Cherry pick 20250224#7

Merged

arthw merged 403 commits intomasterfrom

cherry_pick_20250224

Feb 27, 2025

This pull request is big! We're only showing the most recent 250 commits

Commits on Feb 25, 2025

linenoise.cpp refactoring (#11301)

ericcurtin
authored and
NeoZhangJianyu
committed
rpc : better caching of the base buffer pointer (#11331)

rgerganov
authored and
NeoZhangJianyu
committed
export-lora : fix tok_embd tensor (#11330)

ngxson
authored and
NeoZhangJianyu
committed
Add Jinja template support (#11016)

authored and
NeoZhangJianyu
committed
llava : support Minicpm-omni (#11289)

tc-mb
authored and
NeoZhangJianyu
committed
`common`: utils to split / join / repeat strings (from json converter) (#11342)

ochafik
authored and
NeoZhangJianyu
committed
Adding logprobs to /v1/completions (#11344)

jpodivin
authored and
NeoZhangJianyu
committed
`minja`: sync at https://github.com/google/minja/commit/0f5f7f2b3770eb682fbc11763266d45204173686 (#11352)

ochafik
authored and
NeoZhangJianyu
committed
server : fix draft context not being released (#11354)

slaren
authored and
NeoZhangJianyu
committed
readme : add plugin links (#11355)

ggerganov
authored and
NeoZhangJianyu
committed
main : update README documentation for batch size (#11353)

slaren
authored and
NeoZhangJianyu
committed
vulkan: fix diag_mask_inf (#11323)

jeffbolznv
authored and
NeoZhangJianyu
committed
vulkan: sort shaders for more deterministic binary (#11315)

jeffbolznv
authored and
NeoZhangJianyu
committed
Vulkan-run-test: fix mmq_wg_denoms (#11343)

AMD-dwang
authored and
NeoZhangJianyu
committed
Treat hf.co/ prefix the same as hf:// (#11350)

ericcurtin
authored and
NeoZhangJianyu
committed
server : add more clean up when cancel_tasks is called (#11340)

ngxson
authored and
NeoZhangJianyu
committed
Add -ngl (#11372)

ericcurtin
authored and
NeoZhangJianyu
committed
Update documentation (#11373)

ericcurtin
authored and
NeoZhangJianyu
committed
tests: fix some mul_mat test gaps (#11375)

jeffbolznv
authored and
NeoZhangJianyu
committed
server : (webui) put DeepSeek R1 CoT in a collapsible <details> element (#11364)

authored and
NeoZhangJianyu
committed
Update llama-run README.md (#11386)

ericcurtin
authored and
NeoZhangJianyu
committed
cmake : avoid -march=native when reproducible build is wanted (#11366)

bmwiedemann
authored and
NeoZhangJianyu
committed
CPU/CUDA: fix (GQA) mul mat back, add CUDA support (#11380)

JohannesGaessler
authored and
NeoZhangJianyu
committed
docs : Update readme to build targets for local docker build (#11368)

JafarAbdi
authored and
NeoZhangJianyu
committed
release : pack /lib in the packages (#11392)

ggerganov
authored and
NeoZhangJianyu
committed
rocBLAS: Avoid fp32->fp16->fp32 conversion on cdna (#11356)

IMbackK
authored and
NeoZhangJianyu
committed
CUDA: fix FP16 cuBLAS GEMM (#11396)

JohannesGaessler
authored and
NeoZhangJianyu
committed
hip : Add hipGraph and VMM support to ROCM (#11362)

IMbackK
authored and
NeoZhangJianyu
committed
CANN: Add Ascend CANN build ci (#10217)

authored and
NeoZhangJianyu
committed
ci : fix line breaks on windows builds (#11409)

ggerganov
authored and
NeoZhangJianyu
committed
docker : fix CPU ARM build (#11403)

slaren
authored and
NeoZhangJianyu
committed
server : fix cleaning up stream task (#11418)

ngxson
authored and
NeoZhangJianyu
committed
docker : add GGML_CPU_ARM_ARCH arg to select ARM architecture to build for (#11419)

slaren
authored and
NeoZhangJianyu
committed
build: add /bigobj to MSVC build (#11407)

jeffbolznv
authored and
NeoZhangJianyu
committed
Hip: disable VMM on hip as it seams that it dosent work in some configurations (#11420)

