This website requires JavaScript.
914dde72ba
ggml : unary ops support non-cont src0 + metal F16 unary ops (#19511 )
2026-02-11 18:58:43 +02:00
3136a849db
common : remove unused token util functions (#19506 )
2026-02-11 17:41:35 +01:00
e463bbdf65
model: Add Kimi-K2.5 support (#19170 )
2026-02-11 07:47:30 -08:00
53de59f67d
build : fix case in dSYMs path for build-macos [no ci] (#19515 )
2026-02-11 14:02:29 +01:00
9ab072ebbe
metal : extend l2_norm support for non-cont src0 (#19502 )
2026-02-11 14:53:19 +02:00
ada90bf2ba
docs: ban AI for issues and discussions [no CI] (#19512 )
2026-02-11 12:49:40 +01:00
0c1f39a9ae
common : improve download error reporting (#19491 )
2026-02-11 09:27:55 +01:00
73cd5e1b97
hexagon: Add ARGSORT, DIV, SQR, SQRT, SUM_ROWS, GEGLU (#19406 )
2026-02-10 23:21:12 -08:00
8ee538ce73
llama : correct typos 'occured' and 'occurences' (#19414 )
2026-02-11 06:05:31 +00:00
6d95707827
model : fix wavtokenizer embedding notions (#19479 )
2026-02-11 07:52:20 +02:00
89181c0b6d
ggml : extend bin bcast for permuted src1 (#19484 )
2026-02-11 07:52:00 +02:00
ceaa89b786
metal : consolidate unary ops (#19490 )
2026-02-11 07:51:12 +02:00
2cce9fddb7
llama : refactor sampling_info to use buffer_view template (#19368 )
2026-02-11 05:38:13 +01:00
612db61886
CUDA : Update CCCL-tag for 3.2 to final release from RC (#19486 )
2026-02-10 22:31:19 +01:00
57487a64c8
[WebGPU] Plug memory leaks and free resources on shutdown (#19315 )
2026-02-10 08:04:00 -08:00
fc0fe40049
models : support qwen3.5 series (#19468 )
2026-02-11 00:00:26 +08:00
9a96352729
test: fix IMROPE perf test case (#19465 )
2026-02-10 14:37:50 +01:00
c03a5a46f0
ggml-cpu: arm64: q6_K repack gemm and gemv (and generic) implementations (dotprod) (#19360 )
2026-02-10 10:47:45 +00:00
6948adc90d
ggml : use noexcept overload for is_regular_file in backend registration (#19452 )
2026-02-10 10:57:48 +01:00
854b09f0d7
convert : move experts permutation from Qwen2MoeModel to Qwen3VLMoeTextModel (#19445 )
2026-02-10 09:01:37 +01:00
66d403c480
tts : fix typos in README.md [no ci] (#19463 )
2026-02-10 07:30:41 +01:00
f0bfe54f55
CANN: Remove unnecessary wrapper for gml_backend_buft_is_cann (#18968 )
2026-02-10 06:19:30 +00:00
52e38faf8c
CANN: implement quantized MUL_MAT_ID for MoE models (#19228 )
2026-02-10 14:18:59 +08:00
a0d585537c
cuda : extend GGML_OP_PAD to work with non-cont src0 (#19429 )
2026-02-10 08:07:16 +02:00
98e57ca422
chat: fix case where template accepts type content only (#19419 )
2026-02-09 22:14:12 +01:00
262364e31d
mtmd: Implement tiling for LFM2-VL (#19454 )
2026-02-09 17:30:32 +01:00
820ebfa6f4
Server: log when converting requests to chat completions format (#19457 )
2026-02-10 00:22:57 +09:00
292f6908cd
spec : remove check rate (#19377 )
2026-02-09 14:30:50 +01:00
81ddc60cb3
ci : add metal server workflows (#19293 )
2026-02-09 15:09:30 +02:00
972f323e73
revert : "[Model] Qwen3.5 dense and MoE support (no vision) (#19435 )" (#19453 )
2026-02-09 14:57:51 +02:00
f5e7734ff2
ggml-virtgpu: add backend documentation (#19354 )
2026-02-09 13:15:42 +01:00
1e8924fd65
cmake : add variable to skip installing tests (#19370 )
2026-02-09 06:12:02 +00:00
39bf692af1
[Model] Qwen3.5 dense and MoE support (no vision) (#19435 )
2026-02-09 00:24:08 +01:00
e06088da0f
CUDA: Fix non-contig rope (#19338 )
2026-02-08 14:12:51 +01:00
5fa1c190d9
