llama.cpp-xrip

wmantly/llama.cpp-xrip

0 Fork 0

fa715571ea common: load backends before parsing mmproj-device master wmantly 2026-11-04 20:29:27 +00:00
7e2a519ecf Apply HauhauCS FastMTP d2t vocab trimming patch wmantly 2026-11-04 20:29:27 +00:00
798f5b4847 qwen4exp: fix PLE direct-read block, Linux-only non-compiling xrip 2026-11-01 22:41:44 +03:00
60c62ce9f2 Merge upstream master; take upstream DFlash2, MMQ fusion, and top-k xrip 2026-11-01 22:29:01 +03:00
785322d17f requirements : fix numpy dependencies (#31508) Sigbjørn Skjæret 2026-11-01 18:23:40 +01:00
dde879bc4f sycl: Try PkgConfig for LevelZero includes / libs (#31076) Michael Taylor 2026-11-01 02:46:43 -04:00
060b2fea4e opencl: add initial iq3 support (#31338) shaofeiqiandLi He 2026-10-31 22:26:47 -07:00
a91eab67d8 hexagon : fix DMA transfers with strides > 24 bits (#31442) Mark TodorovichandMax Krasnyansky 2026-10-31 16:49:23 -07:00
d5ef408e20 mtmd : fix InternVL CLS token order (#30886) bosh 2026-11-01 06:30:01 +07:00
79bce0c881 tests : enable Gemma 4 in test-llama-archs (#31401) Niklas Wenzel 2026-10-31 23:55:02 +01:00
bc428e407a llama : enable tensor split for LFM3 (#31431) Tarek Dakhran 2026-10-31 23:40:50 +01:00
f6312ad75e vendor : update miniaudio to 0.11.26 (#31459) Alessandro de Oliveira Faria (A.K.A.CABELO) 2026-10-31 19:37:20 -03:00
0138937e3f llama : fix rotation of attention tensors in layers with shared KV (#31406) Niklas Wenzel 2026-10-31 23:30:49 +01:00
ddf48f85c9 reshaping so mm on 2d instead of 3d tensor for lfm sc (#30831) Anthony Maltsev 2026-10-31 12:34:57 -07:00
f3586e66bc server: cap shared decision prefixes at the first output (#30710) aiqubits 2026-11-01 02:09:48 +08:00
2871678a1d server : fix /v1/responses endpoint to accept extra_params (#30959) David Garcia 2026-10-31 14:08:08 -04:00
7bd1808308 mtmd : resize Gemma 4 images without letterbox padding (#30922) boshandXuan-Son Nguyen 2026-11-01 00:55:33 +07:00
0ec4027a2a opencl: add fwht (#31434) shaofeiqiandLi He 2026-10-30 22:13:48 -07:00
1052f7e1d6 chat : strip the think block in the qwen3-coder with reasoning_format none (#31380) Tarek Dakhran 2026-10-31 04:38:04 +01:00
8956c769e8 sycl: fix f32 to q2_0/q4_0/q5_0 rounding in CPY (#31120) Jesse Symons 2026-10-31 12:54:41 +11:00
5047748f11 rpc : delete fs_path_to_utf8() (#31424) Adrien Gallouët 2026-10-30 20:36:30 +01:00
792fcf4e75 opencl: add initial iq1 support (#31341) shaofeiqiandLi He 2026-10-30 10:43:30 -07:00
a34de8618b common : delete fs_path_to_utf8() (#31423) Adrien Gallouët 2026-10-30 16:20:04 +01:00
aa588a3463 common : drop the Windows argv reconstruction (#31417) Adrien Gallouët 2026-10-30 14:02:00 +01:00
889f2ec96d quant : allow quantization for LFM3-Audio codec decoder (#31386) Tarek Dakhran 2026-10-30 11:52:59 +01:00
1ab23c33b7 vendor : update cpp-httplib to 0.64.0 (#31400) Alessandro de Oliveira Faria (A.K.A.CABELO) 2026-10-30 07:49:58 -03:00
27b8464e82 ci : add missing ccache-clear to windows behavior check (#31413) Sigbjørn Skjæret 2026-10-30 11:34:25 +01:00