IMbackK
authored and
NeoZhangJianyu
committed
vulkan: compile shaders on-demand (#11406)

jeffbolznv
authored and
NeoZhangJianyu
committed
build: apply MSVC /bigobj option to c/cpp files only (#11423)

jeffbolznv
authored and
NeoZhangJianyu
committed
readme : update hot topics

ggerganov
authored and
NeoZhangJianyu
committed
rpc: fix register position (#11424)

thxCode
authored and
NeoZhangJianyu
committed
cmake: add ggml find package (#11369)

authored and
NeoZhangJianyu
committed
docker: add missing vulkan library to base layer and update to 24.04 (#11422)

rare-magma
authored and
NeoZhangJianyu
committed
metal : use residency sets (#11427)

ggerganov
authored and
NeoZhangJianyu
committed
docker : fix ARM build and Vulkan build (#11434)

ngxson
authored and
NeoZhangJianyu
committed
metal: Handle null returned from MTLCreateSystemDefaultDevice() (#11441)

Ihar Hrachyshka
authored and
NeoZhangJianyu
committed
llama: refactor llama_decode_impl (#11381)

JohannesGaessler
authored and
NeoZhangJianyu
committed
llama : minor fixes for up llama load model speed (#11448)

authored and
NeoZhangJianyu
committed
AMD: parse the architecture as supplied by gcnArchName (#11244)

Haus1
authored and
NeoZhangJianyu
committed
Add new hf protocol for ollama (#11449)

ericcurtin
authored and
NeoZhangJianyu
committed
Handle missing model in CLI parameters for llama-run (#11399)

engelmi
authored and
NeoZhangJianyu
committed
SYCL : SOFTMAX F16 mask support and other fixes (#11261)

qnixsynapse
authored and
NeoZhangJianyu
committed
docker: add perplexity and bench commands to full image (#11438)

rare-magma
authored and
NeoZhangJianyu
committed
cmake : don't fail on `GGML_CPU=OFF` (#11457)

someone13574
authored and
NeoZhangJianyu
committed
docker: allow installing pip packages system-wide (#11437)

rare-magma
authored and
NeoZhangJianyu
committed
Add github protocol pulling and http:// (#11465)

ericcurtin
authored and
NeoZhangJianyu
committed
HIP: Only call rocblas_initialize on rocblas versions with the multiple instantation bug (#11080)

sARY77
authored and
NeoZhangJianyu
committed
HIP: Supress transformation warning in softmax.cu

IMbackK
authored and
NeoZhangJianyu
committed
ci : fix build CPU arm64 (#11472)

ngxson
authored and
NeoZhangJianyu
committed
server : Fixed wrong function name in llamacpp server unit test (#11473)

peidaqi
authored and
NeoZhangJianyu
committed
cmake: add hints for locating ggml on Windows using Llama find-package (#11466)

Emreerdog
authored and
NeoZhangJianyu
committed
llama: fix missing k_cache store for rwkv6qwen2 (#11445)

MollySophia
authored and
NeoZhangJianyu
committed
embedding : enable --no-warmup option (#11475)

danbev
authored and
NeoZhangJianyu
committed
ggml-cpu : fix ggml_graph_compute_thread did not terminate on abort. (ggml/1065)

authored and
NeoZhangJianyu
committed
ggml : add option to not print stack on abort (ggml/1081)

authored and
NeoZhangJianyu
committed
sync : ggml

ggerganov
authored and
NeoZhangJianyu
committed
Parse https://ollama.com/library/ syntax (#11480)

ericcurtin
authored and
NeoZhangJianyu
committed
vulkan: Catch pipeline creation failure and print an error message (#11436)

jeffbolznv
authored and
NeoZhangJianyu
committed
server : update auto gen files comments [no ci] (#11484)

danbev
authored and
NeoZhangJianyu
committed
vulkan: implement initial support for IQ2 and IQ3 quantizations (#11360)

authored and
NeoZhangJianyu
committed
server : add /apply-template endpoint for additional use cases of Minja functionality (#11489)

pnb
authored and
NeoZhangJianyu
committed
server : update json snippets in README.md [no ci] (#11492)

danbev
authored and
NeoZhangJianyu
committed
readme : reference examples relative links (#11505)

guspan-tanadi
authored and
NeoZhangJianyu
committed
server : (docs) added response format for /apply-template [no ci] (#11503)

isaac-mcfadyen
authored and
NeoZhangJianyu
committed
server : use lambda instead of std::bind (#11507)

danbev
authored and
NeoZhangJianyu
committed
vocab : correctly identify LF token for GPT-2 style BPE tokenizer (#11496)

mgroeber9110
authored and
NeoZhangJianyu
committed
sync: minja (#11499)

ochafik
authored and
NeoZhangJianyu
committed
CUDA/HIP: add warp_size to cuda_device_info

