Issues

• Enable OpenSSL in FP32/FP16 SYCL builds to restore HF HTTPS downloads

  documentationexamplesSYCL
  Unread
  ggml-org/llama.cpp#30330·raffaeleedidonna opened Oct 11, 2026 · Updated Oct 11, 2026

• examples : fix draft sampling and rejection handling

  examples
  Unread
  ggml-org/llama.cpp#30095·danbev opened Oct 7, 2026 · Updated Oct 9, 2026
  2

• ggml-rpc: validate PAD_REFLECT_1D parameters

  examplesggmltesting
  Unread
  ggml-org/llama.cpp#29915·ZZ2266 opened Oct 3, 2026 · Updated Oct 12, 2026
  1

• ggml: add GGML_TYPE_BF16X - near-lossless bfloat16 recompression (11.5 bpw)

  Apple MetalconversionCUDAexamplesggmltesting
  Unread
  ggml-org/llama.cpp#29896·dfytensor opened Oct 3, 2026 · Updated Oct 3, 2026·Draft
  2

• fix: calc_diff assumes that the last layer will be shrunk

  examples
  Unread
  ggml-org/llama.cpp#29775·cetio opened Sep 30, 2026 · Updated Sep 30, 2026

• metal : per-device tuned mul_mm tiles for every src0 type

  Apple Metaldocumentationexamplesggmltesting
  Unread
  ggml-org/llama.cpp#29743·forforever73 opened Sep 30, 2026 · Updated Oct 3, 2026·Draft
  1

  

• cuda: add PTQ1_0 support

  Apple MetalconversionCUDAexamplesggmlSYCLtesting
  Unread
  ggml-org/llama.cpp#29676·bri-prism opened Sep 29, 2026 · Updated Oct 4, 2026·Draft

• ggml: add PTQ1_0, ternary at group 128

  Apple MetalconversionexamplesggmlSYCLtesting
  Unread
  ggml-org/llama.cpp#29672·bri-prism opened Sep 29, 2026 · Updated Oct 5, 2026
  7

• tools/quantize: add --verify for per-tensor quantization error

  examples
  Unread
  ggml-org/llama.cpp#29430·ceedot-rock opened Sep 25, 2026 · Updated Oct 4, 2026
  2

• laya: Add support for the laya multilingual decision model.

  conversiondocumentationexamplestesting
  Unread
  ggml-org/llama.cpp#29363·Clauszy opened Sep 24, 2026 · Updated Sep 28, 2026
  3

• system-one : typed decision readout for models that answer instead of writing

  conversiondocumentationexamples
  Unread
  ggml-org/llama.cpp#29321·hans00 opened Sep 23, 2026 · Updated Sep 27, 2026·Draft
  2

• llama-bench: fix default thread count on NUMA systems

  examples
  Unread
  ggml-org/llama.cpp#29301·kg290 opened Sep 23, 2026 · Updated Sep 23, 2026
  4

• RPC: multiple graphs cache

  documentationexamplesggmltesting
  Unread
  ggml-org/llama.cpp#29287·Alexander2995 opened Sep 22, 2026 · Updated Sep 26, 2026·Draft
  1

• feat: add a mmap-on-demand that uses pure mmap without prefetch

  examples
  Unread
  ggml-org/llama.cpp#29195·joeldushouyu opened Sep 21, 2026 · Updated Sep 21, 2026

• batched-bench : honor memory fitting options

  examplestesting
  Unread
  ggml-org/llama.cpp#28979·artemg opened Sep 16, 2026 · Updated Sep 16, 2026
  1

• batched-bench: apply CPU affinity options through common threadpools

  examples
  Unread
  ggml-org/llama.cpp#28792·artemg opened Sep 12, 2026 · Updated Sep 12, 2026

• gguf-split: validate --split-max-tensors and --split-max-size

  examples
  Unread
  ggml-org/llama.cpp#28731·linsen458-spec opened Sep 11, 2026 · Updated Sep 24, 2026
  1

• metal: fix tuning include path in subprojects

  examples
  Unread
  ggml-org/llama.cpp#28644·obrr-hhx opened Sep 9, 2026 · Updated Sep 9, 2026·Draft
  1

• llama-bench : add --override-kv

  documentationexamples
  Unread
  ggml-org/llama.cpp#28591·kh0pper opened Sep 8, 2026 · Updated Sep 8, 2026·Draft

• examples : add moe-trace to log MoE expert routing decisions

  documentationexamples
  Unread
  ggml-org/llama.cpp#28544·aryan0078 opened Sep 7, 2026 · Updated Sep 7, 2026
  1

• src : suppress warnings on Windows on ARM (using clang)

  buildexamplesggmlvendor
  Unread
  ggml-org/llama.cpp#28533·danbev opened Sep 7, 2026 · Updated Sep 7, 2026·Draft

• quantize: add --no-fallback option

  documentationexamples
  Unread
  ggml-org/llama.cpp#28474·qdubious opened Sep 6, 2026 · Updated Sep 27, 2026
  5

• ggml : add BPOSIT8 tensor type (b-posit8 W8A8 with exact 256-bit quire accumulation)

  Apple MetalconversionCUDAexamplesggmltesting
  Unread
  ggml-org/llama.cpp#28468·rybruscoe opened Sep 6, 2026 · Updated Sep 11, 2026·Draft
  2

• metal : wide query tile for kernel_flash_attn_ext

  Apple Metaldocumentationexamplesggmltesting
  Unread
  ggml-org/llama.cpp#28439·masterFoad opened Sep 5, 2026 · Updated Sep 26, 2026
  24

  

• cli: allow --server-base to be set via LLAMA_ARG_SERVER_BASE

  documentationexamples
  Unread
  ggml-org/llama.cpp#28428·agustinmista opened Sep 5, 2026 · Updated Sep 7, 2026
  1