d63fbfe794 cmake : embed an app manifest in all Windows exes (#31329) Adrien Gallouët 2026-10-30 10:56:53 +01:00
e216f4ec87 ci : extract most apple jobs into daily workflow (#31408) Daniel Bevenius 2026-10-30 10:01:28 +01:00
344846b4fc gemma3n, gemma4 : create token_embd before the tied output (#31394) Lorenzo Bona 2026-10-30 05:36:27 -03:00
fb5997274e kv-cache : keep rotation constants on the GPU (#31073) Anton IandPascal 2026-10-30 04:15:06 -04:00
ab3768cdc2 memory: refuse rollbacks past the snapshots of the last ubatch (#31368) Pascal 2026-10-30 09:11:58 +01:00
e0f142d5dd opencl: add initial q2_0 support (#31340) shaofeiqiandLi He 2026-10-29 21:30:05 -07:00
3c809f4438 common : end reasoning budget forcing on an end tag from the prompt (#31374) Tarek Dakhran 2026-10-30 01:27:50 +01:00
36a0e2ffc0 sycl: FLASH_ATTN_EXT failures (#30761) Michael Taylor 2026-10-29 18:37:28 -03:00
418f905a4e ggml : add batched tensor copy API (#30881) Georgi Gerganov 2026-10-29 20:52:17 +02:00
eb4e244be7 ROCm/HIP: add RDNA4_M family for gfx1170/gfx1171 (#31373) Slobodan JosicandMario Limonciello 2026-10-29 18:23:29 +01:00
68d20b884c Quant: OCP FP8 E4M3 support (#27512) Oliver SimonsandSigbjørn Skjæret 2026-10-29 18:14:37 +01:00
0893ec98f5 skills: specify rule about ggml_build_forward_expand (#31375) Xuan-Son NguyenandGeorgi Gerganov 2026-10-29 15:44:12 +01:00
78feec062d model: mistral le chonk (#31376) Xuan-Son Nguyen 2026-10-29 15:43:33 +01:00
0ac1a7fd35 llama : reuse the tensor staging buffer and skip the direct-io bounce buffer (#31321) Aman Gupta 2026-10-29 19:22:43 +08:00
7df8b876b9 server : restore the spec checkpoint for drafts longer than n_rs_seq (#31350) 7konigsbergandZeshawn Zahid 2026-10-29 05:28:03 -04:00
8ab29faec4 ui: keep chats at the bottom until the reader scrolls up (#31311) Aleksander Grygier 2026-10-29 09:18:13 +01:00
f5d4d75419 ui: show submit button when the chat input has content (#31324) Aleksander Grygier 2026-10-29 09:15:54 +01:00
3ed915e309 ui: reveal the collapsed sidebar from its logo and toggle it there (#31307) Aleksander Grygier 2026-10-29 09:15:31 +01:00
f8fcef2546 ui: add MCP group checkboxes in tools settings and fix indeterminate checkbox colors (#31303) Aleksander Grygier 2026-10-29 09:15:09 +01:00
b7cde589e4 ui: request persistent storage on startup (#31299) Aleksander Grygier 2026-10-29 09:14:53 +01:00
3dd0f60bb4 doc: update ops for opencl (#31347) lhez 2026-10-29 01:14:29 -07:00
78febbad9d HIP/CUDA: enable sparse attn on CDNA (#31259) uvos 2026-10-29 09:14:11 +01:00
abb76ab380 model : Support Limite 1B - Violetto (#29433) Giuseppe Scrivano 2026-10-29 09:13:41 +01:00
41be0bdc98 common: select A100 cores for SpaceMiT default threads (#30284) Keyu Zhu 2026-10-29 16:12:42 +08:00
22183e51f9 opencl: initial support for iq2 (#31267) lhezandHongqiang Wang 2026-10-29 01:12:11 -07:00