IMbackK
authored and
NeoZhangJianyu
committed
HIP: Prepare reduction operators for wave 64

IMbackK
authored and
NeoZhangJianyu
committed
HIP: require at least HIP 5.5

IMbackK
authored and
NeoZhangJianyu
committed
Tool call support (generic + native for Llama, Functionary, Hermes, Mistral, Firefunction, DeepSeek) w/ lazy grammars (#9639)

authored and
NeoZhangJianyu
committed
`ci`: ccache for all github worfklows (#11516)

ochafik
authored and
NeoZhangJianyu
committed
server : update help metrics processing/deferred (#11512)

danbev
authored and
NeoZhangJianyu
committed
common: Add missing va_end (#11529)

stevegrubb
authored and
NeoZhangJianyu
committed
server : fix --jinja when there's no tools or schema (typo was forcing JSON) (#11531)

ochafik
authored and
NeoZhangJianyu
committed
Fix chatml fallback for unsupported builtin templates (when --jinja not enabled) (#11533)

ochafik
authored and
NeoZhangJianyu
committed
fix stop regression (#11543)

ochafik
authored and
NeoZhangJianyu
committed
`tool-call`: fix llama 3.x and functionary 3.2, play nice w/ pydantic_ai package, update readme (#11539)

authored and
NeoZhangJianyu
committed
`ci`: use sccache on windows instead of ccache (#11545)

ochafik
authored and
NeoZhangJianyu
committed
ci: simplify cmake build commands (#11548)

ochafik
authored and
NeoZhangJianyu
committed
Implement s3:// protocol (#11511)

ericcurtin
authored and
NeoZhangJianyu
committed
`sync`: minja (https://github.com/google/minja/commit/418a2364b56dc9be4ed9a1a2b0fb16fb53a7a22e) (#11574)

ochafik
authored and
NeoZhangJianyu
committed
ci: use sccache on windows HIP jobs (#11553)

ochafik
authored and
NeoZhangJianyu
committed
llama : add support for GLM-Edge and GLM-Edge-V series models (#10573)

authored and
NeoZhangJianyu
committed
sampling : support for llguidance grammars (#10224)

mmoskal
authored and
NeoZhangJianyu
committed
Fix exotic ci env that lacks ostringstream::str (#11581)

ochafik
authored and
NeoZhangJianyu
committed
`tool-call`: support Command R7B (+ return tool_plan "thoughts" in API) (#11585)

ochafik
authored and
NeoZhangJianyu
committed
Name colors (#11573)

ericcurtin
authored and
NeoZhangJianyu
committed
CUDA: use mma PTX instructions for FlashAttention (#11583)

authored and
NeoZhangJianyu
committed
nit: more informative crash when grammar sampler fails (#11593)

ochafik
authored and
NeoZhangJianyu
committed
HIP: add GGML_CUDA_CC_IS_* for amd familys as increasing cc archtectures for amd gpus are not supersets of eatch other (#11601)

IMbackK
authored and
NeoZhangJianyu
committed
CUDA/HIP: add support for selectable warp size to mmv (#11519)

IMbackK
authored and
NeoZhangJianyu
committed
HIP: fix flash_attn_stream_k_fixup warning (#11604)

JohannesGaessler
authored and
NeoZhangJianyu
committed
server : (webui) Fix Shift+Enter handling (#11609)

authored and
NeoZhangJianyu
committed
CUDA: fix Volta FlashAttention logic (#11615)

