Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
246 commits
Select commit Hold shift + click to select a range
1cbfd19
mtmd: support MiMo-V2.5 audio input (RVQ-based model) (#26190)
ngxson Jul 27, 2026
91f8c9c
Disable -ffast-math on HIP (#25495)
Beinsezii Jul 27, 2026
c6292cf
contrib : add guideline about the "merge ready" label (#26178)
ggerganov Jul 28, 2026
f870678
spec: add eagle3-v3 support for gpt-oss model (#25794)
ruixiang63 Jul 28, 2026
f95de97
ggml-metal: FWHT kernel for metal backend (#25924)
nicklafleur Jul 28, 2026
9a3bf2b
server : add extra trace log for prompt similarity (#26218)
ggerganov Jul 28, 2026
d6b61ac
sycl: fix use-after-return of the SDPA scale in the oneDNN flash-atte…
meatposes Jul 28, 2026
6ba5ef2
common/chat: add specialized minimax m3 parser (#26210)
aldehir Jul 28, 2026
8407527
spec: add DSpark speculative decoding (#25173)
wjinxu Jul 28, 2026
b62b350
ggml-cuda: add chunked SSD matmul for Mamba-2 prefill acceleration (#…
BLSharda Jul 28, 2026
8161641
vulkan: add iq4_nl support back to FA (#24585)
jeffbolznv Jul 28, 2026
da5b448
ggml : set output of view src (#25729)
am17an Jul 28, 2026
ee3d1b5
server: abstract llama_memory calls to common_memory (#26221)
ngxson Jul 28, 2026
ad77bd3
docs: Adapt conda-forge package name (#26229)
jjerphan Jul 28, 2026
6e2bc65
ui: rendering performance follow-up (#26097)
allozaur Jul 28, 2026
7e1e28c
mtmd : add Nemotron 3 Nano Omni support (parakeet) (#22520)
danbev Jul 28, 2026
8190848
opencl: skip the Adreno KQ/KQV image kernels for multi-stream batches…
wanghqc Jul 28, 2026
bc71c24
ggml-webgpu: Fix some binding alias issues to support all archs, fix …
reeselevine Jul 28, 2026
e9fa078
model: Add Laguna-S-2.1 LLM_TYPE (#26233)
crusaderky Jul 28, 2026
7be2c65
model: add NextN/MTP speculative decoding support for GLM_DSA (GLM-5.…
satindergrewal Jul 29, 2026
60bccc3
add rdna3.5, and 3 to mmq configs so they can be tuned independently.…
Geramy Jul 29, 2026
f5b9bd3
RPC: add tensor_memset (#25912)
am17an Jul 29, 2026
e1af89a
conversion: fix Qwen2.5-Omni mmproj conversion regression (#26262)
kaben Jul 29, 2026
992c325
server : add trace logging for slot similarity checking (#26271)
ggerganov Jul 29, 2026
e2f59ed
vendor: update BoringSSL to 0.20260728.0 (#26241)
cabelo Jul 29, 2026
11b068d
sycl: contiguous fast path + 32-bit index math for unary elementwise …
Titaniumtown Jul 29, 2026
caa596a
ggml-cuda : disable MMQ on devices with less than 48 KiB shared memor…
KakaruHayate Jul 29, 2026
afeebe1
llama: move suppress_tokens handling to common/sampling (#26276)
ngxson Jul 29, 2026
3018a11
fix: increase greeting spacing on md screens (#26287)
allozaur Jul 29, 2026
64d528b
mimo2: address MTP review feedback (#26228)
tnhnyzc Jul 30, 2026
a6a77bc
[UT] enhance UT to show all real unsupported backends (#25234)
arthw Jul 30, 2026
32703b4
ggml : Fix issue with kleidiai ci and stringop overflow warning (#26277)
JonathanC-ARM Jul 30, 2026
21a5f5b
ui: IndexedDB and Conversations data fixes (#26278)
allozaur Jul 30, 2026
d0bfb19
metal: fix memory unwire if model is freed without any GPU operations…
nikwen Jul 30, 2026
b2f2216
Remove custom cpu op from the M3 graph, express with stock ops (#26297)
timkhronos Jul 30, 2026
9b2a088
CUDA: add Q2_0 support (#25707)
khosravipasha Jul 30, 2026
6a4c335
ggml : bump version to 0.18.0 (ggml/1576)
ggerganov Jul 30, 2026
9ebfc3a
sync : ggml
ggerganov Jul 30, 2026
6b36c23
readme : refresh (#26280)
ggerganov Jul 30, 2026
e1a1abb
ggml-cuda: Allow transpose-free gemmv computation (#26171)
roberteg16 Jul 30, 2026
47f686f
tests : avoid building get-model.cpp many times (#26317)
ggerganov Jul 30, 2026
432d7ff
llama-context : sync pending async copies before clearing embd_seq (#…
o7si Jul 30, 2026
958d9c0
Test support for alternative conv layout (#25617)
pmaybank Jul 30, 2026
ea63b4d
vulkan: Support quantized concat (#25684)
jeffbolznv Jul 30, 2026
b4ca032
server: support inp embd to generate next token (#26313)
ngxson Jul 30, 2026
5f55650
mtmd: add lanczos resize method [no release] (#26341)
ngxson Jul 30, 2026
8f4646a
ggml-webgpu: improve flash_attn_vec for quantized KV at long contexts…
yomaytk Jul 31, 2026
1553725
sycl: fuse RMS_NORM + MUL (#26015)
Titaniumtown Jul 31, 2026
a2be61d
[SYCL] Support q2 mul_mat (#26231)
arthw Jul 31, 2026
1c5b89f
sycl : support dev2dev memcpy by DEV2DEV_MEMCPY_FORWARD (#26234)
arthw Jul 31, 2026
1e22599
ggml-zendnn : group matmul direct API for mul_mat_id (#25918)
z-sachin Jul 31, 2026
69e62fc
llama : enforce the same K and V cache types for DeepSeek V4; enable …
fairydreaming Jul 31, 2026
d5d3e05
[SYCL] support the missed types in cpy (#26005)
arthw Jul 31, 2026
9d9a6d2
SYCL: add oneMKL GEMM flash attention for XMX-accelerated prompt proc…
johnkarlhill Jul 31, 2026
15e755f
cuda: extract Q2_0 elements via __byte_perm (#25603)
dfriehs Jul 31, 2026
0005475
server: correct accepted tokens when need draft token replay (#26320)
ruixiang63 Jul 31, 2026
6f3c0a7
vulkan: update vulkan sdk to 1.4.357.0 (#26303)
jeffbolznv Jul 31, 2026
82dbc4f
llama : load MTP tensors only if they are really used (#26296)
fairydreaming Jul 31, 2026
a09d8ab
Support rotated kv cache quant (#26180)
timkhronos Jul 31, 2026
db7d8b2
mtmd: add n_embd_head (#26342)
ngxson Jul 31, 2026
eb41d50
vulkan: Introduce driver version check for Windows Intel GPU to mitig…
rillomas Jul 31, 2026
876a432
vulkan: add POOL_1D op (#25431)
AnandPatil1 Jul 31, 2026
ddd4ec1
chat : enable tool call in thinking for DS4 (#26269)
pwilkin Aug 1, 2026
de69995
