ggml/hip: fix APU compatibility - soft error handling for hipMemAdviseSetCoarseGrain - #15

Closed

moonshadow-25 wants to merge 329 commits into

pwilkin:masterfrom

moonshadow-25:fix/hip-apu-compatibility

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

Commits on Feb 17, 2026

opencl: optimize mean and sum_row kernels (#19614)

shaofeiqi
and
lhez
authored
983559d
opencl: refactor expm1 and softplus (#19404)

shaofeiqi
and
lhez
authored
e2f19b3

Commits on Feb 18, 2026

common : make small string helpers as inline functions (#19693)
angt
authored
a569bda
vulkan: split mul_mat into multiple dispatches to avoid overflow (#19509)
jeffbolznv
authored
d0061be
Pre-MCP UI and architecture cleanup (#19689)
allozaur
authored
ea00322
ggml webgpu: shader library organization (#19530)

reeselevine
and
neha-ha
authored
238856e
docs: Fix broken links for preparing models in Backends (#19684)
MaciejDromin
authored
e99f108
model: support GLM-OCR (#19677)

ngxson
and
CISC
authored
eeef3cf
server: save generated text for the /slots endpoint (for LLAMA_SERVER_SLOTS_DEBUG=1) (#19622)

authored
b55dcde
ggml webgpu: Fix bug in dispatching large matrix-vector multiplication (#19535)
reeselevine
authored
e7f2f95
Add Jinja support for "indent" string filter (#19529)

pwilkin
and
CISC
authored
8a70973

Commits on Feb 19, 2026

Commits on Feb 20, 2026

Commits on Feb 21, 2026

hexagon : fix build release (#19444) (#19587)
mengshengwu
authored
ba3b9c8
fix: UI single model selection in router mode (#19767)
crsawyer
authored
07968d5
Improve CUDA graph capture (#19754)

authored
a0c91e8
vendor : update cpp-httplib to 0.33.1 (#19778)
angt
authored
99156f3
Add a build target to generate ROCm artifacts using ROCm 7.2 (#19433)
superm1
authored
f75c4e8
Update ROCm docker container to 7.2 release (#19418)
superm1
authored
3571565

Commits on Feb 22, 2026

ci : fix rocm release path [no ci] (#19784)
CISC
authored
e877ad8
server : merge contiguous Responses input items into a single assistant message (#19773)
aldehir
authored
34ec1c3
ci : fix rocm archive name [no ci] (#19808)
CISC
authored
9f0684f
model : add Kanana-2 model support (#19803)
HelloKS
authored
ae2368e
Fix wrong cli-argument in documentation (#19804)
Menkalian
authored
cacc371
common : fix improper trimming in XML parser on complete message (#19805)

aldehir
and
julio75012
authored
ed48378
jinja: correct stats for tojson and string filters (#19785)
ngxson
authored
5452d73
cli : provide model with text filename (#19783)
CISC
authored
e8e2616

Commits on Feb 23, 2026

Commits on Feb 24, 2026

Commits on Feb 25, 2026

gguf : fix ftell/fseek for Windows (#19870)
aldehir
authored
a96a112
ci : update Windows ROCm build to 26.Q1 [no ci] (#19810)
superm1
authored
8fdf269
scripts: update corpus of compare-logprobs (#19326)
ngxson
authored
c747294
server : support multi-modal context checkpoints (#19849)

ggerganov
and
timkhronos
authored
d7d826b
server : enable multi-modal prompt caching (#19877)
ggerganov
authored
f20469d
ci : update the ROCm/HIP toolchain versions [no ci] (#19891)

slojosic-amd
and
CISC
authored
3af34b9
common : add more aliases for sampler CLI params (#19797)
ddh0
authored
832aa94
vulkan: check for memory overlap before doing fusion (#19768)
jeffbolznv
authored
3769fe6

Commits on Feb 26, 2026

support permuted, remove check s0/s10 (#19889)

arthw
and
NeoZhangJianyu
authored
2943210
server : fix typo in server README.md (#19900)
yggdrasil75
authored
bd72300
gguf : avoid too many file size calls (#19919)
ggerganov
authored
1ca3d1d
model : add Jina Embeddings v5 Nano (partial EuroBERT) support (#19826)
maximilianwerk
authored
66287bd
jinja : correct default size for string slices (#19913)
EZForever
authored
9b62913
server: fix load-on-startup not respected in ini file (#19897)

drrros
and
Roman Marchenko
authored
efba35a
ggml-virtgpu: improve the reliability of the code (#19846)
kpouget
authored
ffaafde
llama: Add option to merge gate and exp weights (#19139)

am17an
and
CISC
authored
b68d751
kv-cache : fix can_shift() check to take into account M-RoPE (#19928)
ggerganov
authored
99bd67c
server : fix ctx checkpoint restore logic (#19924)
ggerganov
authored
01cd448
mtmd : fix padding of n_tokens (#19930)
ggerganov
authored
37964f4
vulkan: fix fp16 Flash Attention on Windows AMD RDNA2 and below (#19921)
0cc4m
authored
723c710
ggml : fix AMX and add batched support (#19925)
angt
authored
4e76d24