JohannesGaessler
authored and
NeoZhangJianyu
committed
sync : ggml

ggerganov
authored and
NeoZhangJianyu
committed
server : remove CPPHTTPLIB_NO_EXCEPTIONS define (#11622)

danbev
authored and
NeoZhangJianyu
committed
server : (webui) allow typing and submitting during llm response (#11626)

woof-dog
authored and
NeoZhangJianyu
committed
server : (webui) revert hacky solution from #11626 (#11634)

ngxson
authored and
NeoZhangJianyu
committed
`tool-call`: allow `--chat-template chatml` w/ `--jinja`, default to chatml upon parsing issue, avoid double bos (#11616)

ochafik
authored and
NeoZhangJianyu
committed
ci : do not stale-close roadmap issues

ggerganov
authored and
NeoZhangJianyu
committed
cmake: Add ability to pass in GGML_BUILD_NUMBER (ggml/1096)

ckastner
authored and
NeoZhangJianyu
committed
sync : ggml

ggerganov
authored and
NeoZhangJianyu
committed
authors : update

ggerganov
authored and
NeoZhangJianyu
committed
metal : use residency set for other platforms (#11648)

jhen0409
authored and
NeoZhangJianyu
committed
swift : fix llama-vocab api usage (#11645)

jhen0409
authored and
NeoZhangJianyu
committed
readme : add llm_client Rust crate to readme bindings (#11628)

ShelbyJenkins
authored and
NeoZhangJianyu
committed
`tool-call`: command r7b fix for normal responses (#11608)

ochafik
authored and
NeoZhangJianyu
committed
arg : list RPC devices first when using --list-devices (#11655)

rgerganov
authored and
NeoZhangJianyu
committed
server : add try..catch to places not covered by set_exception_handler (#11620)

ngxson
authored and
NeoZhangJianyu
committed
HIP: force max threads per block to be 1024 (#11621)

fxzjshm
authored and
NeoZhangJianyu
committed
CUDA: non-contiguous (RMS) norm support (#11659)

authored and
NeoZhangJianyu
committed
`sync`: minja (#11641)

ochafik
authored and
NeoZhangJianyu
committed
llava: add quantization for the visual projector LLAVA, Qwen2VL (#11644)

samkoesnadi
authored and
NeoZhangJianyu
committed
CUDA: support for mat. mul. with ne03 != ne13 (#11656)

JohannesGaessler
authored and
NeoZhangJianyu
committed
metal : adjust support conditions for norm operators (#11671)

ggerganov
authored and
NeoZhangJianyu
committed
readme : add link to Autopen under UIs (#11684)

blackhole89
authored and
NeoZhangJianyu
committed
metal : avoid breaking build when metal API predates TARGET_OS_VISION (#11690)

charles-dyfis-net
authored and
NeoZhangJianyu
committed
vulkan: use smaller combined allocations to avoid fragmentation (#11551)

jeffbolznv
authored and
NeoZhangJianyu
committed
vulkan: initial support for IQ4_XS quantization (#11501)

remyoudompheng
authored and
NeoZhangJianyu
committed
vulkan: optimize coopmat2 iq2/iq3 callbacks (#11521)

jeffbolznv
authored and
NeoZhangJianyu
committed
ggml : fix LoongArch compile error with 128-bit SIMD (#11701)

junchao-loongson
authored and
NeoZhangJianyu
committed
build : fix llama.pc (#11658)

angt
authored and
NeoZhangJianyu
committed
llama : add log about loading model tensors (#11699)

ggerganov
authored and
NeoZhangJianyu
committed
SYCL: Adjust support condition for norm operators (#11674)

qnixsynapse
authored and
NeoZhangJianyu
committed
docs: update fedora cuda guide for 12.8 release (#11393)

teihome
authored and
NeoZhangJianyu
committed
server : (webui) migrate project to ReactJS with typescript (#11688)

ngxson
authored and
NeoZhangJianyu
committed
rpc: fix known RCE in rpc-server (ggml/1103)

retr0reg
authored and
NeoZhangJianyu
committed
sync : ggml

ggerganov
authored and
NeoZhangJianyu
committed
llama : fix old glm4 models (#11670)

tv1wnd
authored and
NeoZhangJianyu
committed
ggml : optimize and build warning fix for LoongArch (#11709)