mtmd: add minicpmv46 downsample (#25993)
tc-mb Aug 1, 2026
c629da5
cli : persist reasoning_content in chat history (#26362)
ramicaza Aug 1, 2026
89482bd
agents: clarify comment style and jinja knowledge (#26405)
ngxson Aug 1, 2026
815a2a5
vendor : update BoringSSL to 0.20260730.0 (#26353)
cabelo Aug 1, 2026
a7a6d0d
vulkan: extend topk_moe fusion to support sqrt(softplus) (#26124)
jeffbolznv Aug 1, 2026
11924d4
test: fix some CI errors (#26415)
ngxson Aug 1, 2026
7a2db1a
ggml-webgpu: add support for f16 repeat (#26307)
yomaytk Aug 2, 2026
75587a0
model : load MiMo V2 MTP tensors only if used (#26412)
CISC Aug 2, 2026
272700b
sycl: fix classification of iGPUs (#26105)
KyleHagy Aug 2, 2026
f5919bf
chat : add qwen3 specialized parser (#26252)
aldehir Aug 2, 2026
596a579
DeepseekV4 MTP + DSpark (#25784)
am17an Aug 2, 2026
c745be2
opencl: bugfix increment ref_count in ggml_backend_opencl_init() (#2…
akleine Aug 2, 2026
3581ba0
convert: add option to create separate dspark GGUF (#26452)
am17an Aug 2, 2026
bb4e0e1
common: support the DSpark sidecar resolution (#26458)
ServeurpersoCom Aug 2, 2026
fffbcbd
metal: implement DeepSeek V4 hyper-connections (#26459)
ggerganov Aug 2, 2026
0ab9d6f
opencl: limit local workgroup size for GLU operation (#26383)
mgroeber9110 Aug 2, 2026
9d21b57
metal : add F16 support for bin ops (#26465)
ggerganov Aug 2, 2026
221f0f6
metal : add SILU_BACK (#25982)
Blackcyan30 Aug 2, 2026
1464c62
metal: implement DSv4 Lightning Indexer (#25893)
tarruda Aug 3, 2026
2b63e06
llama : MTP support for DeepSeek V3.2 (#26457)
fairydreaming Aug 3, 2026
4ed2b13
model: MTP support for Qwen3-Next (#25589)
yomaytk Aug 3, 2026
f2b52a8
server: (tools) add x-tool-cwd header (#26420)
ngxson Aug 3, 2026
0b14b87
server: add notice for upcoming default port change 8080 --> 9931 (#2…
ngxson Aug 3, 2026
9bd4c09
CUDA: Fix data-races when reusing SMEM in block_reduce (#26385)
ORippler Aug 3, 2026
96278e3
CUDA: Add backend sampler for penalties sampler (#25262)
kmorennv Aug 3, 2026
563dec8
llama : allocate indexer cache only in "full" indexer layers (#26474)
fairydreaming Aug 3, 2026
67d5978
model: M3: Move MSA into a new memory implementation (#26338)
timkhronos Aug 3, 2026
c50b34a
graph : fix unused input tensors in minimax m3 graph (#26519)
ggerganov Aug 3, 2026
39eab74
opencl: route large q6_K lm_head to the flat GEMV (#26427)
wanghqc Aug 3, 2026
dbadb68
ggml: use dynamic allocation for split graph inputs (#22789)
AgoraPete Aug 3, 2026
e8e06f7
vocab : validate default special token ids (#26506)
CISC Aug 3, 2026
99111b1
server: add get_info tool (#26522)
ngxson Aug 3, 2026
ee0445c
tests: add model resolution test on synthetic repo listings (#26172)
ServeurpersoCom Aug 3, 2026
57c0921
model : support MTP in GLM-4.7-Flash (#24868)
jacekpoplawski Aug 3, 2026
fe2adf0
vendor : update BoringSSL to 0.20260803.0 (#26523)
cabelo Aug 3, 2026
94bc47f
vendor : update cpp-httplib to 0.52.0 (#26485)
cabelo Aug 3, 2026
0ef6e55
chat : add new template for DeepSeek V4 Flash 0731 (#26398)
tarruda Aug 3, 2026
66fa168
Extended SYCL oneDNN SDPA to non-FP16 KV caches (Q4_0–Q8_0 and FP32) …
johnkarlhill Aug 4, 2026
6c8dcaa
sycl: parallelize the non-contiguous concat kernel (#25852)
Titaniumtown Aug 4, 2026
22dc605
ci: fix vulkan llvmpipe runs (#26533)
netrunnereve Aug 4, 2026
935cad6
llama : move n_vocab from llama_sampler_data to penalty_sampler (#26520)
ggerganov Aug 4, 2026
1269cb1
model : allow reshape of tensors during load (#26531)
ggerganov Aug 4, 2026
b06fbc9
convert : import bytes_to_unicode from convert_slow_tokenizer (#26217)
SolshineCode Aug 4, 2026
cf06ad7
vocab : validate plamo2 byte tokens (#26511)
CISC Aug 4, 2026
f26efa0
vulkan backend ops: implemented GATED_LINEAR_ATTN (#25601)
PranavUttarkar Aug 4, 2026
b5746d2
convert : add missing return after setting tekken vocab (#25947)
AngelGalindo7 Aug 4, 2026
15831f5
ggml : bump version to 0.18.1 (ggml/1578)
ggerganov Aug 4, 2026
2e17f69
sync : ggml
ggerganov Aug 4, 2026
5788b51
gguf-py: validate n_dims and guard against uint64 overflow in reader …
hclsys Aug 4, 2026
7bd8282
speculative : refactor enabled configs common_speculative_init (#26510)
danbev Aug 4, 2026
6b5224c
ci: fix pre-built binaries no longer working on macOS 15 and below (#…
nikwen Aug 4, 2026
1c3c967
models : fix dflash wo_a reshape on load (#26577)
ggerganov Aug 4, 2026
0713275
mtmd: support Qwen3-TTS (note: breaking change to llama-tts binary) (…
ngxson Aug 4, 2026
2f56fc3
ui: CWD for agent (#26518)
allozaur Aug 4, 2026
76c956c
gguf-split: Add option to delete split parts during merge (#26538)
guiq83 Aug 4, 2026
a6aa6f5
sampler : remove "full-context windows" from history-based samplers (…
ggerganov Aug 4, 2026
474c92e
mtmd: correcting duplicate empty audio chunks for short inputs (#26536)
Abhinaykrishna2 Aug 4, 2026
4308a4f
server: decode Windows OEM output to UTF-8 in built-in tools (#26597)
ServeurpersoCom Aug 4, 2026
6ea215d
Prefer npm ci over install for security (#26601)
eapache Aug 4, 2026
e031d95
ui: Update vulnerable packages + cleanup Storybook config (#26607)
allozaur Aug 5, 2026
d52ec04
build : remove GGML_METAL_USE_BF16 from all build scripts (#26604)
nikwen Aug 5, 2026
3e3a7a4
ui: show generation statistics by default in chat settings (#26624)
allozaur Aug 5, 2026
61881b1
vendor : apply patches for subprocess.h (#26606)
ngxson Aug 5, 2026
020760a
convert: Add endianness conversion for Q1 and TQ2 quantizations (#26618)
Andreas-Krebbel Aug 5, 2026
a035a88
server: Adding spec-decode counters to /metrics endpoint (#26389)
