llama.cpp-xrip

wmantly/llama.cpp-xrip

0 Fork 0

683bbc0dce gguf-py : detect nfc normalization (#30540) Sigbjørn Skjæret 2026-10-14 20:08:09 +02:00
529ac893e7 sycl: IQ3 code reorder (#29107) Anant Shrivastava 2026-10-14 22:11:23 +05:30
685f70894b ggml-cpu : keep the FP16->FP32 lookup table on x86 with F16C (#30342) (#30343) Leebr Data Consulting 2026-10-14 18:36:06 +02:00
c114bd4c58 docs: update IBM zDNN backend docs (#30527) Aaron Teo 2026-10-15 00:17:50 +08:00
5cdadfc326 ggml-cpu : size set_rows conversion buffer for all threads (#30454) Aman Gupta 2026-10-14 21:01:53 +05:30
62df1815bc cuda, hip: GDN chunked kernel (#29353)
Aman Gupta Aman Gupta gianni-cor
2026-10-14 19:30:58 +05:30
37d6d653e4 skill: Update code-review skill for UI development (#30509) Aleksander Grygier 2026-10-14 15:20:27 +02:00
6868a52cb3 ggml: fix conv_1d result layout for batched input (#28738) Pascal 2026-10-14 13:16:08 +02:00
3fb47677ea common : add --spec-draft-n-ubatch (#30459) Georgi Gerganov 2026-10-14 13:58:05 +03:00
318a471a62 docs : clarify installation recommendation for Intel oneAPI Toolkit (#30407) Francisco Herrera 2026-10-14 05:19:10 -05:00
8c2134a875 ggml : add note about mul_mat_id + disable mul_mat_id_dup tests (#30501) Georgi Gerganov 2026-10-14 12:36:52 +03:00
df18bee11e model: split layers evenly when no device reports free memory (#29677) Pascal 2026-10-14 08:30:00 +02:00
3eb101360f vendor : update cpp-httplib to 0.61.0 (#30494) Adrien Gallouët 2026-10-14 08:28:24 +02:00
af0579c78a ggml : improve RPC tensor transfer logs (#30481) Georgi Gerganov 2026-10-14 09:23:44 +03:00
62bacfa011 cuda : use SIMT implicit GEMM conv2d on AMD without WMMA (#30325) An Longandthelittlefireman 2026-10-14 15:22:11 +09:00
77390baa18 jinja : write floats in tojson as Python does (#30362) Yuri Khrustalev 2026-10-14 02:18:38 -04:00
c5a997624a opencl: use buf+offset instead of subbuffer for adreno moe expert sort (#30418) lhez 2026-10-13 23:17:09 -07:00
21a19cd3be opencl: fix q8_0 dp4a mul_mat_id and mark PREC_F32 mul_mat_id unsupported for adreno (#30365) lhezandHongqiang Wang 2026-10-13 23:16:54 -07:00
9cf967a04d models: drop the t_embd expand of the embedding graphs (#30461) Pascal 2026-10-14 08:15:08 +02:00
4dd6dd4b96 HIP: use 64-row MMQ tiles for some quant types on RDNA3.5 (#28195) liangliangchang 2026-10-13 23:10:29 -07:00
14310190cd vulkan: fix compile error (#30495) Ruben Ortlam 2026-10-14 08:06:09 +02:00
e7268bf6a9 llama : build the k-pool selection in the causal mask layout (#30480) Georgi Gerganov 2026-10-14 08:45:55 +03:00
cdce056a3e vulkan: Add workaround for device free memory > total memory on Intel Windows (#29835) Masato Nakasaka 2026-10-14 14:45:38 +09:00
d524bb660b vulkan: Define the shader count for Intel Arc Pro B65 + B70 (#30431) Matt Corallo 2026-10-14 05:44:22 +00:00
7ec420bc09 vulkan: support mulmat bf16 activations (#29069) Ruben Ortlam 2026-10-14 07:36:32 +02:00