MQ-mengqing
authored and
NeoZhangJianyu
committed
common : add default embeddings presets (#11677)

danbev
authored and
NeoZhangJianyu
committed
SYCL: remove XMX info from print devices (#11712)

qnixsynapse
authored and
NeoZhangJianyu
committed
llama : add llama_sampler_init for safe usage of llama_sampler_free (#11727)

cfillion
authored and
NeoZhangJianyu
committed
vulkan: print shared memory size (#11719)

jeffbolznv
authored and
NeoZhangJianyu
committed
llama : fix progress dots (#11730)

magicse
authored and
NeoZhangJianyu
committed
vocab : ignore invalid UTF-8 input in the BPE tokenizer (#11729)

cfillion
authored and
NeoZhangJianyu
committed
llama : fix defrag logic (#11707)

ggerganov
authored and
NeoZhangJianyu
committed
Make logging more verbose (#11714)

ericcurtin
authored and
NeoZhangJianyu
committed
server : (webui) fix numeric settings being saved as string (#11739)

ngxson
authored and
NeoZhangJianyu
committed
readme : update front-end framework (#11753)

pothitos
authored and
NeoZhangJianyu
committed
CUDA: fix min. version for movmatrix (#11751)

JohannesGaessler
authored and
NeoZhangJianyu
committed
ggml: Fix data race in ggml threadpool (#11736)

kkontny
authored and
NeoZhangJianyu
committed
cont : fix mmap flag print (#11699)

ggerganov
authored and
NeoZhangJianyu
committed
server : minor log updates (#11760)

ggerganov
authored and
NeoZhangJianyu
committed
server : (webui) increase edit textarea size (#11763)

woof-dog
authored and
NeoZhangJianyu
committed
server : (webui) revamp Settings dialog, add Pyodide interpreter (#11759)

ngxson
authored and
NeoZhangJianyu
committed
vulkan: account for lookup tables when checking shared memory size (#11502)

jeffbolznv
authored and
NeoZhangJianyu
committed
There's a better way of clearing lines (#11756)

ericcurtin
authored and
NeoZhangJianyu
committed
vulkan: add environment variable GGML_VK_PREFER_HOST_MEMORY to avoid VRAM allocation (#11592)

wbruna
authored and
NeoZhangJianyu
committed
vulkan: Make Vulkan optional at runtime (#11493). (#11494)

authored and
NeoZhangJianyu
committed
Update README.md [no ci] (#11781)

pascal-lc
authored and
NeoZhangJianyu
committed
sync: minja (https://github.com/google/minja/commit/a72057e5190de2c612d4598bb10b4bfd0f53011f) (#11774)

ochafik
authored and
NeoZhangJianyu
committed
server : correct signal handler (#11795)

ngxson
authored and
NeoZhangJianyu
committed
llama-mmap: fix missing include (#11796)

wgottwalt
authored and
NeoZhangJianyu
committed
server : (webui) introduce conversation branching + idb storage (#11792)

ngxson
authored and
NeoZhangJianyu
committed
docs: utilize the forward slash (/) as the path separator for Unix-like systems (#11770)

MambaWong
authored and
NeoZhangJianyu
committed
fix: typos in documentation files (#11791)

maximevtush
authored and
NeoZhangJianyu
committed
CUDA: use arch list for compatibility check (#11775)

authored and
NeoZhangJianyu
committed
server : use common_token_to_piece instead of common_detokenize (#11740)

danbev
authored and
NeoZhangJianyu
committed
Fix #11802: Compile bug - RegQueryValueExA changed to RegQueryValueEx (#11803)

sheldonrobinson
authored and
NeoZhangJianyu
committed
docs: add OpenCL (#11697)

lhez
authored and
NeoZhangJianyu
committed
llama : fix typo in llama-grammar.h [no ci] (#11816)

danbev
authored and
NeoZhangJianyu
committed
CUDA: fix CUDART_VERSION checks (#11821)

JohannesGaessler
authored and
NeoZhangJianyu
committed
ggml-cpu: Fix duplicate MATMUL_INT8 (#11817)

ownia
authored and
NeoZhangJianyu
committed
ggml : fix multi-threaded clamp_f32 (#11824)

Burton2000
authored and
NeoZhangJianyu
committed
cleanup: fix compile warnings associated with gnu_printf (#11811)

bandoti
authored and
NeoZhangJianyu
committed
HIP: Switch to std::vector in rocblas version check (#11820)

IMbackK
authored and
NeoZhangJianyu
committed
sync : ggml

ggerganov
authored and
NeoZhangJianyu
committed
Fix: Compile failure due to Microsoft STL breaking change (#11836)

