Commit Graph

Select branches

Hide Pull Requests

master

Mono Color

• aa39d7a3e1 [SYCL] Fix function signature for ggml_backend_sycl_split_buffer_type (#28981) Michael Taylor 2026-09-16 23:15:47 -03:00
• 4bc272fd72 vulkan: work around NV bug with argsort_large.comp (#28975) Jeff Bolz 2026-09-16 18:40:07 -05:00
• fb27a525d2 TP: fix split state and granularity for fused QKV gemma4, qwen35 (#28965) David Friehs 2026-09-16 21:02:12 +02:00
• c6824a9e42 ci: switch fast jobs back to github (#28959) Eve 2026-09-16 16:50:26 +00:00
• 2f3fd02526 Enable CUDA graph for MTP draft (#28549) Gaurav Garg 2026-09-16 21:46:54 +05:30
• 1ec8188094 hexagon: Support for K-Quants Q4_K and Q6_K (#28994) Marco ColomboandMax Krasnyansky 2026-09-16 18:00:31 +02:00
• 82324fc508 hexagon: accept the zeroed rope probe in supports_op (#28995) Marco Colombo 2026-09-16 17:38:44 +02:00
• 7ceed8737f models : allow Nemotron-H models to only define layer_norm_epsilon (#28989) Kartik Gulia 2026-09-16 18:54:47 +05:30
• 7d6f5d02bb model : add support for HrmTextForCausalLM (DFM Mimir 1B) (#27625) GeorgeandSigbjørn Skjæret 2026-09-16 16:18:45 +03:00
• 83078fec0d CUDA/HIP: improve access patterns in im2col (#28013) uvos 2026-09-16 13:46:21 +02:00
• f266648fa9 spacemit : fix wrong transpose function for int16 data (#25161) I3eg1nner 2026-09-16 19:19:47 +08:00
• 60199339bc rpc : invalidate cached compute graph when a referenced buffer is freed (#24292) y198 2026-09-16 18:03:11 +07:00
• b04d4e567c Change max context length for auto-fitting with unified KV (#28849) Gaurav Garg 2026-09-16 16:08:50 +05:30
• 37b53fd454 qwen4exp: add hc ops (#28901) Aman Gupta 2026-09-16 16:00:01 +08:00
• fccf7166fb HIP: broaden MoE ncols_opt tile heuristic on RDNA3.5 architecture (#28935) WenqiangJia2026 2026-09-16 15:55:02 +08:00
• 0bec16e388 chat : force \n</think> on reasoning budget end for qwen3-coder (#28869) Aldehir Rojas 2026-09-16 01:47:28 -05:00
• d4365d9554 vulkan: make MUL_MAT_ID BN/2 tail unconditional (#28923) SG-Amadeus 2026-09-16 14:45:44 +08:00
• 0a8b29a607 metal: fix NaN in mul_mm_id when activations exceed f16 range (#26223) 3 people 2026-09-16 08:37:40 +02:00
• 583926e3ac ci : add self-hosted webgpu to hf-jobs (#28712) Sigbjørn SkjæretandGeorgi Gerganov 2026-09-16 08:23:58 +02:00
• e13469a323 llama-bench: support --version to print build info (#28971) asbelin 2026-09-16 08:39:43 +03:00
• 930e2fa599 hexagon: add back missing contiguous fast-path and hvx_copy_uu for each run (#28886) Jhen-Jie Hong 2026-09-16 07:02:06 +08:00
• 72b590d65f hex-cpy: use dma if src and dst are contiguous (#28906) Trivikram Reddy 2026-09-15 17:45:28 -05:00
• 38a5b42d9a HIP: Enable AllReduce for ROCm (#27825) Sandro Steeger 2026-09-15 20:57:41 +02:00
• 9f31776c37 opencl: choose the MoE expert matmul by batch size for speculative decoding/MTP (#27637) Hongqiang WangandLi He 2026-09-15 11:21:05 -07:00