10da8c463f unicode : add opt-in NFC normalization for Unigram tokenizers (#30390) Sandro Hanea 2026-10-14 07:33:24 +02:00
eca5458ce0 convert : export YaRN truncation config (#30288) Takamichi Ohara 2026-10-14 14:31:40 +09:00
091b6a7fc5 Enable OpenSSL in FP32/FP16 SYCL builds to restore HF HTTPS downloads (#30330) Raffaele Di DonnaandRaffaele Di Donna 2026-10-14 04:25:26 +02:00
fcac19079f webgpu: (Fix CI) fix mul_mat_id with duplicate expert ids (#30451) Masashi Yoshimura 2026-10-14 08:24:23 +09:00
04f7226a2c server : escape the special tokens without std::regex (#30243) Yuri Khrustalev 2026-10-13 19:17:11 -04:00
ff026f8c0a chat: replace output string with parts (#30389) Xuan-Son Nguyen 2026-10-14 01:03:56 +02:00
1b70350c92 ggml : simplify backend search path error handling (#28367) Adrien Gallouët 2026-10-13 20:22:19 +02:00
d71e324a44 docs : update zDNN supported ops (#30472) Aaron Teo 2026-10-14 02:20:41 +08:00
45451c366d kleidiai: add SME2 Q4_K and Q6_K matmul support (#29219) Charles Xu 2026-10-13 18:37:48 +02:00
77c4540ce4 vulkan: Optimize pool_2d when the pooling window is large (#30119) Jeff Bolz 2026-10-13 10:17:59 -05:00
ecdd29e8b2 meta: recreate all subgraphs when the arena is reallocated (#30453) Igor Matuszewski 2026-10-13 16:58:35 +02:00
ec0bfac179 hexagon: add configurable HTP DCVS/HMX power mode (#30423) Pranav DhinakarandMax Krasnyansky 2026-10-13 07:47:55 -07:00
ace20f679e graph: pad the KQ bias of the no-cache attention (#30446) Pascal 2026-10-13 16:34:10 +02:00
b9f3599fb1 llama : do not map foreign ranges into a backend buffer (#30405) Georgi Gerganov 2026-10-13 17:28:15 +03:00
f441d2d9db model: load YaRN beta metadata for laguna models (#30331) Adam CaldwellandAdam Caldwell 2026-10-13 07:18:01 -07:00
cb55ffe6e2 graph: skip the LM head for pooled embeddings (#30445) Pascal 2026-10-13 16:16:14 +02:00
a17e94d229 tests : consolidate context state tests into test-llama-context (#30023) Georgi Gerganov 2026-10-13 16:21:55 +03:00
b476d9a762 llama : fix log message (#30452) Georgi Gerganov 2026-10-13 16:07:54 +03:00
e0bcaccbe5 vulkan: allow multiple devices to access gpu-visible host memory (#29741) Ruben Ortlam 2026-10-13 15:05:10 +02:00
0a731cced9 ggml-cpu : add AVX2 dot product for Q2_0 (#30447) homeofeandSigbjørn Skjæret 2026-10-13 14:36:47 +02:00
f851e244d0 cuda : add GGML_CUDA_GRAPH_DEBUG and read env vars once (#30327) Georgi Gerganov 2026-10-13 15:24:34 +03:00
bfe65e1dfa cuda : fix compiler warnings in mmq-load-tiles.cuh (#30443) Adrien Gallouët 2026-10-13 13:48:49 +02:00
411d548515 vulkan : split matmul dispatch when it would exceed maxComputeWorkGroupCount (#29533) losewayy 2026-10-13 19:03:47 +08:00
296538adea spec : remove tokens assert from dsv4 graph_mtp (#30444) Oliver Simons 2026-10-13 12:39:10 +02:00
7f51953d7d model : allow Qwen3.8 Flash Next to accept image embds for MTP (#30397) Oliver Simons 2026-10-13 12:34:47 +02:00