BLSharda Aug 5, 2026
9303cdd
security : clarify about AI-generated reports (#26579)
ngxson Aug 5, 2026
9a688e5
fit: Fix memory allocation for MTP layers (#26605)
smalinin Aug 5, 2026
717dad5
mtmd: support multi-row batching for deepseek-ocr (#26154)
ngxson Aug 5, 2026
cd0fa60
grammar : degrade max repetition >= 2000 to unbounded (#26613)
aldehir Aug 5, 2026
b06aa77
mtmd: Unlimited-OCR fix max_tiles, setting in converter (#25614)
sfallah Aug 5, 2026
360e134
tests: re-enable MiniMax M3 in `test-llama-archs` (#26633)
nikwen Aug 5, 2026
f9e832c
server: harden the file_glob_search directory walk (#26626)
ServeurpersoCom Aug 5, 2026
c8e03ce
mtmd/ggml: add ggml_build_forward_order (#26649)
ServeurpersoCom Aug 5, 2026
803b7fc
vulkan: fix submission batching size, add debug tools for diagnosing …
0cc4m Aug 6, 2026
9de0fcf
model-conversion : add --model-name to conversion scripts (#26665)
danbev Aug 6, 2026
a1f96d4
ci : onboard AMD ROCm CI with gfx1151 fixes (#26544)
jimw567 Aug 6, 2026
e700bfb
convert : accept "ExaoneMoeForCausalLM" arch spelling (#26660)
csabakecskemeti Aug 6, 2026
3db4ff8
model-loader : fix quantized reshaped tensor strides (#26672)
ggerganov Aug 6, 2026
eb5667a
convert : fix DeepseekV4 rope parameters with transformers 5.x (#26673)
CISC Aug 6, 2026
6a32c29
server: fix empty response for /cors-proxy (#26656)
ngxson Aug 6, 2026
15586e2
mtmd: add chunk save/load function (#26645)
ngxson Aug 6, 2026
e40bf88
metal : avoid `threadgroup` matrix array instantiation in kernel_ligh…
JamePeng Aug 7, 2026
d8d9887
ci: abort if build requirements are missing (#26368)
nikwen Aug 7, 2026
5b87ed3
cuda: fix warnings for unused variable/function (#26688)
dfriehs Aug 7, 2026
c074cb3
sycl : enhance OP set_rows to support all missed data types (#26515)
arthw Aug 7, 2026
eef5f3e
sycl : fix error Error OP FLASH_ATTN_EXT on arc770 (#26441)
arthw Aug 7, 2026
c1f4109
sycl : update guide Q&A and script for device setting (#26442)
arthw Aug 7, 2026
31558db
sycl : Support DSv4 OPs: LIGHTNING_INDEXER,DSV4_HC_COMB,DSV4_HC_POST,…
arthw Aug 7, 2026
6b5c2ef
sycl: *glu flat path (#26354)
Titaniumtown Aug 7, 2026
fc3f10b
sycl: fix UE4M3 parsing (#25608)
clee Aug 7, 2026
42e9881
Mitigate crashing issue on Windows MSYS2 UCRT64 environment (GCC 16.1…
rillomas Aug 7, 2026
82bb485
ui: read model modalities from the router model list (#26709)
ServeurpersoCom Aug 7, 2026
cb26014
ggml : add aarch64 HWCAP fallbacks and fix fp16 variant detection (#2…
kilian-hu Aug 7, 2026
217df17
mtmd: stop feeding the text stream again during Qwen3-TTS generation …
ServeurpersoCom Aug 7, 2026
e1470ee
server: (router) do not evict busy models (#26567)
ngxson Aug 7, 2026
dff15d4
server: (router) add LRU scheduler (#26572)
ngxson Aug 7, 2026
34e9ee5
ui: set npm `min-release-age` to protect against supply-chain attacks…
nikwen Aug 7, 2026
f4f7758
webui: load the model selected via ?model= when ?load=true (#26707)
erusev Aug 7, 2026
9ba73fd
server : clarify comment in eval_llama_cmpl_schema [no ci] [no releas…
danbev Aug 7, 2026
933f46f
ggml : bump version to 0.19.0 (ggml/1581)
ggerganov Aug 7, 2026
4cf5cab
sync : ggml
ggerganov Aug 7, 2026
4cb22cd
mtmd: fix longest_edge ignoring min/max pixels (#26638)
ngxson Aug 7, 2026
2363478
ui: Filesystem `@mentions` for Chat Form (#26715)
ServeurpersoCom Aug 7, 2026
a194a75
metal : fix NORM/RMS_NORM for row lengths that leave a partial simdgr…
robertomeroni Aug 7, 2026
f8e3026
sycl: coalesce the ssm_conv window loads (#26612)
Titaniumtown Aug 7, 2026
6de1b63
allozaur/feat/chat slash commands (#26716)
ServeurpersoCom Aug 7, 2026
1621a3d
tests : speed-up server test suite 3x (#26734)
ggerganov Aug 7, 2026
fc6545d
allozaur/feat/chat form contenteditable (#26717)
ServeurpersoCom Aug 7, 2026
3653e6d
tts: account for the vocoder pass in the timings line (#26733)
ServeurpersoCom Aug 7, 2026
69bf643
CUDA: fix thread/block count in quantized cpy kernel launches (#26731)
grafail Aug 8, 2026
dd2c7c4
server: add initial tool isolation support (via docker) (#26507)
ngxson Aug 8, 2026
18f7ad7
server, ui: only offer a working directory when a tool reads it (#26762)
ServeurpersoCom Aug 8, 2026
687e778
CUDA: fuse rms_norm + mul + rope (+ view + set_rows) (#26767)
grafail Aug 8, 2026
7ba604f
server: report the isolate working directory from get_info (#26773)
ServeurpersoCom Aug 8, 2026
61141f1
ci: rm `GGML_HIP_ROCWMMA_FATTN` (#26760)
taronaeo Aug 9, 2026
0865990
ggml-cpu : fix missing Q5_0 dispatch in SpaceMiT backend (#26792)
Hao-Chen2337 Aug 9, 2026
9369185
ci: add pr-draft-label (#26801)
ngxson Aug 9, 2026
74ce157
ui: degrade the working directory picker when file search is off (#26…
ServeurpersoCom Aug 9, 2026
f401bb1
ggml-webgpu : refactor several wgsl files and simplify flash_attn wgs…
yomaytk Aug 10, 2026
aea252f
ci: fix the ctest sanitize runs (#26593)
netrunnereve Aug 10, 2026
0377426
model-saver : fix expert shared/chunk FFN length key clobber (#26693)
SolshineCode Aug 10, 2026
1e396e7
server: gate the docker tools runtime tests on a real container run (…
ServeurpersoCom Aug 10, 2026
92d1bb0
ui: Linting & Formatting scripts (#26819)
allozaur Aug 10, 2026
6ad4ab0
readme : remove dev branches (#26832)
ggerganov Aug 10, 2026
157b81f
model : Granite-Switch Architecture (#25107)
barvhaim Aug 10, 2026
e23e944
vendor : update cpp-httplib to 0.53.0 (#26821)
cabelo Aug 10, 2026
7a20b41
model: add MTP support for Nemotron model (#26725)
ruixiang63 Aug 10, 2026
2e2d99c
ci: Add support for CUDA 13.4 ARM64 builds for Windows (#26650)
shivamkumard-ctrl Aug 10, 2026