rpc : update from common.cpp (#19400 )
2026-02-08 09:06:45 +01:00
eb449cdfa4
server : improve context checkpoint logic (#19408 )
2026-02-08 09:40:04 +02:00
5999b50eb0
llama-quantize : cleanup --help output (#19317 )
2026-02-08 01:22:38 -06:00
9a5f57795c
ci : remove server job from webui and move slow test (#19424 )
2026-02-08 01:20:00 +01:00
96441c955e
ci : use -j param correctly when building with sanitizers (#19411 )
2026-02-08 00:50:47 +02:00
8872ad2125
metal : consolidate bin kernels (#19390 )
2026-02-07 10:35:56 +02:00
34ba7b5a2f
metal : fix event synchronization in cpy_tensor_async (#19402 )
2026-02-07 07:37:15 +02:00
b83111815e
model : support Step3.5-Flash (#19283 )
2026-02-07 04:06:14 +08:00
3228e77287
gguf-py : bump sentencepiece version (#19319 )
2026-02-06 15:05:19 -05:00
7fbd36c50c
ggml-webgpu: JIT compile binary operators and handle binding overlaps (#19310 )
2026-02-06 10:33:30 -08:00
537eadb1b9
sycl: add F16 support for GGML_OP_CEIL (#19306 )
2026-02-06 17:13:44 +02:00
db6adb3c88
tests: reduce number of FA test permutations (#19381 )
2026-02-06 08:50:30 -06:00
dfde5993ea
common : add common_speculative_is_compat() (#19270 )
2026-02-06 16:47:22 +02:00
06bf3796f4
unicode : MSVC regex fix (#19340 )
2026-02-06 14:56:13 +01:00
3688c4f504
Kimi-Linear support (backend agnostic + MLA KV cache) (#18755 )
2026-02-06 18:39:58 +08:00
1946e46f4c
vulkan: For coopmat2 FA, use fp16 accumulators for the final result (#19376 )
2026-02-06 02:15:13 -06:00
f9bd518a6b
vulkan: make FA mask/softcap enables spec constants (#19309 )
2026-02-06 01:49:58 -06:00
7fcf1ef45d
metal : skip loading all-zero mask (#19337 )
2026-02-06 09:25:11 +02:00
e696cfc016
llama : rename llama-sampling to llama-sampler (#19363 )
2026-02-06 07:26:54 +01:00
3e21647666
cuda : cuda graphs now compare all node params (#19383 )
2026-02-06 07:55:06 +02:00
22cae83218
metal : adaptive CPU/GPU interleave based on number of nodes (#19369 )
2026-02-05 19:07:22 +02:00
449ec2ab07
vulkan: Preprocess FA mask to detect all-neg-inf and all-zero. (#19281 )
2026-02-05 09:26:38 -06:00
3795cc1e89
benches : update models + numbers (#19359 )
2026-02-05 14:34:07 +02:00
b828e18c75
docker : fix vulkan build (#19352 )
2026-02-05 11:10:39 +01:00
a4ea7a188f
vendor : update BoringSSL to 0.20260204.0 (#19333 )
2026-02-05 09:53:35 +01:00
7a4f97d196
metal : add diag (#19330 )
2026-02-05 10:08:45 +02:00
a498c75ad1
vulkan: fix GPU deduplication logic. (#19222 )
2026-02-05 03:06:59 -05:00
3409ab842d
vulkan: Set k_load_shmem to false when K is too large (#19301 )
2026-02-05 01:48:33 -06:00
c342c3b93d
vulkan: fix non-contig rope (#19299 )
2026-02-05 01:38:59 -06:00
af252d0758
metal : add missing includes (#19348 )
2026-02-05 01:05:09 -05:00
11fb327bf3
vendor : add missing llama_add_compile_flags (#19322 )
2026-02-05 02:27:38 +01:00
e6e934c5ea
vendor: update cpp-httplib version (#19313 )
2026-02-05 05:15:03 +08:00
b536eb0233
codeowners : add danbev for examples/debug (#19332 )
2026-02-04 20:20:40 +01:00
e0c93af2a0
debug: make common_debug_print_tensor readable (#19331 )
2026-02-04 17:55:31 +01:00
423bee462b
ci : fix sanitize workflow to enable ggml sanitizers too (#19323 )
2026-02-04 15:12:03 +02:00
8abcc70a74
model: (qwen3next) correct vectorized key_gdiff calculation (#19324 )
2026-02-04 13:09:58 +01:00
eaba92c3dc
tests : add non-cont, inplace rope tests (#19296 )