• d1d3c3396a ci: build MUSA for only 1 arch (#28944) Aman Gupta 2026-09-15 21:48:15 +08:00
• 6011c34ce6 docs: Rule of thumb for AI review time [no ci] (#28945) Johannes Gäßler 2026-09-15 14:11:16 +02:00
• 7609846557 rpc : hash-cache only weights (#28789) 3 people 2026-09-15 13:50:20 +02:00
• 5431581326 cuda: support row-contiguous SUM_ROWS (#26308) Mohamed Elashri 2026-09-15 12:39:29 +02:00
• 9e71716247 models : move build_arch_graph() after graph() template specialization (#28934) Chris Peterson 2026-09-15 02:33:26 -07:00
• fc82583e65 vulkan: support sparse Flash Attention (#28105) Ruben Ortlam 2026-09-15 11:30:27 +02:00
• 77d554b26d OpenVINO: optimize stateful decode and GPU MoE inference (#28638) 5 people 2026-09-15 17:29:19 +08:00
• 6ec1a7e956 opencl: add generic ssm_scan (#28881) lhez 2026-09-15 02:29:02 -07:00
• 1af6c65de0 ci: bump kleidiai runners from 22.04 to 24.04 (#28885) Aaron Teo 2026-09-15 17:23:16 +08:00
• 1e7bcf3da4 metal : add FA kernels for HSK=96, HSV=64 (MiniCPM3) (#28599) Yanzhao Wang 2026-09-15 01:10:40 -07:00
• 0ecb159c9e ci: Bump CUDA Windows x64 builds to 13.4.1 (#28930) shivamkumard-ctrl 2026-09-15 13:34:17 +05:30
• 987498f459 ci : fix android release (#28936) Sigbjørn Skjæret 2026-09-15 09:04:35 +02:00
• 4c9233c034 cuda : enable i16 and i32 for DUP (#28897) Aman Karki 2026-09-15 09:12:21 +05:30
• 69eb250670 cmake : use PROJECT_SOURCE_DIR instead of CMAKE_SOURCE_DIR (#28771) Daniel Bevenius 2026-09-15 05:26:09 +02:00
• 1bc7a5af0d webui: stop re-probing disabled /tools endpoint on every message (#28646) Abhiram 2026-09-15 04:41:26 +05:30
• 7cf1c54a96 ci : reuse build tag name when used instead of safe one (#28911) Sigbjørn Skjæret 2026-09-14 22:21:39 +02:00
• 96ffdc41ce CI: hip-quality-check: ignore spill added in bfdc32183d (#28909) uvos 2026-09-14 21:25:22 +02:00
• bfdc32183d HIP: fattn-mma: use fp32 accumulation on MFMA devices (#28576) uvos 2026-09-14 20:26:22 +02:00
• 391fac1646 ci : add ubuntu-cuda builds to release (#28186) Oliver SimonsandSigbjørn Skjæret 2026-09-14 19:07:53 +02:00
• 41abbfd599 qwen4exp: enable rms_norm + mul fusion (#28896) Aman Gupta 2026-09-14 22:16:30 +08:00
• b4fa47d226 release : added gfx1103 to ubuntu rocm build (#28423) Apoorv Parle 2026-09-14 07:09:46 -07:00
• f3a184b153 cmake : remove precompiled headers (#28892) Daniel Bevenius 2026-09-14 16:07:28 +02:00
• dfe45163e1 scripts: Add script to verify API/ABI compatibility (#28579) Christian Kastner 2026-09-14 16:02:30 +02:00
• b29c606e28 llama.cpp : bump version to 0.4.1 (#28900) Georgi Gerganov 2026-09-14 17:01:05 +03:00
• d9e03f1074 sync : ggml Georgi Gerganov 2026-09-14 16:43:29 +03:00
• eeea731613 ggml : bump version to 0.24.0 (ggml/1627) Georgi Gerganov 2026-09-14 16:34:19 +03:00
• bbdd9f246e tests : add fusion baseline README and broaden fusion CI triggers (#28893) Georgi Gerganov 2026-09-14 15:45:05 +03:00