86c298f
llama: Restore quantization of mmprojs (#26818)
pcuenca Aug 10, 2026
4c6766f
vendor: sync subprocess.h and drop local patches (#26808)
ServeurpersoCom Aug 10, 2026
a52077c
chat : Align Laguna-S-2.1 chat template to huggingface (#26232)
crusaderky Aug 10, 2026
62bf73d
model: Muse Glimmer Support (#26841)
pcuenca Aug 10, 2026
4ae84de
server: add more tool isolation support (ssh remote + podman rootless…
ServeurpersoCom Aug 10, 2026
e5275f6
ci : don't specify python version in server-sanitize for broader runn…
CISC Aug 10, 2026
4dee52f
ui: UI/chat form follow ups (#26743)
ServeurpersoCom Aug 10, 2026
f8def7f
ggml : require contiguous src for ROLL on CUDA and Metal (#25928)
devYRPauli Aug 10, 2026
d2f8305
ggml-cpu : fix CPU affinity mask being ignored on Android (#26838)
hiteshchopra11 Aug 10, 2026
dd1ea52
llama : support multi-output backend sampling (#25532)
gaugarg-nv Aug 10, 2026
0666ad2
ci : target ROCm 7.14 for build and release (#25775)
superm1 Aug 10, 2026
689e227
opencl: transpose the K tile in local memory for FA prefill kernels (…
wanghqc Aug 10, 2026
030ebb5
Address review comment of PR 25532 (#26852)
gaugarg-nv Aug 10, 2026
84f7129
ggml-webgpu: fix CI errors from #25025 and #25262 (#26566)
yomaytk Aug 11, 2026
48d22e2
common/peg : suppress incomplete escape sequences (#26780)
aldehir Aug 11, 2026
14e78dd
model : fix SWA not being enabled for EXAONE 4.5 (#26848)
junmo-kim Aug 11, 2026
4801e3c
tests : disable backend sampler hip multi output (#26878)
jimw567 Aug 11, 2026
b3df572
tests : clean-up server test, use `tests.sh` in ci (#26886)
ggerganov Aug 11, 2026
153d324
llama: add default load-mode auto, which avoids mmap on iGPUs (#26081)
0cc4m Aug 11, 2026
9afff1b
tests : fix running server tests on windows (#26889)
ggerganov Aug 11, 2026
1138b85
model-conversion : use save_output_data for causual embeddings [no ci…
danbev Aug 11, 2026
7044859
ci: hip-quality-check: update vgpr spill ignore list (#26859)
IMbackK Aug 11, 2026
8d274dd
ui: fix context gauge for single-model usage (#25738)
intel00000 Aug 11, 2026
6e62ba5
mtmd: support pocket-tts (#26871)
ngxson Aug 11, 2026
cc078b4
Dflash support for nemotron-3.5 (#26905)
lnigam Aug 11, 2026
5d16e81
convert : keep quantization scales for nemotron --mtp export (#26903)
ynankani Aug 11, 2026
2468576
requirements: use stable torch packages on s390x (#26864)
nikwen Aug 11, 2026
38406d5
imatrix.cpp: Move finite check and only check touched experts (#26861)
bartowski1182 Aug 11, 2026
70dfba5
ci : add windows-rocm to check-release (#26897)
CISC Aug 11, 2026
ba360ef
chat : tighten bare function parsing for Qwen models (#26793)
aldehir Aug 11, 2026
f785fc9
spec : update speculative-simple (#26904)
ggerganov Aug 11, 2026
5988633
cuda : add warp-per-row wkv7 kernel for single-token decode (#26111)
123123213weqw Aug 11, 2026
ebb546b
CUDA: only disable CUDA graphs when mul_mat_id actually needs a strea…
grafail Aug 11, 2026
7b13a84
ci : add missing release check (#26923)
CISC Aug 11, 2026
0b1bad1
chat : fix muse-glimmer detection of tool calls after EOM (#26879)
ruanslv Aug 11, 2026
cb27fe9
opencl: use flat mv q5_k when weight exceeds image1d_buffer_t limit (…
lhez Aug 12, 2026
6eff593
convert : handle per_layer_config in Gemma4 (transformers 5.15) (#26882)
pluvium27 Aug 12, 2026
55f453b
wavtokenizer-dec : bound posnet/convnext block_count against n_layer_…
oakkaya Aug 12, 2026
a7cd2f0
vulkan: add TQ2_0 (ternary) support (#25850)
michaeltrabalka-tech Aug 12, 2026
a4a4c51
tests : update speculative params (#26925)
ggerganov Aug 12, 2026
89e0aa6
opencl: default FA c8 cluster width to 16 on X1E (#26433)
wanghqc Aug 12, 2026
4dd1275
ui: add read_media tool (#25877)
parabelboi Aug 12, 2026
5d9e5ac
server : support slot save/restore with media inputs (#26640)
CHIPMUNK-T0T Aug 12, 2026
13fd0bb
cmake : add config version support (ggml/1582)
danbev Aug 12, 2026
af05a42
sync : ggml
ggerganov Aug 12, 2026
ece98b8
model : disallow integer dflash sliding_window_pattern (#26900)
CISC Aug 12, 2026
132753b
kleidiai: Add runtime feature detection mechanism for aarch64/kleidia…
JonathanC-ARM Aug 12, 2026
d8a8bea
gguf : harden loader against malformed tensor dims and metadata types…
harrison001 Aug 12, 2026
680a9ae
cmake : introduce semantic versioning (#26839)
danbev Aug 12, 2026
7a9ff95
disable rocm cache (#26962)
ggerganov Aug 12, 2026
9558fa4
ci : disable ubuntu-rocm (#26969)
ggerganov Aug 12, 2026
84e908c
ci: fix thread sanitizer + remove ccache (#26927)
netrunnereve Aug 12, 2026
2d1a4e8
Merge beellama-llaamacpp-base into bunch-moe-transfer
crusaderky Aug 13, 2026
0efd589
Fix merge errors
crusaderky Aug 13, 2026
44e6fb2
cosmetic
crusaderky Aug 13, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
1 change: 0 additions & 1 deletion .devops/rocm.Dockerfile
Original file line number Diff line number Diff line change
Expand Up @@ -57,7 +57,6 @@ COPY --from=web /app/tools/ui/dist tools/ui/dist
RUN HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
cmake -S . -B build \
-DGGML_HIP=ON \
-DGGML_HIP_ROCWMMA_FATTN=ON \
-DAMDGPU_TARGETS="$ROCM_DOCKER_ARCH" \
-DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON \
-DCMAKE_BUILD_TYPE=Release -DLLAMA_BUILD_TESTS=OFF \
Expand Down
27 changes: 27 additions & 0 deletions .github/actions/windows-setup-cuda/action.yml
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,10 @@ inputs:
cuda_version:
description: "CUDA toolkit version"
required: true
cuda_arch:
description: "CUDA target architecture"
required: false
default: "x64"