2026-02-04 12:45:21 +02:00
6ab881b7c3
model-conversion : add tensor-info.py utility (#18954 )
2026-02-04 10:40:53 +01:00
d838c22bb3
spec : fix the check-rate logic of ngram-simple (#19261 )
2026-02-04 10:39:53 +02:00
25f40ca65f
completion : simplify batch (embd) processing (#19286 )
2026-02-04 05:43:28 +01:00
015deb9048
ggml-virtgpu: make the code thread safe (#19204 )
2026-02-04 03:46:18 +01:00
2ceda3f662
ggml-cpu: use LUT for converting e8->f32 scales on x86 (#19288 )
2026-02-04 09:43:29 +08:00
44008ce8f9
metal : add solve_tri (#19302 )
2026-02-03 23:43:14 +02:00
6a9bf2f788
ci : add sanitizer runs for server (#19291 )
2026-02-03 22:41:20 +02:00
faa1bc26ee
sampling : delegate input allocation to the scheduler (#19266 )
2026-02-03 22:16:16 +02:00
32b17abdb0
vulkan: disable coopmat1 fa on Nvidia Turing (#19290 )
2026-02-03 17:37:32 +01:00
8bece2eb20
CUDA: use mmvq for mul-mat-id for small batch sizes (#18958 )
2026-02-03 23:31:23 +08:00
a6fd8ca1fe
models : remove unnecessary cont in openelm (#19289 )
2026-02-03 14:20:57 +01:00
c55bce4159
metal : minor cleanup (#19251 )
2026-02-03 13:43:29 +02:00
1f1e57f2bf
CUDA: Fix loop unrolling for BW in mul_mat_q_stream_k_fixup (#19053 )
2026-02-03 11:33:14 +01:00
e9a859db3c
ggml: added cleanups in ggml_quantize_free (#19278 )
2026-02-03 08:43:39 +02:00
41e3f02647
cuda : revert CUDA_SCALE_LAUNCH_QUEUES override until investigated (#19227 )
2026-02-03 12:11:02 +05:30
1efb5f7ae1
vocab: add Falcon-H1-Tiny-Coder FIM tokens (#19249 )
2026-02-03 09:31:01 +03:00
aeb827a3cc
spec : simplify time measurement using common_time_meas (#19262 )
2026-02-03 08:20:15 +02:00
91ea44e89b
opencl: refactor some ops, concat, repeat, tanh and scale (#19226 )
2026-02-02 15:54:43 -08:00
0dfcd3b607
jinja : add missing 'in' test to template engine (#19004 ) (#19239 )
2026-02-02 12:00:55 -08:00
07a7412a3b
mtmd: add min/max pixels gguf metadata (#19273 )
2026-02-02 20:59:06 +01:00
9f682fb640
ggml-cpu: FA split across kv for faster TG (#19209 )
2026-02-03 01:19:55 +08:00
a3fa035822
server: print actual model name in 'model not found" error (#19117 )
2026-02-02 16:55:27 +01:00
15818ac44c
ci: add test-backend-ops test for CPU (#19268 )
2026-02-02 22:40:28 +08:00
bf38346d13
Remove support for Nvidia & AMD GPU, because the oneAPI plugin for Nvidia & AMD GPU is unavailable: download/installation channels are out of work. (#19246 )
2026-02-02 21:06:21 +08:00
4d5e972673
sycl: implement GGML_OP_TOP_K (#19242 )
2026-02-02 15:05:51 +02:00
6fdddb4987
metal : support virtual devices (#18919 )
2026-02-02 14:29:44 +02:00
6156ae5111
model-conversion : add debug option to conversion script (#19265 )
2026-02-02 11:29:57 +01:00
59377a6c87
ggml-backend: fix async set/get fallback sync (#19179 )
2026-02-02 10:00:05 +01:00
1239267cc4
authors : update (#19263 )
2026-02-02 08:51:25 +02:00
Georgi Gerganov
Daniel Bevenius
AesSedai
Johannes Gäßler
Adrien Gallouët
Max Krasnyansky
thecaptain789
Oliver Simons
Nikhil Jain
JJJYmmm
Xuan-Son Nguyen
Alberto Cabrera Pérez
k4ss4n
Piotr Wilkin (ilintar)
Raul Torres
hipudding
Tarek Dakhran
손희준
Sascha Rogmann
Kevin Pouget
Hugo
ddh0
Sigbjørn Skjæret
forforever73
Alex Trotta
Abhijit Ramesh
Nechama Krashinski
Jeff Bolz
Lasse Lauwerys
ymcki
Oleksandr Kuvshynov
will-lms
Aaron Teo
Aman Gupta
Ruben Ortlam
George
Gaurav Garg
Alexey Dubrov
lhez
Sid Mohan
Matthieu Coudron
Neo Zhang
Tamar