]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
metal: implement DeepSeek V4 hyper-connections (#26459)
authorGeorgi Gerganov <redacted>
Sun, 2 Aug 2026 18:06:02 +0000 (21:06 +0300)
committerGitHub <redacted>
Sun, 2 Aug 2026 18:06:02 +0000 (21:06 +0300)
commitfffbcbdb9d5e56105a8842867a59bb9736520ca8
treeb3af22e48becabb4fb3c00c393f331d93290a196
parentbb4e0e1b3f6bb38960769a1c9bcd2081016154cd
metal: implement DeepSeek V4 hyper-connections (#26459)

- Implement GGML_OP_DSV4_HC_COMB, GGML_OP_DSV4_HC_PRE, and
  GGML_OP_DSV4_HC_POST with SIMDgroup register and shuffle optimized kernels.
- Add Metal dispatch and support plumbing and test the production Sinkhorn
  iteration count and embedding width.

Assisted-by: Codex
Co-authored-by: Thiago Padilha <redacted>
ggml/src/ggml-metal/ggml-metal-device.cpp
ggml/src/ggml-metal/ggml-metal-device.h
ggml/src/ggml-metal/ggml-metal-device.m
ggml/src/ggml-metal/ggml-metal-impl.h
ggml/src/ggml-metal/ggml-metal-ops.cpp
ggml/src/ggml-metal/ggml-metal-ops.h
ggml/src/ggml-metal/ggml-metal.metal
tests/test-backend-ops.cpp