runs:
using: "composite"
Expand Down Expand Up @@ -127,3 +131,26 @@ runs:
echo "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
echo "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
echo "CUDA_PATH_V13_3=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.3" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8

- name: Install Cuda Toolkit 13.4 for ARM64
if: ${{ inputs.cuda_version == '13.4' && inputs.cuda_arch == 'arm64' }}
shell: pwsh
run: |
mkdir -p "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4"
choco install unzip -y
curl -O "https://packages.nvidia.com/bin-archive/pool/windows-x86_64/5B515474-7E78-11F1-8656-C51E4F4B317F/cccl-windows-x86_64-13.3.4.1.2-archive.zip"
curl -O "https://packages.nvidia.com/bin-archive/pool/windows-x86_64/5B515474-7E78-11F1-8656-C51E4F4B317F/cuda_crt-windows-x86_64-13.4.46-archive.zip"
curl -O "https://packages.nvidia.com/bin-archive/pool/windows-x86_64/5B515474-7E78-11F1-8656-C51E4F4B317F/cuda_nvcc-windows-x86_64-13.4.46-archive.zip"
curl -O "https://packages.nvidia.com/bin-archive/pool/windows-x86_64/5B515474-7E78-11F1-8656-C51E4F4B317F/libnvvm-windows-x86_64-13.4.46-archive.zip"
curl -O "https://packages.nvidia.com/bin-archive/pool/windows-arm64/5B515474-7E78-11F1-8656-C51E4F4B317F/cuda_cudart-windows-arm64-13.4.46-archive.zip"
curl -O "https://packages.nvidia.com/bin-archive/pool/windows-arm64/5B515474-7E78-11F1-8656-C51E4F4B317F/libcublas-windows-arm64-13.7.0.10-archive.zip"
unzip '*.zip' -d "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4"
xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cccl-windows-x86_64-13.3.4.1.2-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cuda_crt-windows-x86_64-13.4.46-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cuda_nvcc-windows-x86_64-13.4.46-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\libnvvm-windows-x86_64-13.4.46-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\cuda_cudart-windows-arm64-13.4.46-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
xcopy "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\libcublas-windows-arm64-13.7.0.10-archive\*" "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" /E /I /H /Y
echo "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4\bin" | Out-File -FilePath $env:GITHUB_PATH -Encoding utf8 -Append
echo "CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
echo "CUDA_PATH_V13_4=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.4" | Out-File -FilePath $env:GITHUB_ENV -Append -Encoding utf8
28 changes: 23 additions & 5 deletions .github/actions/windows-setup-rocm/action.yml
Original file line number Diff line number Diff line change
Expand Up @@ -8,8 +8,26 @@ inputs:
runs:
using: "composite"
steps:
- name: Setup ROCm
uses: ./.github/actions/install-exe
with:
url: https://download.amd.com/developer/eula/rocm-hub/AMD-Software-PRO-Edition-${{ inputs.version }}-Win11-For-HIP.exe
args: -install
- name: Install ROCm with Wheels
shell: pwsh
run: |
$ErrorActionPreference = "Stop"
write-host "Setting up Python virtual environment"

