-06 01:56:11 +08:00
7a63fdede1
ggml: Update VMM Pool allocation ggml-cuda.cu - Turing P2P access fix (fixes #24489 ) (#24491 )
Vexxie and Johannes Gäßler
2026-07-05 18:10:09 +01:00
78d2f52468
cuda : concat implementation for quantized types (#25303 )
fairydreaming and Stanisław Szymczyk
2026-07-05 17:26:24 +02:00
a4107133a6
llama : add guard for K/V rotation input when buffer is unallocated (#25215 )
liminfei-amd
2026-07-05 04:37:38 +08:00
665892536d
ui: add sync blocks so display/behavior settings can be set via --ui-config-file (#25132 )
Pascal
2026-07-04 16:12:27 +02:00
ef2d770117
ggml : fix broken CPU concat implementation for quantized types (#25247 )
fairydreaming and Stanisław Szymczyk
2026-07-04 13:37:37 +02:00
2d973636e2
chat: trim messages sent to StepFun parser (fixes long reasoning loops) (#25238 )
2026-07-03 23:12:11 +02:00
d4cff114c0
ui: Improve performance when streaming (#25225 )
Nick Towle and Pascal
2026-07-03 10:03:51 -07:00
f113e02d5a
ui: strip path and weight extension from model id in single model mode (#25137 )
Pascal
2026-07-03 17:32:48 +02:00
152d337fad
spec: support spec-draft-p-min in DFlash (#25246 )
Ruixiang Wang
2026-07-03 15:40:06 +02:00
75a48a9055
cuda: enable topk-moe fusion for 288 experts (#25267 )
Piotr Wilkin (ilintar) and Oliver Simons
2026-07-03 15:36:55 +02:00
067de93718
ui: align persisted config with strict server schema and enable thinking by default (#25242 )
Pascal
2026-07-03 13:14:52 +02:00
b5315e16e0
server + ui: ping silent SSE streams every 1s and kick only after 3s so slow prefill never drops healthy connections (#25241 )
Pascal
2026-07-03 12:47:04 +02:00
94875285e4
ui: Add MCP Servers Opt-In for first time visitors (#25239 )
Aleksander Grygier
2026-07-03 12:16:29 +02:00
5a460dea9f
Remove redundant CUDA copies after gated_delta_net. (#23940 )
Gaurav Garg
2026-07-03 14:36:29 +05:30
c8ae9a750c
vendor : update cpp-httplib to 0.49.0 (#25218 )
Alessandro de Oliveira Faria (A.K.A.CABELO)
2026-07-03 05:26:54 -03:00
fdb1db877c
llama : add llama_model_ftype_name() (#25134 )
Adrien Gallouët
2026-07-02 17:26:47 +02:00
4fc4ec5541
opencl: allow loading precompiled binary kernels from library (#23042 )
lhez
2026-07-01 10:29:22 -07:00
a6647b1a32
common : use hf primary split as model path (#25194 )
Adrien Gallouët
2026-07-01 18:33:00 +02:00
13e673863b
hexagon: flash attention rework (optimizations, accuracy improvements, etc) (#25085 )
Max Krasnyansky
2026-07-01 06:59:19 -07:00
b820cc8e6f
CUDA: consistent use of __restrict__ + PDL for FA (#25185 )
Johannes Gäßler
2026-07-01 10:55:14 +02:00
6dbc1174b8
ggml-cpu: add AVX2 optimization for nvfp4 dot product and use UE4M3 LUT (#23961 )
ragz4125
2026-07-01 13:01:20 +05:30
9d88e7cedd
ui Prevent tool messages from incorrectly appending to other conversations (#25177 )
Aleksander Grygier and Pascal
2026-07-01 09:25:18 +02:00
7af4279f45
ui: Remove PWA navigate fallback to prevent caching API endpoint requests (#25174 )
Aleksander Grygier
2026-07-01 07:32:55 +02:00
fd1a05791d
opencl: initial q1_0 support (#25160 )
lhez
2026-06-30 21:43:20 -07:00
0eca4d490e
cuda : prevent integer truncation and overflow errors when using KQ mask strides in flash_attn_mask_to_KV_max kernel (#24945 )
fairydreaming and Stanisław Szymczyk
2026-06-30 20:47:05 +02:00
4f31eedb0c
model : register t_layer_inp for qwen3next (#25141 )
Jürgen Schmied
2026-06-30 17:57:14 +02:00
799fcc04a5
common,server: handle bracketed IPv6 literals in URL authority (#25140 )
Pascal
2026-06-30 16:16:44 +02:00
931eb37f8c
CUDA: fix get_rows_back for tables with more than 65535 rows (grid-y clamp + stride) (#25103 )
