llama.cpp-xrip

wmantly/llama.cpp-xrip

0 Fork 0

7951f601e8 vendor : update cpp-httplib to 0.62.0 (#30785) Alessandro de Oliveira Faria (A.K.A.CABELO) 2026-10-19 09:23:23 -03:00
33d6135cbe moe-cache : replace MoE LRU cache with a 50/50 SLRU cache (#30637) Evan Huus 2026-10-19 07:28:50 -04:00
098ce526b5 metal : fix moe_reduce fusion race when dst aliases its inputs (#30795) Georgi Gerganov 2026-10-19 11:55:02 +03:00
f3fb2fcdbc cann : refactor #ifdef blocks to avoid unreachable code after return (#25273) Raul Torres 2026-10-19 09:17:54 +01:00
fda00f18cd tests: add big-endian conversion tests (#30705) Aaron Teo 2026-10-19 16:14:58 +08:00
4866db350d CUDA: fix warning for nonexistent extra semicolon (#30774) Johannes Gäßler 2026-10-19 10:08:21 +02:00
42b70032b1 ggml-cpu: vectorize ggml_v_silu and ggml_v_expf (#30791) Aaron Teo 2026-10-19 15:59:49 +08:00
4c1004c492 cuda: cap MMQ Stream-K blocks at available work (#30553) JoelandJohannes Gäßler 2026-10-19 03:41:14 -04:00
412a363c65 qwen3 : bidirectional attention and dense_2 output projection (#30607) fzowlandSigbjørn Skjæret 2026-10-19 08:29:02 +02:00
2b57148b08 ui: Fix markdown tables and lists being cut off on phones (#30687) Aleksander Grygier 2026-10-19 08:28:16 +02:00
8f97d82da3 tools : reject truncated GGUF tensor data in split and export tools (#30497) ppAlexa 2026-10-19 11:57:46 +05:30
27af58168a tests : update Metal fusion baseline for beam (#30764) Georgi Gerganov 2026-10-19 09:11:38 +03:00
7db0d3d781 pi : turn .pi/gg into a system prompt addendum (#30762) Georgi Gerganov 2026-10-19 09:11:13 +03:00
99db719bdf CUDA: fix shared-memory WAR hazard in flash_attn_ext_vec (#30573) Paul MairhoferandJohannes Gäßler 2026-10-19 07:25:04 +02:00
31288c9c66 hexagon: FLASH_ATTN_EXT improvements for longer context and SWA models (#30731) Max Krasnyansky 2026-10-18 18:08:11 -07:00
fe8841a78a hexagon: allow multistream FA (#30218) Jhen-Jie Hong 2026-10-19 09:07:53 +08:00
ff5a233767 cuda : fuse qwen4exp's SCALE -> SIGMOID -> SCALE -> hc_post chain (#30602) Aman Gupta 2026-10-18 23:18:47 +05:30
0878f7c304 ggml: support non-contiguous SUM (CPU, CUDA, Vulkan, WebGPU) (#28851) Andrea Rossetti 2026-10-18 18:57:36 +02:00
5aeac37a35 ggml-cuda: fattn do not check oob if k_VKQ_max = (k_VKQ_max - nbatch_fa) (#30572) thelittlefireman 2026-10-18 10:59:06 +02:00
547208dca2 Restore ESIMD kernels on FFN path (#30239) Todd Malsbary 2026-10-17 21:02:38 -07:00
848bbda904 hexagon: add support for hadamard MUL_MAT and GET_ROW for flat and tiled Q1_0 and Q2_0 (#30627) Todor BoinovskiandMax Krasnyansky 2026-10-17 19:17:33 -07:00
797de63b2f hexagon: add support for Q1_0 and Q2_0 (#30122) Todor BoinovskiandMax Krasnyansky 2026-10-17 13:36:38 -07:00
9d1878a435 ggml-cuda: HIP: use __builtin_amdgcn_perm for q1_0 unpacking (#29927) thelittlefireman 2026-10-17 19:04:39 +02:00
c0fd786264 ui : Discover & Download Models from HF Hub (#29584)
Aleksander Grygier Pascal Adrien Gallouët