Commits on Feb 27, 2026

Commits on Feb 28, 2026

Commits on Mar 1, 2026

Commits on Mar 2, 2026

Commits on Mar 3, 2026

Commits on Mar 4, 2026

completion : Fix a typo in warning message (#20082)
standby24x7
authored
54910bd
Fix locale-dependent float printing in GGUF metadata (#17331)

ssam18
and
JohannesGaessler
authored
cb8f4fa
impl : use 6 digits for tensor dims (#20094)
ddh0
authored
c99909d
ggml : use a simple std::thread in AMX without OpenMP (#20074)
angt
authored
66199c9
ggml: fix ggml_is_contiguous_n for ne == 1 (#20092)
JohannesGaessler
authored
7f5ee54
tools : add missing clocale include in mtmd-cli [no ci] (#20107)
CISC
authored
d969e93
Add concat op to webgpu. (#20068)
yomaytk
authored
541bf37
[WebGPU] Fix wait logic for inflight jobs (#20096)
nikhilJain17
authored
24d2ee0
hexagon: add llama-completion runner script (#20095)
tboinovski1
authored
1a29907

Commits on Mar 5, 2026

opencl: add `SET`, support i32 for `CPY`, minor refactor for cpy (#20101)
lhez
authored
69fd345
hexagon: Flash Attention optimizations (dma, mpyacc, multi-row) and MatMul updates (#20118)

max-krasnyansky
and
chraac
authored
7a99dc8
chore : correct typos [no ci] (#20041)

authored
92f7da0
webui: Improvements for Models Selector UI (#20066)
allozaur
authored
5e335ba
convert : register Qwen 3.5 ForCausalLM for text only (#20119)
CISC
authored
cf23251
cli : add command and file auto-completion (#19985)
CISC
authored
b5ed0e0
model : update Qwen3.5 model type detection (#20126)

EZForever
and
CISC
authored
872646b
CUDA: Improve performance via less synchronizations between token (#17795)

aendk
and
ggerganov
authored
2cd20b7
models : kda chunk size = 16 (#19827)

ymcki
and
ggerganov
authored
a0ed91a

Commits on Mar 6, 2026

hexagon: add fp16 support for binary ops: add,sub,mul,div (#20139)

YardenTal44
and
max-krasnyansky
authored
2b10b62
opencl: add neg, exp and diag (#20127)
lhez
authored
6c97bff
cli : Don't clear system prompt when using '/clear' (#20067)
roj234
authored
f7db3f3
kv-cache : fix M-RoPE checkpoints (#20132)
ggerganov
authored
17a4258
ggml-cpu: fix data race for debug asserts (#20148)
JohannesGaessler
authored
2850bc6
webui: Agentic Loop + MCP Client with support for Tools, Resources and Prompts (#18655)
allozaur
authored
f6235a4
Checkpoint every n tokens: squash (#20087)
pwilkin
authored
f5ddcd1
context: ignore zero scale LoRAs when checking sameness (#20166)
TimNN
authored
388baab
CUDA: use shared mem for ssm_conv (#20128)

am17an
and
JohannesGaessler
authored
1e38a7a
ggml-cpu: Fix gcc 15 ICE on ppc64le (#20083) (#20130)
shalinib-ibm
authored
c6980ff
ggml: update comments for backends which have no memory to report (#20157)
taronaeo
authored
ba2ff79
ggml-cuda: add mem check for fusion (#19916)

am17an
and
JohannesGaessler
authored
d48e876
cpu: skip redudant ROPE cache updates (#20149)
max-krasnyansky
authored
ba2fd11
server : preserve anthropic thinking blocks in conversion (#20120)

T0mSIlver
and
root
authored
e68f2fb
hexagon: add f32 ssm_conv op (#20122)

tboinovski1
and
max-krasnyansky
authored
34df42f
Autoparser - complete refactoring of parser architecture (#18675)
pwilkin
authored
566059a
Add @pwilkin to CODEOWNERS for autoparser code (#20174)
pwilkin
authored
7463687
quants : Add memsets and other fixes for IQ quants (#19861)
bartowski1182
authored
649f064
Autoparser: add optional argument reshuffle capability (#20171)
pwilkin
authored
2f2923f