882d67166d cuda : fix warnings from dead breaks (#30441) Adrien Gallouët 2026-10-13 12:29:27 +02:00
30f7df1a83 cuda : warnings cleanup (#30420) Adrien GallouëtandSigbjørn Skjæret 2026-10-13 11:44:37 +02:00
9d7d504980 convert : fix LoRA conversion with --outtype auto (#30433) Chanho Lee 2026-10-13 17:29:09 +09:00
ed019cd8b9 llama : skip logits buffer allocation when graph produces no logits (#30255) Animesh 2026-10-13 13:54:20 +05:30
9880157881 graph : avoid reallocations in the "no cache" path (#30439) Georgi Gerganov 2026-10-13 11:22:19 +03:00
af50f2f248 ggml-zdnn : introduce Q8_0 mat mul (#29836) Aaron Teo 2026-10-13 15:53:48 +08:00
071b553fb1 models : remove redundant ggml_cont calls now supported by all unary kernels (#30198) Muhammad SaadandSigbjørn Skjæret 2026-10-13 03:26:52 -04:00
4413a19241 hexagon: extend scalar DMA broadcasting support for binary ops (#30245) kurquharandMax Krasnyansky 2026-10-12 20:33:05 -07:00
8f574b0ad4 ci : build OpenVINO releases with GGML_NATIVE=OFF and CPU backend variants (#30324) Pratyush Kumar 2026-10-13 07:52:05 +05:30
a49f58dcc3 ci: fix OpenVINO download links (#30401) Pratyush Kumar 2026-10-13 07:50:04 +05:30
cbe28bb3cb model: fix vllm d3 multi seq case (#30408) Xuan-Son Nguyen 2026-10-13 00:23:01 +02:00
358a420ec4 graph: pad the KV dimension of the no-cache attention to 256 (#30416) Pascal 2026-10-12 23:00:27 +02:00
48e9857567 metal: fix metal tensor kernel_mul_mm error for large tensor (#28748)
SecondSeed YiChen Lv 向锡铭
2026-10-13 03:57:26 +08:00
8bb3cf0d21 hexagon: optimize CPY/CONCAT — round 3, short-rows edition (#30341)
Max Krasnyansky Ruslan Karymov Yiwei Shao
2026-10-12 07:39:55 -07:00
0f701c8829 vulkan: pack FMA for f16 accumulation, no dot product op (#29877) Alex Morgan 2026-10-12 09:19:20 -05:00
efe3626b11 model: support vllm d3 models (#30383) Xuan-Son Nguyen 2026-10-12 15:01:52 +02:00
097cf33d94 model: support vllm decision 2 model (#30378) Xuan-Son Nguyen 2026-10-12 13:39:55 +02:00
ccdbe29053 tests : link each test against only the libraries it uses (#30299) Georgi Gerganov 2026-10-12 13:32:56 +03:00
073b74034b metal : fix mul_mat_id with duplicate ids (#30373) Georgi Gerganov 2026-10-12 13:31:41 +03:00
77757a3467 examples : fix draft sampling and rejection handling (#30095) Daniel Bevenius 2026-10-12 12:02:24 +02:00
972b5fcc19 llama : report compute graph reallocations during inference (#30329) Georgi Gerganov 2026-10-12 11:53:58 +03:00
2f160ba59a ui: fix text attachment preview overflow (#29898) shamnadxyz 2026-10-12 13:50:16 +05:30
f9a95aeb20 sycl: guard DMMV reads at row tails (#29338) Frosty40 2026-10-12 03:06:27 -05:00
08ffba21e5 vulkan: use ternary expressions to avoid switch lowering for CM2 quant block size (#30164) Amadeus dyw 2026-10-12 14:00:26 +08:00
a05e58c981 ggml: Change default visibility of symbols to hidden (#30014) Christian Kastner 2026-10-12 07:50:12 +02:00
b2f036c0f1 vulkan: fix buffer logging on 32-bit platforms after source split (#30326) miyan 2026-10-12 12:51:24 +08:00
