]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/shortlog
pkg/ggml/sources/whisper.cpp
2026-08-04 Neo Zhangsycl : support dev2dev memcpy by DEV2DEV_MEMCPY_FORWARD...
2026-08-04 Neo ZhangSupport q2 mul_mat (llama/26231)
2026-08-04 Titaniumtownsycl: fuse RMS_NORM + MUL (llama/26015)
2026-08-04 Masashi Yoshimuraggml-webgpu: improve flash_attn_vec for quantized KV...
2026-08-04 Jeff Bolzvulkan: Support quantized concat (llama/25684)
2026-08-04 pmaybankTest support for alternative conv layout (llama/25617)
2026-08-04 Robert Esclapezggml-cuda: Allow transpose-free gemmv computation ...
2026-08-03 Daniel Beveniusci : set GGML_NATIVE=OFF for build-vad workflow (#3966)
2026-08-03 Igor Koshenskiiexamples : fix VAD min silence argument parsing (#3963)
2026-07-31 Georgi Gerganovsync : ggml (#3962)
2026-07-30 Bhargav Krishparakeet : verify hparams loaded from parakeet model...
2026-07-29 ToddAdding ruby bindings change log (#3959)
2026-07-28 KITAITI Makotoruby : add VAD speech segments API (#3931)
2026-07-11 Cappuccinoexamples : Remove leading space from txt output (#3921)
2026-07-10 Georgi Gerganovggml : use ggml_vqtbl1q_u8 for 32-bit compat (#0)
2026-07-10 Georgi Gerganovtalk-llama : sync llama.cpp
2026-07-10 Georgi Gerganovsync : ggml
2026-07-10 Georgi Gerganovggml : bump version to 0.16.0 (ggml/1559)
2026-07-10 Georgi Gerganovggml : fix conv 2d dw (llama/25490)
2026-07-10 Georgi Gerganovcommon : adapt to q2_0 (ggml/0)
2026-07-10 Oliver SimonsOnly index by compile times + always multiply/add ...
2026-07-10 Sou-lymetal : add CONV_2D_DW (depthwise convolution) support...
2026-07-10 RapidMarkggml-hip: enable -funsafe-math-optimizations (llama...
2026-07-10 Pascalcuda: align snake fusion matcher with the other backend...
2026-07-10 Aparna M Phexagon: add VISION RoPE support (llama/25216)
2026-07-10 Masashi Yoshimuraggml-webgpu: tune subgroup split (d_split) in flash_att...
2026-07-10 Hongqiang Wangopencl: Q6_K GEMM/GEMV fix for ne01 of weights that...
2026-07-10 Ruben Ortlamvulkan: disable FA mask_opt on GCN to improve performan...
2026-07-10 Hongqiang Wangopencl: ragged-tile MoE prefill FP16 GEMM optimization...
2026-07-10 Ruben Ortlamvulkan: for small AMD GPUs, reduce submission threshold...
2026-07-10 Max Krasnyanskyhexagon: new vtcm layouts and improved pipelines for...
2026-07-10 fairydreamingcuda : add support for f16->f16 GGML_OP_SET_ROWS (llama...
2026-07-10 fairydreamingmetal : add set_rows with src0 f16 (llama/25434)
2026-07-10 tyronecaiggml : fix A indexing in simd_gemm scalar tail-column...
2026-07-10 fairydreamingggml : add support for CPU f16->f16 GGML_OP_SET_ROWS...
2026-07-10 lhezopencl: fix potential crash in aos reconstruct (llama...
2026-07-10 Pasha KhosraviAdd Q2_0 quantization: type definition and CPU backend...
2026-07-10 Oliver SimonsCUDA: Fuse MMVQ post-scale for NVFP4 (llama/24481)
2026-07-10 asf0ggml-hip : add -fno-finite-math-only alongside -ffast...
2026-07-10 Neo Zhangfix unsupported UT cases of CONT & CPY (llama/25231)
2026-07-10 Neo Zhangsupport op col2im_1d (llama/25264)
2026-07-10 Neo Zhangsupport OP cross_entropy_loss, cross_entropy_loss_back...
2026-07-10 Todd Malsbarysycl : set K_QUANTS_PER_ITERATION to 1 on DMMV path...
2026-07-10 Neo Zhangfix unsupport ACC UT cases for noncontiguous (llama...
2026-07-10 Neo Zhangsycl : enhance argsort to support all UT cases (llama...
2026-07-10 Neo Zhangsycl : use sycl func to fix AOT double type issue ...
2026-07-10 Neo Zhangsycl : rename the env vars from "disable" to "enable...
2026-07-10 An Longggml : make ggml_time_init idempotent (llama/24422)
2026-07-10 fairydreamingvulkan : check src0 type in GGML_OP_SET_ROWS to avoid...
2026-07-10 Hongqiang Wangopencl: general flash attention decode performance...