2026-10-17 17:20:51 +02:00
2744e48d1e CUDA: faster MMVQ for low-bit types (#30517) Aman Guptaandthelittlefireman 2026-10-17 19:05:30 +05:30
10b8b75534 tests: give the random model generator a per-tensor filler table (#30674) Pascal 2026-10-17 15:07:55 +02:00
9d1e03616e cuda : fix -Wtautological-compare in fattn vec dispatch (#30519) Adrien Gallouët 2026-10-17 14:36:17 +02:00
f2f667b02d cuda : add grouped experts top-k fusion (#29181) Sigbjørn Skjæret 2026-10-17 13:55:33 +02:00
62318cd57d model : add Kolibri-1 (Aleph-Alpha) support (#30673) Xue ZhenliangandSigbjørn Skjæret 2026-10-17 19:00:17 +08:00
83f9389419 ci : added ABI reports as github artifact (#30619) Sandro Hanea 2026-10-17 11:41:53 +02:00
1268edac57 model : Granite switch decision 256 options (#30693) Gabe Goodhart 2026-10-17 01:14:32 -06:00
f74e2fa6a3 sycl: fix prompt-processing regression after Q4_0/Q8_0 weight reorder (#30374) Marcus 2026-10-16 19:03:41 -07:00
ea0c4c2c85 skills: add slice-model (#30677) Xuan-Son Nguyen 2026-10-17 02:11:56 +02:00
6420cf5dd5 ci: update device matrix for Snapdragon jobs by removing SM8750 (#30620) Zack Li 2026-10-16 16:25:05 -07:00
27c7a769da hexagon: vectorize softplus with HVX (#30435) wes 2026-10-17 06:20:43 +08:00
2f2763db6e opencl: Add MxFP4 MoE bin kernel for Adreno X1 devices (#30691) Shawn GuandLi He 2026-10-16 15:15:04 -07:00
01fc3aae84 model: add Beam (Reflection) (#30670) Daniel Han 2026-10-16 15:07:55 -07:00
23f7aabc16 model : Granite Switch Decision (#30676) Gabe Goodhart 2026-10-16 12:00:41 -06:00
7182ef71e6 common: honor HTTP_PROXY/HTTPS_PROXY env vars in http client (#21752)
texasich texasich texasich marecabo
2026-10-16 10:01:37 -05:00
2a667032dc convert : respect Mistral format for DeepseekV2Model (#30669) aiqubits 2026-10-16 21:17:30 +08:00
f3363593f4 Add more benchmarks to llama-eval (#26174) Piotr Wilkin (ilintar) 2026-10-16 11:52:18 +02:00
ceefd16daa hexagon: correct HMX FA sinks (#30581) Jhen-Jie HongandMax Krasnyansky 2026-10-16 16:51:14 +08:00
9a16f1dbe4 model : MTP support for Mellum2.1 (#30612) Mikhail Podvitskii 2026-10-16 10:20:03 +02:00
b84ea5cf5e convert : speed up model endian conversion and add more datatype support (#30647) Aaron Teo 2026-10-16 16:18:29 +08:00
a30fb4924d [SYCL]: fix memory errors in mul_mat, split buffer, host pool (#29889) Michael Taylor 2026-10-16 04:28:50 -03:00
495c5805ce gguf-py : add numpy Q1_0 and Q2_0 de/quantization support (#30630) Sh1m13 2026-10-16 12:56:14 +05:30
623c16393c CUDA: HIP: rms_norm: add variants that load float4 at a time and store x in registers (#29720) uvos 2026-10-16 09:07:34 +02:00
13b8048920 common : read ngram cache parts by reference instead of copying (#30061) hayder 2026-10-16 02:06:32 -05:00
0f4376268f convert : support text-only checkpoints in pplx-decider (#30575) FINAL-Bench 2026-10-16 15:56:00 +09:00
3dbcb7fca5 cpu : reduce AVX2 tiled accumulators together (#30296) Trivalik 2026-10-16 08:48:23 +02:00