6d194531e1 metal : faster MXFP4 mat-vec, full threadgroups for unary/binary/GLU (#31075) Aman Gupta 2026-10-29 16:11:29 +08:00
fade717ed8 tests : remove duplicate FA test cases added in #31125 (#31357) Jesse Symons 2026-10-29 19:10:25 +11:00
c9daee149d ggml-metal : fix unary kernels compile on older Metal toolchains (#31360) Georgi Gerganov 2026-10-29 10:09:51 +02:00
8a199ad31a hexagon: support for UPSCALE op (#31331) Max Krasnyansky 2026-10-29 00:22:34 -07:00
da6c30ade8 hexagon: improve DDR bw utilization in transposed CONCAT (#31277) Trivikram ReddyandMax Krasnyansky 2026-10-29 01:57:18 -05:00
f7c01c5226 [SYCL] add DPAS flash attention kernel for quantized KV cache (#31125) Jesse Symons 2026-10-29 17:01:21 +11:00
bf48103ad1 sycl : vectorize reordered-layout dequantization stores (Q4_0, K-quants, IQ3, MXFP4) (#30976) Anant Shrivastava 2026-10-29 11:27:53 +05:30
6d5ac3c8d5 tests : fix comp_padding CUDA graph reservation (#31344) Yonghye Kwon 2026-10-29 14:04:28 +09:00
7ac5727d60 ggml-openvino: slice view inputs of SCALE (#31334) Ravi Panchumarthy 2026-10-28 21:11:13 -07:00
2e0d86bebc Hexagon: IM2COL improvement - block based HVX parallelization (#31050) Marco ColomboandMax Krasnyansky 2026-10-29 03:04:18 +01:00
e09a78a25d tests: fix CI flakes in server cancel and sleep tests (#31330) Pascal 2026-10-29 01:35:01 +01:00
369dcf309c ggml-cpu: add SVE implementation of ggml_gemv_q8_0_4x8_q8_0 (#30585) Nikhil Sharma 2026-10-29 05:51:05 +05:30
d0ff6317cc docs : fix DSpark confidence threshold flag name (--spec-draft-p-min) (#31178) Abhijith Jajam 2026-10-29 05:47:38 +05:30
cf24317bea common, mtmd: bound control vector layer indices and M-RoPE position loops (#31336) Pascal 2026-10-28 23:15:01 +01:00
74a4fb0037 opencl: add tiled q4_k GEMM for adreno (#30412) Hongqiang WangandLi He 2026-10-28 14:55:28 -07:00
5220213255 hexagon: add conv2d support (#30995)
Todor Boinovski Pranav Dhinakar Max Krasnyansky
2026-10-28 13:43:40 -07:00
c3df4b7e22 Speed up IQ4 quant creation with pre-computed halfway points (#31201) Bartowski 2026-10-28 14:35:56 -04:00
ac8e1694d8 ggml-cpu: SVE 256 bit implementation of gemm_q6_K_8x8_q8_K kernel (#27472) anubhavfujitsu 2026-10-28 21:31:37 +05:30
c02b5c641e ggml-cpu: SVE 256-bit Implementation of mxfp4 gemv kernel (#28141) abhijain1204fujitsuandAnubhav Srivastava 2026-10-28 21:28:20 +05:30
ff0d396e3d ci(s390x): switch to Ubuntu 26.04 runners (#31317) Aaron Teo 2026-10-28 23:32:38 +08:00
838093f192 llama : remove simple batch splitting for DSV4 (#31300) Yonghye Kwon 2026-10-29 00:25:47 +09:00
604cc43023 codeowners : refresh teams and maintainer PRs link (#31316) Georgi Gerganov 2026-10-28 17:00:54 +02:00
39febd2adb sycl: work-group argmax for short TOP_K rows (#31110) Frosty40 2026-10-28 09:18:07 -05:00
f5222f25c1 cuda: fuse MoE (MUL_MAT_ID) gate/up and GLU in MMQ (#31188) anujj 2026-10-28 18:34:28 +05:30
28cb47c7c8 gguf-py: fix np_roundf dropping the sign (#31310) Xuan-Son NguyenandSigbjørn Skjæret 2026-10-28 14:03:59 +01:00