# Create the venv directly at the cache location to avoid relocation issues
New-Item -Path "C:\TheRock\build" -ItemType Directory -Force | Out-Null
python -m venv C:\TheRock\build\.venv
& C:\TheRock\build\.venv\Scripts\Activate.ps1

write-host "Upgrading pip"
python -m pip install --upgrade pip

write-host "Installing ROCm wheels for multi-arch support"
# Install ROCm wheels for multi-arch support (this may take several minutes)
python -m pip install --index-url https://repo.amd.com/rocm/whl-multi-arch/ "rocm[libraries,devel]==${{ inputs.version }}"

# Pre-expand the devel tree so it is included in the cache
write-host "Initializing ROCm devel tree"
rocm-sdk init
if ($LASTEXITCODE -ne 0) { throw "rocm-sdk init failed with exit code $LASTEXITCODE" }
write-host "Completed ROCm wheel installation to C:\TheRock\build"
8 changes: 2 additions & 6 deletions .github/workflows/build-apple.yml
Original file line number Diff line number Diff line change
Expand Up @@ -60,10 +60,10 @@ jobs:
-DCMAKE_BUILD_RPATH="@loader_path" \
-DLLAMA_FATAL_WARNINGS=ON \
-DLLAMA_BUILD_BORINGSSL=ON \
-DGGML_METAL_USE_BF16=ON \
-DGGML_METAL_EMBED_LIBRARY=OFF \
-DGGML_METAL_SHADER_DEBUG=ON \
-DGGML_RPC=ON
-DGGML_RPC=ON \
-DCMAKE_OSX_DEPLOYMENT_TARGET=13.3
time cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
leaks -atExit -- ./build/bin/test-thread-safety -hf ggml-org/gemma-3-270m-qat-GGUF -ngl 99 -p "$(printf 'hello %.0s' {1..128})" -n 16 -c 512 -ub 32 -np 2 -t 2 -lv 1
Expand Down Expand Up @@ -126,7 +126,6 @@ jobs:
run: |
sysctl -a
cmake -B build -G Xcode \
-DGGML_METAL_USE_BF16=ON \
-DGGML_METAL_EMBED_LIBRARY=ON \
-DLLAMA_OPENSSL=OFF \
-DLLAMA_BUILD_APP=OFF \
Expand Down Expand Up @@ -177,7 +176,6 @@ jobs:
run: |
sysctl -a
cmake -B build -G Xcode \
-DGGML_METAL_USE_BF16=ON \
-DGGML_METAL_EMBED_LIBRARY=ON \
-DLLAMA_BUILD_COMMON=OFF \
-DLLAMA_BUILD_APP=OFF \
Expand Down Expand Up @@ -211,7 +209,6 @@ jobs:
run: |
sysctl -a
cmake -B build -G Xcode \
-DGGML_METAL_USE_BF16=ON \
-DGGML_METAL_EMBED_LIBRARY=ON \
-DLLAMA_BUILD_COMMON=OFF \
-DLLAMA_BUILD_APP=OFF \
Expand Down Expand Up @@ -256,7 +253,6 @@ jobs:
run: |
sysctl -a
cmake -B build -G Xcode \
-DGGML_METAL_USE_BF16=ON \
-DGGML_METAL_EMBED_LIBRARY=ON \
-DLLAMA_OPENSSL=OFF \
-DLLAMA_BUILD_APP=OFF \
Expand Down
48 changes: 24 additions & 24 deletions .github/workflows/build-cache.yml
Original file line number Diff line number Diff line change
Expand Up @@ -119,27 +119,27 @@ jobs:
version_major: ${{ env.OPENVINO_VERSION_MAJOR }}
version_full: ${{ env.OPENVINO_VERSION_FULL }}