e937b17a2d hexagon: optimize q2_k bpw and activation (#30035) Jhen-Jie Hong 2026-10-12 11:52:01 +08:00
b13a93a3d5 hexagon: add IM2COL_3D (#30305) kurquharandMax Krasnyansky 2026-10-11 20:24:47 -07:00
f2cc7282ce opencl: improve fa, allow dk512 for gemma-4, improve dk64 (#30266) lhezandHongqiang Wang 2026-10-11 11:49:25 -07:00
c83f3058b6 opencl: fix image limit for q4_k dense bin kernels, q4_0 bcast and rms_norm (#30310) lhezandHongqiang Wang 2026-10-11 09:41:29 -07:00
5c8854fd88 rpc : turn GGML_RPC_DEBUG into a verbosity level and add logs (#29544) Georgi Gerganov 2026-10-11 17:43:35 +03:00
a9d27ac693 model : support for Prism Bonsai 2 27B (#29600) bri-prismandGeorgi Gerganov 2026-10-11 05:39:16 -07:00
1cf7767765 cpu: fix AVX2 CPUs not using F16C when available (#30270) Ploo 2026-10-11 12:24:41 +00:00
966baae76b vulkan: handle mul_mat_id duplicates in the prepass rather than looping (#29998) Jeff BolzandAnjielon 2026-10-11 07:02:57 -05:00
62a6f74bab server: add component tests (#30312) Xuan-Son Nguyen 2026-10-11 12:51:07 +02:00
23b0202a18 server: leave a busy slot untouched when a request pins it (#30295) Pascal 2026-10-10 21:25:38 +02:00
69f201a205 model : support MiniCPM-V 4.7 (#29416)
tc-mb Sigbjørn Skjæret Xuan Son Nguyen
2026-10-11 02:41:59 +08:00
abee0c8476 cmake(s390x): disable z17 target for unsupported compilers (#30297) Aaron Teo 2026-10-11 02:00:56 +08:00
aa94f20861 args: add LLAMA_ARG_SLOT_SAVE_PATH env for --slot-save-path (#30272) Mikolaj Kucharski 2026-10-10 13:34:40 +00:00
781dbc5ac9 spec: properly handle mtmd input for mtp (#30257) Xuan-Son Nguyen 2026-10-10 11:22:38 +02:00
0fd868cbca mtmd: add build_inp_attn_mask (#30259) Xuan-Son Nguyen 2026-10-10 11:22:12 +02:00
1623d8ce47 cuda: always use MMVQ for MUL_MAT_ID on sm_60 (#27828) David Friehs 2026-10-10 08:15:42 +02:00
1bb2b9fcbe CUDA/HIP: fix race in flash_attn_ext_f16_process_tile when nbatch_combine != DKQ/2 (#30103) uvos 2026-10-10 08:14:51 +02:00
2bbca8f202 ggml-cpu: vectorize fp32 to fp16 conversion (#30157) Aaron Teo 2026-10-10 14:08:52 +08:00
404f557b5b vulkan: use 4 rows for NVIDIA MUL_MAT_ID MMVQ (#29274) Amadeus dyw 2026-10-10 14:08:19 +08:00
b797c82c7d ggml: fix s390x all cpu build (#30140) Aaron Teo 2026-10-10 14:07:52 +08:00
f2918cabbf opencl: add bin kernels kernel_gemm_moe_q4_k_q8_1_dp4a_bin, kernel_gemm_moe_q6_k_q8_1_dp4a_bin (#30187) Shawn Gu 2026-10-09 21:02:11 -07:00
1e6f04a75e sycl : accelerate MXFP4 MoE with arithmetic decoding and weight reordering (#29809) Captain-Tripps 2026-10-09 21:56:23 -05:00
10a60cf303 vendor: apply deep nested json patch from upstream (#30253) Xuan-Son Nguyen 2026-10-10 00:07:44 +02:00
79e2e74eb1 CUDA: fix round issue, under MSVC the CPU and GPU agree (#30229) Dante 2026-10-09 23:29:50 +05:30

... 3 4 5 6 7 ...

Commit Graph Select branches Hide Pull Requests master Mono Color

Commit Graph

Select branches

Hide Pull Requests

master