• 97e4ca7358 models : fix incorrect uses of get_key_or_arr (#28868) Sigbjørn Skjæret 2026-09-14 13:05:37 +02:00
• 1aca1f9fcd models : fix mimo2 swa pattern load (#28865) Sigbjørn Skjæret 2026-09-14 13:05:17 +02:00
• be2c6d7d1f tests(s390x): add non-vxe build to tests (#28776) Aaron Teo 2026-09-14 19:04:58 +08:00
• 3d10bcd197 llama: add Maple 20B-A1B ternary MoE architecture (CPU) (#27000) Alex 2026-09-14 13:04:05 +02:00
• 21f6b0d22c sycl: rfc: Use radix select for top_k (#28670) cwriterandcwriter 2026-09-14 13:02:44 +02:00
• 2f539596c6 ggml-cpu : disable PCH and fix CACHE_LINE_SIZE ambiguity to fix heap corruption (#28882) Georgi Gerganov 2026-09-14 13:03:41 +03:00
• 89fe242405 ci : trigger self-hosted CI on changes to ci/run.sh (#28859) Georgi Gerganov 2026-09-14 11:51:06 +03:00
• 15d8f2d592 ci : remove gg_sum summary logic (#28857) Georgi Gerganov 2026-09-14 11:50:43 +03:00
• 661643e430 sycl : fix oneDNN scratchpad breaking the pool free order (#28704) Łukasz Ślusarczyk 2026-09-14 08:24:06 +02:00
• 093a2f86c3 common : move llama_n_rs_seq to before llama_decode (#28749) Daniel Bevenius 2026-09-14 05:24:05 +02:00
• ad6c66839a ggml-cuda: fallback to F32 on device without BF16 hardware acceleration (#28846) thelittlefiremanandJohannes Gäßler 2026-09-14 00:05:10 +02:00
• 7a16a6ce32 grammar : coalesce find + insert into a single insert and adjust move/copy mechanics (#26885) Clint Herron 2026-09-13 17:56:46 -04:00
• 5f436dddb4 tests : exclude HY_V4 from WebGPU test-llama-archs tests (#28855) fairydreamingandStanisław Szymczyk 2026-09-13 19:24:11 +02:00
• e49d2c2760 models : guard the expert FFN size fallback in nemotron-h against a zero divisor (#28779) Yaniss Amazouz 2026-09-13 20:20:50 +03:00
• 6978052985 ggml-cpu(s390x): guard VXE-only repack helpers (#28775) Bernard Ladenthin 2026-09-13 19:18:53 +02:00
• 243a3082d4 tests : fix typo in test-quant-type-selection for nemotron 3 nano (#28835) Michael Taylor 2026-09-13 13:50:46 -03:00
• b6b003d2cb sycl : Fix get mem error (#28227) 4 people 2026-09-13 23:31:34 +08:00
• c95f8e47b8 ci : run editorconfig and code-style checks on ubuntu-slim (#28854) Georgi Gerganov 2026-09-13 18:16:45 +03:00
• bc52a12b38 pi : prefer PI_MODEL_NAME env var for model disclosure (#28853) Georgi Gerganov 2026-09-13 18:13:32 +03:00
• 4a89937354 tests : reduce FA test sizes (#28842) Georgi Gerganov 2026-09-13 13:05:28 +03:00
• 37b3a9e0cc ci : remove leftover command (#28839) Sigbjørn Skjæret 2026-09-13 09:41:27 +02:00
• 002a12ad25 ci : cap test-backend-ops parallel jobs at 2 and add a 3600s timeout (#28833) Georgi Gerganov 2026-09-13 09:18:28 +03:00
• f1e44dcc11 vulkan: workaround NV queuesubmit driver bug (#28830) Jeff Bolz 2026-09-13 01:18:19 -05:00
• 56b9eb280a opencl: apply the noshuffle row-alignment rule to q4_K, q5_K and q8_0, not just q6_K (#28575) Hongqiang Wang 2026-09-12 21:33:23 -07:00
• 790cf51aab chat : improve parsing of complex types in qwen3-coder (#28742) Aldehir Rojas 2026-09-12 19:08:52 -05:00