windows-2022-rocm-cache:
runs-on: windows-2022

env:
# Make sure this is in sync with build.yml
HIPSDK_INSTALLER_VERSION: "26.Q1"

steps:
- name: Clone
id: checkout
uses: actions/checkout@v6

- name: Setup Cache
uses: actions/cache@v5
id: cache-rocm
with:
path: C:\Program Files\AMD\ROCm
key: cache-gha-rocm-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ runner.os }}

- name: Setup ROCm
if: steps.cache-rocm.outputs.cache-hit != 'true'
uses: ./.github/actions/windows-setup-rocm
with:
version: ${{ env.HIPSDK_INSTALLER_VERSION }}
# windows-2022-rocm-cache:
# runs-on: windows-2022

# env:
# # Make sure this is in sync with release.yml and build-cuda-windows.yml
# ROCM_VERSION: "7.14.0"

# steps:
# - name: Clone
# id: checkout
# uses: actions/checkout@v6

# - name: Setup Cache
# uses: actions/cache@v5
# id: cache-rocm
# with:
# path: C:\TheRock\build
# key: rocm-wheels-${{ env.ROCM_VERSION }}-multi-arch-${{ runner.os }}

# - name: Setup ROCm
# if: steps.cache-rocm.outputs.cache-hit != 'true'
# uses: ./.github/actions/windows-setup-rocm
# with:
# version: ${{ env.ROCM_VERSION }}
14 changes: 10 additions & 4 deletions .github/workflows/build-cmake-pkg.yml
Original file line number Diff line number Diff line change
Expand Up @@ -5,7 +5,7 @@ on:

jobs:
linux:
runs-on: [self-hosted, Linux, CPU]
runs-on: [self-hosted, Linux]
steps:
- uses: actions/checkout@v6
with:
Expand All @@ -21,15 +21,21 @@ jobs:
-DLLAMA_BUILD_TOOLS=OFF \
-DLLAMA_BUILD_EXAMPLES=OFF \
-DLLAMA_BUILD_APP=OFF \
-DLLAMA_BUILD_IS_DEV=OFF \
-DCMAKE_BUILD_TYPE=Release
cmake --build build --config Release
cmake --build build --config Release -j $(nproc)
cmake --install build --prefix "$PREFIX" --config Release
export LLAMA_CONFIG="$PREFIX"/lib/cmake/llama/llama-config.cmake
tclsh <<'EOF'
set build(commit) [string trim [exec git rev-parse --short HEAD]]
set build(number) [string trim [exec git rev-list --count HEAD]]
set build(version) "0.0.$build(number)"
set cmakelists [read [open "CMakeLists.txt" r]]
regexp {set\(LLAMA_VERSION_MAJOR\s+(\d+)\)} $cmakelists -> major
regexp {set\(LLAMA_VERSION_MINOR\s+(\d+)\)} $cmakelists -> minor
regexp {set\(LLAMA_VERSION_PATCH\s+(\d+)\)} $cmakelists -> patch
set build(version) "$major.$minor.$patch"
set llamaconfig [read [open "$env(LLAMA_CONFIG)" r]]
set checks [list "set\\(LLAMA_VERSION \\s+$build(version)\\)" \
Expand All @@ -48,4 +54,4 @@ jobs:
cd examples/simple-cmake-pkg
cmake -S . -B build -DCMAKE_PREFIX_PATH="$PREFIX"/lib/cmake
cmake --build build
cmake --build build -j $(nproc)
6 changes: 4 additions & 2 deletions .github/workflows/build-cpu.yml
Original file line number Diff line number Diff line change
Expand Up @@ -94,8 +94,10 @@ jobs:
id: cmake_build
run: |
cmake -B build \
-DGGML_NATIVE=OFF \
-DLLAMA_FATAL_WARNINGS=ON \
-DGGML_RPC=ON
-DGGML_RPC=ON \
-DGGML_NATIVE=OFF
time cmake --build build --config Release -j $(nproc)

- name: Test
Expand All @@ -121,7 +123,7 @@ jobs:
env:
OPENBLAS_VERSION: 0.3.23
SDE_VERSION: 9.33.0-2024-01-07
VULKAN_VERSION: 1.4.313.2
VULKAN_VERSION: 1.4.357.0

strategy:
matrix:
Expand Down
1 change: 0 additions & 1 deletion .github/workflows/build-cuda-ubuntu.yml
Original file line number Diff line number Diff line change
Expand Up @@ -99,7 +99,6 @@ jobs:
run: |
cmake -B build -S . \
-DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" \
-DGGML_HIP_ROCWMMA_FATTN=ON \
-DGPU_TARGETS="gfx1030" \
-DGGML_HIP=ON
cmake --build build --config Release -j $(nproc)
Expand Down
85 changes: 50 additions & 35 deletions .github/workflows/build-cuda-windows.yml
Original file line number Diff line number Diff line change
Expand Up @@ -83,7 +83,7 @@ jobs:

env:
# Make sure this is in sync with build-cache.yml
HIPSDK_INSTALLER_VERSION: "26.Q1"
ROCM_VERSION: "7.14.0"

strategy:
matrix:
Expand All @@ -97,66 +97,81 @@ jobs:
id: checkout
uses: actions/checkout@v6