Matt Jallo
2026-06-30 05:16:24 -07:00
e495d1e748
CUDA: fix Gemma E4B MTP FlashAttention (#25148 )
Johannes Gäßler
2026-06-30 14:06:54 +02:00
f708a5b2ca
vulkan: roll bk loop in matmul for asahi linux (#24663 )
Kevin Liu
2026-06-30 06:27:38 -04:00
d9df11006f
HIP: use hipBLAS for dense prefill on gfx900, keep MMQ for MoE (#24588 )
zduford
2026-06-30 05:51:38 -04:00
6c5de1cc83
ggml-webgpu: add support for NVFP4 (#25143 )
Masashi Yoshimura
2026-06-30 17:20:04 +09:00
86b94708f2
Revert "sched : reintroduce less synchronizations during split compute (#20793 )" (#25138 )
Oliver Simons
2026-06-30 02:41:45 +02:00
6f4f53f2b7
common : dedup preset and cached model entries in /v1/models (#25131 )
Adrien Gallouët
2026-06-29 17:37:23 +02:00
25a1d63f43
vulkan: use flops instead of weight tensor size for submission heuristic (#25005 )
Ruben Ortlam
2026-06-29 15:24:44 +02:00
8c146a8366
DeepSeek V4 (#24162 )
2026-06-29 16:58:51 +08:00
6cb18b2f2e
tools/ui: restore Tailwind scanning in ignored worktrees (#24879 )
seryogakovalyov
2026-06-29 11:55:52 +03:00
277a105dc8
common : remove unused regex-partial (#25118 )
o7si
2026-06-29 14:48:39 +08:00
b3fed31b99
jinja, chat: add --reasoning-preserve flag (#25105 )
Xuan-Son Nguyen
2026-06-28 23:33:51 +02:00
dbdaece23d
Revert "ui: fix accessibility for hover-gated interactive elements assisted by claude(in debugging and tests) (#24727 )" (#25098 )
Aleksander Grygier
2026-06-28 21:30:03 +02:00
7cb8576e7c
ui: fix stop and reasoning skip in single-model mode (#25084 )
Pascal
2026-06-28 21:06:43 +02:00
fa72bc6826
dflash: refactor draft model conversion (#25110 )
Ruixiang Wang
2026-06-28 20:31:48 +02:00
c818263f2a
chat : implement minicpm5 parser (#24889 )
2026-06-28 09:53:32 -05:00
f68a788b0b
jinja: add --dump-prog for debugging (#25086 )
Xuan-Son Nguyen and Sigbjørn Skjæret
2026-06-28 15:50:31 +02:00
d1b34251bc
spec : add DFlash support (#22105 )
Ruixiang Wang and Kashif Rasul
2026-06-28 15:01:34 +02:00
c1a1c8ee94
common : allow --offline in llama download (#25091 )
Adrien Gallouët
2026-06-28 12:34:11 +02:00
27c8bb4f63
logs : reduce v2 (#25078 )
Georgi Gerganov
2026-06-28 08:52:15 +03:00
ebd048fc5e
opencl: flash attention improvement (#25069 )
Hongqiang Wang and Li He
2026-06-27 15:36:06 -07:00
0ed235ea2c
[CUDA] Added a cudaMemcpy2DAsync fast path to ggml_cuda_cpy (#25057 )
Gaurav Garg
2026-06-27 17:46:21 +05:30
9bebfcb4bc
sycl : fix failed ut cases of norm (#25044 )
Neo Zhang
2026-06-27 17:13:43 +08:00
0b6529d818
vulkan: fix step operator for 0 input (#25036 )
Ruben Ortlam
2026-06-27 10:57:31 +02:00
c299a92c38
binaries : Improve rpc-server and export-graph-ops names. (#25045 )
Christian Kastner
2026-06-27 09:31:29 +02:00
0275c0f800
ci : add windows-openvino to check-release (#25022 )
Sigbjørn Skjæret
2026-06-27 09:30:56 +02:00
83d385b429
tests : fix test-chat-template --no-common option (#25075 )
Sigbjørn Skjæret
2026-06-27 09:30:19 +02:00
050ee92d04
app : allow --version, --licenses & --help (#25054 )
Adrien Gallouët
2026-06-26 23:18:11 +02:00
3fc4e10527
sched : reintroduce less synchronizations during split compute (#20793 )
2026-06-26 16:18:30 +02:00
5d8ccdf9d1
devops : add llama in all docker images (#25035 )
Adrien Gallouët
2026-06-26 15:15:48 +02:00
024930c6ad
arg: fix handling --spec-draft-hf and --hf-repo-v (#25043 )
Xuan-Son Nguyen
2026-06-26 14:36:03 +02:00
5397c36194
openvino: Update to OV 2026.2.1, self-contained release packages, operator improvements (#24974 )
2026-06-26 05:07:19 -07:00
e7ea94afcb
sync : ggml
Georgi Gerganov
2026-06-26 15:04:05 +03:00
96183e9820