8642541b25 llama.cpp : bump version to 0.8.0 (#31309) Georgi Gerganov 2026-10-28 13:30:38 +02:00
d4f1b18b3a mtmd: enable video input support for LFM3 model family (#31283) Tarek DakhranandXuan Son Nguyen 2026-10-28 12:23:06 +01:00
d1eb0debad server : support tool calls in /v1/realtime (#31287) Tarek DakhranandXuan Son Nguyen 2026-10-28 12:17:13 +01:00
5d7793d6fc sync : ggml Georgi Gerganov 2026-10-28 12:34:54 +02:00
ec941ee852 ggml : bump version to v.28.0 (ggml/1676) Georgi Gerganov 2026-10-28 12:24:54 +02:00
e486a4b559 ui: build the reasoning menu from the template effort levels (#31273) Pascal 2026-10-28 10:35:57 +01:00
f1a7dea99a tests : adjust constants to reduce irrelevant warnings (#31254) Georgi Gerganov 2026-10-28 11:26:38 +02:00
a7af149a2a cuda : add top-k MoE fusion for 384 experts (#31288) Yonghye Kwon 2026-10-28 17:13:01 +09:00
2864937a77 vulkan: clamp A prefetch to end_k in coopmat1 int8 matmul (#30283) Kris 2026-10-28 09:12:22 +01:00
a71c00f0bb chat : stop qwen3-coder reasoning and content before template separators (#30557)
7konigsberg Zeshawn Zahid Alde Rojas
2026-10-28 04:11:47 -04:00
f4d8af608a ui : show save and reset settings buttons only when relevant (#31248) Aleksander Grygier 2026-10-28 07:54:34 +01:00
02d502f314 ui: Support OGG audio attachments by converting them to WAV (#31237) Aleksander Grygier 2026-10-28 07:54:15 +01:00
b1756b23f3 ui: Fix draft for new message disappearing after using "Add System Message" action (#31214) Aleksander Grygier 2026-10-28 07:53:56 +01:00
6886628cb1 ui : Search dialog (#31173) Aleksander Grygier 2026-10-28 07:53:41 +01:00
8ca4060b7e server : add realtime conversation and transcription on /v1/realtime (#31269) Tarek DakhranandXuan Son Nguyen 2026-10-28 01:26:12 +01:00
d05706b4f3 model : add LiquidAI/d1-2B and LiquidAI/d1-omni-2B (#31275) Tarek DakhranandXuan Son Nguyen 2026-10-27 23:36:32 +01:00
1b4554eff1 server: expose the reasoning effort levels of the chat template (#31271) Pascal 2026-10-27 22:39:37 +01:00
a919179942 tests: fix CI riscv timeout in test-llama-context (#31265) Pascal 2026-10-27 22:18:31 +01:00
338fce6511 HIP/CUDA: remove pragma unroll from loops in conv2d where itteration count cant be infered at compile time (#31255) uvos 2026-10-27 19:42:04 +01:00
19ff817fb2 model : add LiquidAI/LFM3-Audio-2B and LiquidAI/LFM3-Omni-2B (#31257) Tarek DakhranandXuan Son Nguyen 2026-10-27 18:47:14 +01:00
eb7818c0b8 common : support deferred tools and tool_reference content parts (#31256) Tarek DakhranandAldehir Rojas 2026-10-27 18:39:11 +01:00
6c44591049 snapdragon: switch QDC artifact upload and log download to v2 API (#31240) Mengsheng Wu 2026-10-28 01:06:41 +08:00
19cce69199 model : add LiquidAI/LFM3-1.7B and LiquidAI/LFM3-VL-2B (#31252)
Tarek Dakhran Sigbjørn Skjæret Xuan-Son Nguyen Tarek Dakhran
2026-10-27 17:06:40 +01:00

1 2 3 4 5 ...

Commit Graph Select branches Hide Pull Requests master Mono Color

Commit Graph

Select branches

Hide Pull Requests

master