Commits on Mar 7, 2026

Autoparser: True streaming (#20177)
pwilkin
authored
c024d85
opencl: add l2_norm (#20160)
lhez
authored
6fce5c6
ggml: add GATED_DELTA_NET op (#19504)

am17an
and
ggerganov
authored
c5a7788

Commits on Mar 8, 2026

[SYCL] supprt Flash Attention for fp32/fp16/Q4/Q5/Q8 (#20190)
arthw
authored
213c4a0
server : correct index on finish in OAI completion streams (#20226)
decahedron1
authored
ff52ee9
Revert to OAI-compatible args (#20213)
pwilkin
authored
b283f6d
readme : update infra list (#20212)
Defilan
authored
a950479
llama: end-to-end tests (#19802)
JohannesGaessler
authored
a976ff0
vulkan: Fix data races in coopmat1 mul_mat(_id) (#20084)
jeffbolznv
authored
cd18a50
ggml-vulkan: Add ELU op support (#20183)
GiantPrince
authored
d088d5b
Fix structured outputs (#20223)

pwilkin
and
aldehir
authored
62b8143
Fix compile bug (#20203)

pwilkin
and
CISC
authored
9b24886
common : gracefully handle incomplete output (#20191)
aldehir
authored
451ef08
graph : remove redundant scale_w parameter (#20235)
CISC
authored
35bee03
server : do not create checkpoints right after mtmd chunks (#20232)
ggerganov
authored
d417bc4

Commits on Mar 9, 2026

PEG parser for LFM2 (#20251)
pwilkin
authored
97c64fb
llama-bench: introduce `-hf` and `-hff` flags & use `--mmap 1` by default (#20211)
taronaeo
authored
ae87863
cuda : display total and free VRAM capacity during device initialization (#20185)
tehsiuhuang
authored
5f4cdac
vulkan: skip zero size tensors in backend copies (#20233)
0cc4m
authored
b2f460b
ggml-vulkan: add SGN operator, auto-generate Vulkan.csv and ops.md (#20219)
bertaye
authored
0beb8db
contributing: limit open PRs for new contributors to 1 (#20036)
am17an
authored
e2763a6
llama-quant : left-align tensor names in output (#20117)
ddh0
authored
b518195
ggml-cuda: disable gdn for musa (#20278)
am17an
authored
e8bbc73
server : add kill switch when server is stuck (#20277)
ggerganov
authored
107d599
models : fix assert in mamba2 graph (#20270)
ggerganov
authored
43e1cbd
common: map developer role to system (#20215)
pwilkin
authored
f76565d
server : fix off-by-1 in server_tokens::size_up_to_pos() (#20279)
ggerganov
authored
d6e1556
server : warn swa-full is not supported for non-SWA models (#20291)
ggerganov
authored
344ee2a
metal : add upscale (#20284)
ggerganov
authored
ed0007a
server : fix checkpoints n_tokens calculation (#20287)
ggerganov
authored
96cfc49
metal : extend mul_mv_ext to BF16, Q2_K, Q3_K (#20250)

arkavo-com
and
claude
authored
e22cd0a
server: Parse port numbers from MCP server URLs in CORS proxy (#20208)

eapache
and
ngxson
authored
23fbfcb
llama: dynamic head_dim and n_rot for SWA (#20301)
ngxson
authored
59db9a3
model: fix step3.5 n_rot (#20318)
ngxson
authored
0842b9b
common: consolidate PEG string parsers (#20263)
aldehir
authored
c96f608

Commits on Mar 10, 2026

Commits on Mar 11, 2026

fix for failed UT case: ACC, L2_NORM, UPSCALE, fused_glu, unary (#20283)
arthw
authored
b2e1427
fix op rope, add rope_back (#20293)
arthw
authored
0cec84f
vendor : update miniaudio to 0.11.25 (#20209)
cabelo
authored
4f2f0a1
vendor : update cpp-httplib to 0.37.0 (#20207)
cabelo
authored
e1a3999
Fix agentic mcp image single model (#20339)
ServeurpersoCom
authored
00de615
cuda/hip: fix loop unrolling in ssm-conv (#20369)
IMbackK
authored
9ef7523
ggml-cuda: gdn use shared mem for HIP (#20366)
IMbackK
authored
5f91b1d
common/parser: handle reasoning budget (#20297)

authored
acb7c79
common/parser: use nlohmann::ordered_json to preserve parameter order (#20385)
aldehir
authored
b5fe455
ci: disable coopmat on ubuntu-24-cmake-vulkan job (#20294)
0cc4m
authored
182acfe
[SYCL] Update SYCL.md for binary package for Windows (#20401)
arthw
authored
ecac98e
metal : add env var to trigger graph capture (#20398)
ggerganov
authored
c363256
metal : fix q5_k mul_mv register spill (#20399)
ggerganov
authored
b541241
compare-llama-bench: check remotes as well (#20406)
am17an
authored
bd1ec81
metal : fix capture_compute counter logic (#20410)
ggerganov
authored
76ea1c1
llama : add support for Nemotron 3 Super (#20411)