ggml : bump version to 0.15.3 (ggml/1550)
Georgi Gerganov
2026-06-26 14:37:43 +03:00
487a6cc164
vulkan: opt mul_mat_vecq for mi50 (#22933 )
nullname
2026-06-26 19:49:24 +08:00
5a6a0dd7e1
vulkan: add INTEL_XE1 arch enum and enable coopmat1 on Intel Xe-LPG Plus (#24404 )
2026-06-26 11:26:22 +00:00
ded1561b42
ui: fix accessibility for hover-gated interactive elements assisted by claude(in debugging and tests) (#24727 )
Sanjay Ahari
2026-06-26 16:25:38 +05:30
9df06805ee
vulkan: Workaround compiler bug in conv2d coopmat2 path (#24924 )
Jeff Bolz
2026-06-26 04:53:32 -05:00
2f18fe13c5
CUDA: add cublasSgemmBatched mapping for HIP/MUSA vendor headers (#25033 )
leonardHONG
2026-06-26 17:42:56 +08:00
c16c35b814
ggml-cpu: fix SVE leftover path in ggml_vec_dot_f32 (#24699 )
Tarek Dakhran and Georgi Gerganov
2026-06-26 09:41:56 +02:00
1a87dcdc45
server + ui: SSE Replay Buffer (#23226 )
Pascal and Xuan Son Nguyen
2026-06-26 09:31:29 +02:00
e7e3f35090
sycl : clamp softmax input to avoid underflow (#24941 )
Jassieluo
2026-06-26 15:02:42 +08:00
b11f7c16bc
mtmd: add more validations (#25013 )
Xuan-Son Nguyen
2026-06-26 08:43:29 +02:00
f818065d75
CUDA: batch out_prod broadcast (dps2>1) path with cublasSgemmBatched (#24426 )
leonardHONG
2026-06-26 13:51:25 +08:00
960d628f46
mamba2: remove hardcoded 2x expansion factor and invalid d_inner % d_state check (#23082 )
2026-06-26 08:50:54 +03:00
5c7c22c3e1
opencl: flush profiling batch at shutdown for incomplete batches (#25016 )
shaofeiqi
2026-06-25 18:48:24 -07:00
beac5309f1
xcframework : disable mtmd video on i/tv/visionos (#25018 )
Sigbjørn Skjæret
2026-06-26 00:13:59 +02:00
9d5d882d8c
model : Add label for LFM2.5-230M (#25008 )
Tarek Dakhran
2026-06-25 18:58:52 +02:00
1ec44d178d
CUDA: Various fixes to cpy.cu (#25000 )
Oliver Simons
2026-06-25 17:29:23 +02:00
c7cddefcbd
misc: fix labeler (#25012 )
Xuan-Son Nguyen
2026-06-25 17:23:37 +02:00
e9d1b76d0a
server: use status code 403 for disabled features (#24970 )
Xuan-Son Nguyen
2026-06-25 16:36:40 +02:00
099bf06952
misc: update lables (#24920 )
Xuan-Son Nguyen
2026-06-25 16:26:56 +02:00
60bc8866b1
common: refactor model handling (#24980 )
Xuan-Son Nguyen
2026-06-25 15:17:51 +02:00
e8ecce53b8
docs : Eagle3 qwen3 draft model support (#24977 )
Kashif Rasul
2026-06-25 14:58:00 +02:00
683b04cc4a
app : add the llama download subcommand (#24982 )
Adrien Gallouët
2026-06-25 13:36:36 +02:00
f728adab68
ggml : address integer overflows in binary ops CUDA implementation (#24706 )
fairydreaming and Stanisław Szymczyk
2026-06-25 10:06:44 +02:00
3e61ea0e2f
ui: fix always-show-sidebar-on-desktop setting after navigation refactor (#24979 )
Pascal
2026-06-25 09:45:55 +02:00
fdbd6abee2
tests : synchronize contexts at end of test-thread-safety (#24935 )
Christopher Albert
2026-06-25 08:22:51 +02:00
e12a0128ab
build: include libmtmd in Apple XCFramework (#21935 )
Abraham Gonzalez and Abraham Gonzalez
2026-06-25 01:37:30 -04:00
b3ce5cedf4
quant : fix quantizing moe with mtp (#24986 )
Sigbjørn Skjæret
2026-06-25 07:36:49 +02:00
e9fb3b3fc0
sycl : support --split-mode tensor (#24152 )
David Spruill
2026-06-25 01:35:21 -04:00
9c10954865
sycl : fix the failed UT cases of conv_3d (#24900 )
Neo Zhang
2026-06-25 13:27:58 +08:00
fdb2c11c70
opencl: support non-contig rows in norm (#24965 )
lhez
2026-06-24 19:21:25 -07:00
09cedfd699
chat: harden caps check (#24973 )
Piotr Wilkin (ilintar)
2026-06-25 02:49:22 +02:00
8be759e6f7
hexagon: MUL_MAT and MUL_MAT_ID rework : 32x32 tiled weight repack, kernel-params, cached graphs (#24954 )
Max Krasnyansky
2026-06-24 12:14:25 -07:00
Pascal