• 8e330954ad common: add LOG_JSON macro to log structured data (#28586) Xuan-Son Nguyen 2026-09-13 01:36:34 +02:00
• acecd56032 common : implement common_schema internal representation for JSON schemas (#28736) Aldehir Rojas 2026-09-12 16:14:50 -05:00
• ae9afff8d2 jinja : support dot property integer literals (#28817) Sigbjørn Skjæret 2026-09-12 22:49:53 +02:00
• 737e0980fe cmake: leave the timestamp out of precompiled headers on clang (#28816) Pascal 2026-09-12 22:47:14 +02:00
• 3057bb66c8 ui : add cache (#28802) Adrien Gallouët 2026-09-12 16:09:46 +02:00
• 56381e407c server : allow model downloads at model limit fix issue #26809 (#28530) MiaoMing Chen 2026-09-12 17:50:35 +08:00
• c8edceb061 ggml-cuda: hip add specific config table for AMD GCN (#27841) thelittlefireman 2026-09-12 11:26:53 +02:00
• e192abb406 server : add missing headers (#28795) Adrien Gallouët 2026-09-12 11:23:54 +02:00
• 718f7b4175 vendor : update cpp-httplib to 0.56.0 (#28787) Alessandro de Oliveira Faria (A.K.A.CABELO) 2026-09-12 04:15:08 -03:00
• 2a3005c23f syscl : Handle (fail gracefully) unsupported tq1_0 quants (#28681) Michael Taylor 2026-09-12 04:05:38 -03:00
• f3a33dff26 rpc : fix linking when compiling with BUILD_SHARED_LIBS=OFF (#28492) Ed Addario 2026-09-12 07:22:57 +01:00
• c069aa7f5f server: frame the router child state command as a whole line (#28747) Pascal 2026-09-12 07:38:50 +02:00
• 8a56aedd61 opencl: fix several bugs where the backend aborts (#27630) Hongqiang Wang 2026-09-11 22:11:12 -07:00
• 07fc97716f opencl: add bin kernel kernel_gemm_noshuffle_q4_k_f32_32b_trans_ila_a8_bin (#28677) shaofeiqiandLi He 2026-09-11 22:10:08 -07:00
• 3f5e94d7c2 webgpu: align tensor bindings to the type block size (#28382) Pascal 2026-09-12 06:40:21 +02:00
• eafe15a5e3 hexagon: support for multi-device model split (aka row-split) (#28589) Max KrasnyanskyandAlexander Lu 2026-09-11 20:46:51 -07:00
• d3146f2b56 ggml-webgpu: Update to a recent version of Dawn (#28683) Mendy BergerandMasashi Yoshimura 2026-09-11 21:47:29 -04:00
• 82d6bb284d server: refactor subproc handling (#28555) Xuan-Son NguyenandPascal 2026-09-12 00:53:07 +02:00
• 8ea290247c cmake : skip PCH for llama-server PCH when using MSVC (#28763) 3 people 2026-09-11 21:36:52 +02:00
• b78a39a2f9 ci : run test-backend-ops as a dedicated ci/run.sh test (#28740) Georgi Gerganov 2026-09-11 22:00:57 +03:00
• 982937a333 tests: extend test-quantize-fns to test nrc=2 (i8mm) kernels (#16234) Rohanjames1997 2026-09-11 13:19:37 -05:00
• 8172e6577a tests: tolerate a shared pool abort in test_completion_unified (#28759) Pascal 2026-09-11 15:50:12 +02:00
• 43f3dda623 ggml: skip 0-sized ids tensor when offloading selected experts (#28739) Aman Gupta 2026-09-11 21:17:08 +08:00
• 5bda51bfbc metal : skip the empty half of the mul_mm_id token tile (#28301) Foad Abo Dahood 2026-09-11 14:12:55 +03:00