MrSMlT
authored and
NeoZhangJianyu
committed
HIP: Remove GCN from list of devices that avoid MMQ (#11831)

IMbackK
authored and
NeoZhangJianyu
committed
server : (webui) Give copy button back to all message bubbles (#11814)

woof-dog
authored and
NeoZhangJianyu
committed
ggml : x2 speed for WASM by optimizing SIMD (#11453)

authored and
NeoZhangJianyu
committed
ggml-cpu : add chunking support to mul_mat_id (#11666)

slaren
authored and
NeoZhangJianyu
committed
llama : update llama_decode_internal ref [no ci] (#11840)

danbev
authored and
NeoZhangJianyu
committed
llama.cpp: fix warning message (#11839)

okuvshynov
authored and
NeoZhangJianyu
committed
sampling: add Top-nσ sampler (#11223)

authored and
NeoZhangJianyu
committed
`server`: fix tool-call of DeepSeek R1 Qwen, return reasoning_content (Command 7RB & DeepSeek R1) unless `--reasoning-format none` (#11607)

authored and
NeoZhangJianyu
committed
musa: bump MUSA SDK version to rc3.1.1 (#11822)

yeahdongcn
authored and
NeoZhangJianyu
committed
llama : add --completion-bash option (#11846)

danbev
authored and
NeoZhangJianyu
committed
server : (docs) Update wrong tool calling example (#11809)

RezaRahemtola
authored and
NeoZhangJianyu
committed
llamafile: use member variable instead of constant for iq4nlt (#11780)

jmorganca
authored and
NeoZhangJianyu
committed
readme : minor

ggerganov
authored and
NeoZhangJianyu
committed
llama-bench : fix unexpected global variable initialize sequence issue (#11832)

authored and
NeoZhangJianyu
committed
vulkan: linux builds + small subgroup size fixes (#11767)

netrunnereve
authored and
NeoZhangJianyu
committed
ggml: optimize some vec dot functions for LoongArch ASX (#11842)

MQ-mengqing
authored and
NeoZhangJianyu
committed
llama : add completion for --chat-template-file (#11860)

danbev
authored and
NeoZhangJianyu
committed
docker : drop to CUDA 12.4 (#11869)

ggerganov
authored and
NeoZhangJianyu
committed
cuda : add ampere to the list of default architectures (#11870)

slaren
authored and
NeoZhangJianyu
committed
opencl: Fix rope and softmax (#11833)

lhez
authored and
NeoZhangJianyu
committed
llguidance build fixes for Windows (#11664)

mmoskal
authored and
NeoZhangJianyu
committed
vulkan: initial support for IQ1_S and IQ1_M quantizations (#11528)

remyoudompheng
authored and
NeoZhangJianyu
committed
server: fix type promotion typo causing crashes w/ --jinja w/o tools (#11880)

ochafik
authored and
NeoZhangJianyu
committed
repo : update links to new url (#11886)

ggerganov
authored and
NeoZhangJianyu
committed
readme : add notice about new package registry (#11890)

ggerganov
authored and
NeoZhangJianyu
committed
metal : optimize dequant q6_K kernel (#11892)

akretz
authored and
NeoZhangJianyu
committed
examples: fix typo in imatrix/README.md (#11884)

authored and
NeoZhangJianyu
committed
scripts: fix compare-llama-bench commit hash logic (#11891)

JohannesGaessler
authored and
NeoZhangJianyu
committed
metal : fix the crash caused by the lack of residency set support on Intel Macs. (#11904)

halechan
authored and
NeoZhangJianyu
committed
vulkan: support multi/vision rope, and noncontiguous rope (#11902)

jeffbolznv
authored and
NeoZhangJianyu
committed
ci : fix (again) arm64 build fails (#11895)

ngxson
authored and
NeoZhangJianyu
committed
common : Fix a typo in help (#11899)

standby24x7
authored and
NeoZhangJianyu
committed
server : bump httplib to 0.19.0 (#11908)

ngxson
authored and
NeoZhangJianyu
committed
vulkan: implement several ops relevant for ggml_opt (#11769)

remyoudompheng
authored and
NeoZhangJianyu
committed
server : fix divide-by-zero in metrics reporting (#11915)

aviallon
authored and
NeoZhangJianyu
committed
update release requirements (#11897)

netrunnereve
authored and
NeoZhangJianyu
committed
CUDA: use async data loading for FlashAttention (#11894)

authored and
NeoZhangJianyu
committed
docs : Fix duplicated file extension in test command (#11935)

xiaobing318
authored and
NeoZhangJianyu
committed
scripts: corrected encoding when getting chat template (#11866) (#11907)