- name: Grab rocWMMA package
id: grab_rocwmma
run: |
curl -o rocwmma.deb "https://repo.radeon.com/rocm/apt/7.2.1/pool/main/r/rocwmma-dev/rocwmma-dev_2.2.0.70201-81~24.04_amd64.deb"
7z x rocwmma.deb
7z x data.tar

- name: Use ROCm Installation Cache
uses: actions/cache@v5
id: cache-rocm
with:
path: C:\Program Files\AMD\ROCm
key: cache-gha-rocm-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ runner.os }}
# - name: Cache ROCm Installation
# uses: actions/cache@v5
# id: cache-rocm
# with:
# path: C:\TheRock\build
# key: rocm-wheels-${{ env.ROCM_VERSION }}-multi-arch-${{ runner.os }}

- name: Setup ROCm
if: steps.cache-rocm.outputs.cache-hit != 'true'
# if: steps.cache-rocm.outputs.cache-hit != 'true'
uses: ./.github/actions/windows-setup-rocm
with:
version: ${{ env.HIPSDK_INSTALLER_VERSION }}
version: ${{ env.ROCM_VERSION }}

- name: Setup ROCm Environment
run: |
$ErrorActionPreference = "Stop"

# Activate venv from cache or fresh install
& C:\TheRock\build\.venv\Scripts\Activate.ps1

# Expand the devel tree (idempotent; no-op if already done during install)
rocm-sdk init
if ($LASTEXITCODE -ne 0) { throw "rocm-sdk init failed with exit code $LASTEXITCODE" }

# Get ROCm installation paths using the rocm-sdk CLI tool
$rocmPath = (rocm-sdk path --root)
if (-not $rocmPath) { throw "rocm-sdk path --root returned empty - devel package may not be installed" }
$rocmPath = $rocmPath.Trim()
$cmakePath = (rocm-sdk path --cmake).Trim()
$binPath = (rocm-sdk path --bin).Trim()
write-host "ROCm root: $rocmPath"

echo "HIP_PATH=$rocmPath" >> $env:GITHUB_ENV
echo "CMAKE_PREFIX_PATH=$cmakePath" >> $env:GITHUB_ENV
echo "HIP_DEVICE_LIB_PATH=$rocmPath\lib\llvm\amdgcn\bitcode" >> $env:GITHUB_ENV
echo "HIP_PLATFORM=amd" >> $env:GITHUB_ENV
echo "LLVM_PATH=$rocmPath\lib\llvm" >> $env:GITHUB_ENV
echo "$binPath" >> $env:GITHUB_PATH

# Keep venv in PATH for subsequent steps
echo "C:\TheRock\build\.venv\Scripts" >> $env:GITHUB_PATH

- name: Verify ROCm
id: verify
run: |
# Find and test ROCm installation
$clangPath = Get-ChildItem 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | Select-Object -First 1
if (-not $clangPath) {
Write-Error "ROCm installation not found"
exit 1
}
& $clangPath.FullName --version
# Test the ROCm clang shipped in the installed wheel
& "${env:HIP_PATH}\lib\llvm\bin\clang.exe" --version

- name: ccache
uses: ggml-org/ccache-action@v1.2.21
with:
# TODO: this build does not match the build in release.yml, so we use a different cache key
# ideally, the builds should match, similar to the CUDA build above so that we would be able
# to populate the ccache for the release with manual runs of this workflow
#key: release-windows-2022-x64-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ matrix.name }}
key: cuda-windows-2022-x64-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ matrix.name }}
#key: release-windows-2022-x64-hip-${{ env.ROCM_VERSION }}-${{ matrix.name }}
key: cuda-windows-2022-x64-hip-${{ env.ROCM_VERSION }}-${{ matrix.name }}

- name: Build
id: cmake_build
run: |
$env:HIP_PATH=$(Resolve-Path 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' | split-path | split-path)
$env:CMAKE_PREFIX_PATH="${env:HIP_PATH}"
cmake -G "Unix Makefiles" -B build -S . `
-DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" `
-DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" `
-DCMAKE_CXX_FLAGS="-I$($PWD.Path.Replace('\', '/'))/opt/rocm-7.2.1/include/" `
-DCMAKE_PREFIX_PATH="${env:HIP_PATH}" `
-DCMAKE_C_COMPILER="${env:HIP_PATH}\lib\llvm\bin\clang.exe" `
-DCMAKE_CXX_COMPILER="${env:HIP_PATH}\lib\llvm\bin\clang++.exe" `
-DCMAKE_HIP_COMPILER="${env:HIP_PATH}\lib\llvm\bin\clang.exe" `
-DCMAKE_BUILD_TYPE=Release `
-DLLAMA_BUILD_BORINGSSL=ON `
-DROCM_DIR="${env:HIP_PATH}" `
-DHIP_PATH="${env:HIP_PATH}" `
-DGGML_HIP=ON `
-DGGML_HIP_ROCWMMA_FATTN=ON `
-DGPU_TARGETS="gfx1100" `
-DGPU_TARGETS="gfx1100" `
-DGGML_RPC=ON
cmake --build build -j ${env:NUMBER_OF_PROCESSORS}

- name: ccache-clear
uses: ./.github/actions/ccache-clear
with:
#key: release-windows-2022-x64-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ matrix.name }}
key: cuda-windows-2022-x64-hip-${{ env.HIPSDK_INSTALLER_VERSION }}-${{ matrix.name }}
#key: release-windows-2022-x64-hip-${{ env.ROCM_VERSION }}-${{ matrix.name }}
key: cuda-windows-2022-x64-hip-${{ env.ROCM_VERSION }}-${{ matrix.name }}
Loading