c804754529 ggml-cpu: refactor s309x repack helper functions (#30588) Aaron Teo 2026-10-16 14:45:35 +08:00
c0083e3052 sycl : reduce the build threads to half (#30236) Neo Zhang 2026-10-16 14:44:24 +08:00
13de11e78f sycl: wide stores for the q4_K/q5_K weight dequant (#29696) Titaniumtown 2026-10-15 23:43:07 -07:00
88aaf325a7 tests : catch vanished rows in the fusion baseline (#30618) Georgi Gerganov 2026-10-16 09:35:19 +03:00
a8d9a7e1e5 vulkan: fix gpu-visible host memory issues (#30468) Ruben Ortlam 2026-10-16 08:29:23 +02:00
f12aaa344c Fix failing RocM tests (#30611) Karthik VijayanandKarthik Vijayan 2026-10-16 04:46:55 +00:00
5e03a1ff45 graph : add precision handling for MELLUM architecture (#30584) Mikhail Podvitskii 2026-10-16 05:32:26 +02:00
b797b9ca81 hexagon: tensor repacking rework and support for tiled and flat quantized and fp weights (#30635) Max Krasnyansky 2026-10-15 20:30:11 -07:00
6afa9a3134 llama : share the k-pool input views between layers (#30613) Aman Gupta 2026-10-16 08:44:36 +05:30
b6334e37e6 Limit max CUDA pinned host buffer size to 4 GB (#30449) Gaurav Garg 2026-10-16 08:20:26 +05:30
e707790c73 sycl: support reordered Q8_0 MoE weights (#29337) Frosty40 2026-10-15 20:59:38 -05:00
43d57fb8ca [SYCL] Enable ssl in build script and CI (#30456) Neo ZhangandSigbjørn Skjæret 2026-10-16 09:58:07 +08:00
2851b3bcbf mtmd: add mtmd_audio_stream + support voxtral-realtime (#30629) Xuan-Son Nguyen 2026-10-16 01:46:42 +02:00
59d5fad7c1 convert : fix Granite Switch block_count for MultiSwitch checkpoints (#30621) Nigel JonesandGabe Goodhart 2026-10-15 23:41:01 +01:00
e287bd8729 convert : fix regression in mapping of Granite speech tensors (#30628) Matt Rutkowski 2026-10-15 17:31:14 -05:00
2f729cfa4c mtmd: lightonocr does not require longest edge to always be 1540, native resolution is the default (#30600) Said Taghadouini 2026-10-15 23:53:02 +02:00
ba3c05f71d model: add Muse Spark Open (text, vision and audio) (#30616)
Pascal Xuan Son Nguyen Pedro Cuenca Sigbjørn Skjæret
2026-10-15 23:27:12 +02:00
b76b13379d pi : expand AI disclosure with hardware and stats (#30615) Georgi Gerganov 2026-10-15 20:32:52 +03:00
620d2225cc llama.cpp : bump version to 0.7.0 (#30603) Georgi Gerganov 2026-10-15 16:48:22 +03:00
cd0a5b6a18 models : add Nemotron Super 3.5 VL and Nemotron Bolt 74B A9B support (#30520) Kartik Gulia 2026-10-15 19:04:56 +05:30
a11b6530b5 scripts : add ci-status.sh to list active CI runs (#30586) Georgi Gerganov 2026-10-15 16:28:09 +03:00
61ad80bf93 server: fix the off-by-two in the SWA reuse threshold (#30580) Pascal 2026-10-15 15:25:19 +02:00
a6b91d8c5e model : Mellum - read MoE gating params from GGUF, unify RoPE (#30578) Mikhail PodvitskiiandSigbjørn Skjæret 2026-10-15 15:24:47 +02:00
ea736c1d5c server: save the draft state in the slot file (#30574) Pascal 2026-10-15 15:24:19 +02:00
b0d239164b server: prefer empty slots and keep states overwritten by child tasks (#30567) Pascal 2026-10-15 15:24:01 +02:00