2026-07-10 Pascalmetal: add col2im_1d op (f32/f16/bf16) (llama/25176)
2026-07-10 Johannes GäßlerCUDA: remove -sm row, refactor cuBLAS (llama/24216)
2026-07-10 Alexey KopytkoCUDA: extend K-type validation to V-types for flash...
2026-07-10 ragz4125ggml-cpu: use UE4M3 LUT in ARM NVFP4 dot product (llama...
2026-07-10 shalinib-ibmggml-cpu: Enable tiled matmul on AIX (llama/25199)
2026-07-10 hokanosekaivulkan: fix 32-bit integer overflow in CEIL_DIV (llama...
2026-07-10 a-hukggml-hip: enable -ffast-math for HIP builds (llama...
2026-07-10 adavyasggml-cuda: optimize conv_transpose_1d indexing (llama...
2026-07-10 Eveabort if we see a multi buffer (llama/25276)
2026-07-10 liminfei-amdggml : fix tensor-parallel + -ncmoe crash on MoE models...
2026-07-10 Vexxieggml: Update VMM Pool allocation ggml-cuda.cu - Turing...
2026-07-10 fairydreamingcuda : concat implementation for quantized types (llama...
2026-07-10 fairydreamingggml : fix broken CPU concat implementation for quantiz...
2026-07-10 Piotr Wilkin... cuda: enable topk-moe fusion for 288 experts (llama...
2026-07-10 Gaurav GargRemove redundant CUDA copies after gated_delta_net...
2026-07-10 lhezopencl: allow loading precompiled binary kernels from...
2026-07-10 Max Krasnyanskyhexagon: flash attention rework (optimizations, accurac...
2026-07-10 Johannes GäßlerCUDA: consistent use of __restrict__ + PDL for FA ...
2026-07-10 ragz4125ggml-cpu: add AVX2 optimization for nvfp4 dot product...
2026-07-10 lhezopencl: initial q1_0 support (llama/25160)
2026-07-10 fairydreamingcuda : prevent integer truncation and overflow errors...
2026-07-10 Matt JalloCUDA: fix get_rows_back for tables with more than 65535...
2026-07-10 Johannes GäßlerCUDA: fix Gemma E4B MTP FlashAttention (llama/25148)
2026-07-10 Kevin Liuvulkan: roll bk loop in matmul for asahi linux (llama...
2026-07-10 zdufordHIP: use hipBLAS for dense prefill on gfx900, keep...
2026-07-10 Masashi Yoshimuraggml-webgpu: add support for NVFP4 (llama/25143)
2026-07-10 Oliver SimonsRevert "sched : reintroduce less synchronizations durin...
2026-07-10 Ruben Ortlamvulkan: use flops instead of weight tensor size for...
2026-07-10 Hongqiang Wangopencl: flash attention improvement (llama/25069)
2026-07-10 Gaurav GargAdded a cudaMemcpy2DAsync fast path to ggml_cuda_cpy...
2026-07-10 Neo Zhangsycl : fix failed ut cases of norm (llama/25044)
2026-07-10 Ruben Ortlamvulkan: fix step operator for 0 input (llama/25036)
2026-07-10 Andreas Kieslingersched : reintroduce less synchronizations during split...
2026-07-10 Ravi Panchumarthyopenvino: Update to OV 2026.2.1, self-contained release...
2026-07-10 nullnamevulkan: opt mul_mat_vecq for mi50 (llama/22933)
2026-07-10 Jiang, Fishvulkan: add INTEL_XE1 arch enum and enable coopmat1...
2026-07-01 Lin Xiaodongwhisper : expose internal VAD speech segments (#3916)
2026-07-01 Daniel Beveniusci, examples : add package-lock.json to examples addon...
2026-07-01 Lin Xiaodongwhisper : map token timestamps to original time when...
2026-07-01 Lin Xiaodongwhisper : make voice_length() utf-8 aware for CJK ...
2026-06-30 Daniel Beveniusci : set GGML_NATIVE=OFF for sanitize builds (#3920)
2026-06-30 Yongmin Yoo 유용민examples : Improved inference performance of Android...
2026-06-26 Georgi Gerganovtalk-llama : sync llama.cpp
2026-06-26 Georgi Gerganovsync : ggml
2026-06-26 Georgi Gerganovggml : bump version to 0.15.3 (ggml/1550)
2026-06-26 Jeff Bolzvulkan: Workaround compiler bug in conv2d coopmat2...
2026-06-26 leonardHONGCUDA: add cublasSgemmBatched mapping for HIP/MUSA vendo...
2026-06-26 Tarek Dakhranggml-cpu: fix SVE leftover path in ggml_vec_dot_f32...
2026-06-26 Jassieluosycl : clamp softmax input to avoid underflow (llama...
2026-06-26 leonardHONGCUDA: batch out_prod broadcast (dps2>1) path with cubla...
next