MoonRide303
authored and
NeoZhangJianyu
committed
server : add TEI API format for /rerank endpoint (#11942)

authored and
NeoZhangJianyu
committed
tool-call: refactor common chat / tool-call api (+ tests / fixes) (#11900)

ochafik
authored and
NeoZhangJianyu
committed
server : (webui) Enable communication with parent html (if webui is in iframe) (#11940)

authored and
NeoZhangJianyu
committed
llama : fix indentation in llama-grammar [no ci] (#11943)

danbev
authored and
NeoZhangJianyu
committed
speculative : update default params (#11954)

ggerganov
authored and
NeoZhangJianyu
committed
common : add llama.vim preset for Qwen2.5 Coder (#11945)

danbev
authored and
NeoZhangJianyu
committed
doc: add links to ggml examples [no ci] (#11958)

JohannesGaessler
authored and
NeoZhangJianyu
committed
run : add --chat-template-file (#11961)

engelmi
authored and
NeoZhangJianyu
committed
ggml: aarch64: implement SVE kernels for q3_K_q8_K vector dot (#11917)

authored and
NeoZhangJianyu
committed
ggml-cpu: Add CPU backend support for KleidiAI library (#11390)

chaxu01
authored and
NeoZhangJianyu
committed
server (webui): Fix Premature Submission During IME Conversion (#11971)

authored and
NeoZhangJianyu
committed
clip : fix visual encoders with no CLS (#11982)

alex-jw-brooks
authored and
NeoZhangJianyu
committed
MUSA: support ARM64 and enable dp4a .etc (#11843)

authored and
NeoZhangJianyu
committed
CUDA: correct the lowest Maxwell supported by CUDA 12 (#11984)

authored and
NeoZhangJianyu
committed
doc: update contributing guidelines [no ci] (#11969)

JohannesGaessler
authored and
NeoZhangJianyu
committed
llama : skip loading unused tensors (#12004)

ggerganov
authored and
NeoZhangJianyu
committed
llama.swiftui : add "Done" dismiss button to help view (#11998)

danbev
authored and
NeoZhangJianyu
committed
cuda: Add Q5_1, Q5_0, Q4_1 and Q4_0 to F32 conversion support. (#12000)

gcp
authored and
NeoZhangJianyu
committed
server : disable Nagle's algorithm (#12020)

ggerganov
authored and
NeoZhangJianyu
committed
ci : Build on Github-hosted arm64 runners (#12009)

Rohanjames1997
authored and
NeoZhangJianyu
committed
CUDA: optimize FA for GQA + large batches (#12014)

JohannesGaessler
authored and
NeoZhangJianyu
committed
ci : fix arm upload artifacts (#12024)

ggerganov
authored and
NeoZhangJianyu
committed
llava: build clip image from pixels (#11999)

authored and
NeoZhangJianyu
committed
CUDA: app option to compile without FlashAttention (#12025)

JohannesGaessler
authored and
NeoZhangJianyu
committed
ggml-cpu: Support s390x SIMD Instruction Set (#12019)

authored and
NeoZhangJianyu
committed
Some llama-run cleanups (#11973)

ericcurtin
authored and
NeoZhangJianyu
committed
run: allow to customize prompt by env var LLAMA_PROMPT_PREFIX (#12041)

benoitf
authored and
NeoZhangJianyu
committed
SYCL: Fix GGML_SYCL_DEBUG macro (#11995)

qnixsynapse
authored and
NeoZhangJianyu
committed
gguf_convert_endian.py: implement byteswapping for q4_k and q6_k (#11349)

AlekseiNikiforovIBM
authored and
NeoZhangJianyu
committed

Commits on Feb 26, 2025

[SYCL] Optimize mul_mat for Q4_0 on Intel GPU (#12035)

NeoZhangJianyu
and
arthw
committed
llava : Add Granite Vision Support (#11794)

authored and
NeoZhangJianyu
committed
opencl: fix for small models (#11950)

authored and
NeoZhangJianyu
committed
add new line at end of file
NeoZhangJianyu
committed