c0cb53ac4e sync : ggml Georgi Gerganov 2026-10-15 16:16:53 +03:00
4ed1c8507e ggml : bump version to 0.27.0 Georgi Gerganov 2026-10-15 15:38:20 +03:00
711780ffc7 ci : bump ty to 0.0.88 (#30587) Sigbjørn Skjæret 2026-10-15 14:15:55 +02:00
fd55cb98fe fix(ui): take the agentic turns default from DEFAULT_AGENTIC_CONFIG (#30582) Aleksander Grygier 2026-10-15 12:36:06 +02:00
3e4dfa8f39 models : add maion-coder (#29778) Karthik VijayanandKarthik Vijayan 2026-10-15 09:18:46 +00:00
08698c81ed jinja : improve context copy constructor performance (#30562) William Tan 2026-10-15 05:00:57 -04:00
82e1e2fd84 jinja : write nan and inf in tojson as json.dumps does (#30483) Yuri Khrustalev 2026-10-15 04:49:40 -04:00
47f5a64ae1 vulkan : fix concurrent first device init (#29766) losewayy 2026-10-15 16:37:35 +08:00
e36df60470 sycl : replace oneAPI package by DLE+oneDNN package (#30131) Neo Zhang 2026-10-15 15:26:01 +08:00
9166ceaf16 opencl: fix OOB tail access for adreno noshuffle gemv kernels (#30421) lhezandHongqiang Wang 2026-10-15 00:21:18 -07:00
2130ca8476 jinja : fix loop body variable scoping (#30533) William Tan 2026-10-15 03:14:01 -04:00
23f83e056d ggml-cpu : optimize pad_f32 (#30526) Adrien Gallouët 2026-10-15 09:13:18 +02:00
2ce20d22fd unicode : reject overlong UTF-8 and out-of-range codepoints (#30485) Simone IacominoandSigbjørn Skjæret 2026-10-15 09:12:56 +02:00
68ac765726 llama : split llama_prefetch() into merge and prefetch steps (#30532) Adrien Gallouët 2026-10-15 09:12:45 +02:00
77c1f39929 opencl: apply flash attn softcap before mask (#30546) lhez 2026-10-15 00:11:51 -07:00
733f57a999 vulkan : copy strided f16 KV once on AMD so reads spread across memory channels (#27703) Nathan WilsonandRuben Ortlam 2026-10-15 17:39:02 +10:30
5a638aedb7 cpu: add F16 SWIGLU_OAI reference implementation and validation (#29200) cqderekandravi9 2026-10-15 15:07:45 +08:00
6e66cb5d4e metal : add BF16 variant of the tensor API flash attention kernel (#30491) Ethan Guo 2026-10-15 14:39:47 +08:00
e467e5b5e5 kv-cache: honor PARTIAL_ONLY in the plain KV cache (#30514) PascalandGeorgi Gerganov 2026-10-15 06:39:18 +02:00
1ddc262558 graph: keep the graph non empty without embeddings (#30522) Pascal 2026-10-15 06:38:12 +02:00
2639a3b4e0 server: save only restorable slot states to the prompt cache (#30498) Pascal 2026-10-15 06:30:17 +02:00
843e7dcc15 spec : add lilicorr support (#30506) Kartik GuliaandGeorgi Gerganov 2026-10-15 09:19:25 +05:30
3e390af18d vulkan: Use L2-friendly workgroup scheduling for cm2 mulmat (#30215) Ruben Ortlam 2026-10-15 05:26:10 +02:00
a0e6dc7384 opencl: support CLAMP on non-contiguous views (#30186) Hongqiang Wang 2026-10-14 13:02:27 -07:00
23deec0255 devops: incorrect COPY instruction (#30529) stormy 2026-10-14 20:09:28 +02:00

... 2 3 4 5 6 ...

Commit Graph Select branches Hide Pull Requests master Mono Color

Commit Graph

Select branches

Hide Pull Requests

master