authored
eaf1d79
benches : add nemotron super (#20420)
ggerganov
authored
3ca19b0
ggml : add NVFP4 quantization type support (#19769)

richarddd
and
CISC
authored
5eae9cb
llama : whitespace cleanup (#20422)
CISC
authored
f90bd1d
llama : enable chunked fused GDN path (#20340)

authored
d28961d
ggml-webgpu: Add supports for `GGML_OP_REPEAT` (#20230)
yomaytk
authored
f2ab047
common : fix --n-cpu-moe, --cpu-moe for models with fused gate + up (#20416)
ddh0
authored
4a748b8
graph : add optional scale parameter to build_lora_mm [no ci] (#20427)
richarddd
authored
1eea6a2
model : add support for Phi4ForCausalLMV (#20168)

dranger003
and
ngxson
authored
fdb1764

Commits on Mar 12, 2026

common/parser: add GigaChatV3/3.1 models support (#19931)

Mishusha
and
Mishusha
authored
a8304b4
hip: compile debug builds with -O2 on hip to avoid a compiler bug (#20392)
IMbackK
authored
d63aa39
opencl: add cumsum op (#18981)

shaofeiqi
and
lhez
authored
3d9ab22
opencl: use larger workgroup size for get_rows (#20316)
lhez
authored
0516e04
vulkan: Fix ErrorOutOfHostMemory on Intel GPU when loading large models with --no-mmap (#20059)
rillomas
authored
5866e3b
vulkan: fix OOB check in flash_attn_mask_opt (#20296)
jeffbolznv
authored
aa429cf
vulkan: fix l2_norm epsilon handling (#20350)
jeffbolznv
authored
246ffc4
ci: Setup self-hosted CI for Intel Linux Vulkan backend (#20154)
rillomas
authored
4cc6eb1
metal : avoid divisions in bin kernel (#20426)
ggerganov
authored
e4cff09
ggml-virtgpu: Fix some build commands (#20341)
yomaytk
authored
0503996
New conversations now auto-select the first loaded model (#20403)
ServeurpersoCom
authored
de19015
vulkan: fix SSM_CONV PP scaling with large ubatch sizes (#20379)

ProgenyAlpha
and
claude
authored
40c550d
convert : better mtp check and fix return [no ci] (#20419)
CISC
authored
c3e3f9e
vulkan: add GATED_DELTA_NET op support (#20334)

ProgenyAlpha
and
claude
authored
deee238
grammar: Fix grammar root symbol check (#19761)
AsbjornOlling
authored
0a10c34
common : update completion executables list [no ci] (#19934)
danbev
authored
6de1bc6
test-backend-ops: allow loading tests from file and parsing model operators into file (#19896)
0cc4m
authored
128142f
tests : use `reasoning` instead of `reasoning_budget` in server tests (#20432)
pwilkin
authored
0e81041
vendor : update cpp-httplib to 0.37.1 (#20390)
cabelo
authored
557fe2d
llama : disable graph reuse with pipeline parallelism (#20463)
ggerganov
authored
57819b8

Commits on Mar 13, 2026

convert : fix/suppress pyright errors (#20442)
danbev
authored
983df14
metal : fix l2 norm scale (#20493)
ggerganov
authored
73c9eb8
general: CONTRIBUTING.md - guidelines for quantization schemes (#19762)

pwilkin
and
JohannesGaessler
authored
2948e60
mtmd : rename mtmd_get_audio_bitrate to mtmd_get_audio_sample_rate (#20105)
danbev
authored
8f974d2
ggml : fix typo gmml (#20512)
angt
authored
b5e1212
ggml-cpu: add RVV vec dot kernels for quantization types (#18859)

authored
fbaa95b
server: reset counter related to kill-switch on client error (#20513)

SoftwareRenderer
and
ggerganov
authored
d7ba99c
llama : fix pooling assertion crash in chunked GDN detection path (#20468)

ZeroV0LT
and
Domenico Crupi
authored
f17b3be
common/parser: gracefully handle undetected tool parser, print error message. (#20286)
pwilkin
authored
1430c35
graph : remove redundant GDN state transposes (#20443)

authored
e30f1fd
tools : enable kvu in perplexity for hellaswag, winogrande, multiple-choice (#19954)
angt
authored
463b6a9

Commits on Mar 14, 2026

ggml/hip: fix APU compatibility - soft error handling for hipMemAdviseSetCoarseGrain

moonshadow-25
and
claude
committed
3e28c43