localai-backend/acestep-cpp
LocalAI music-generation backend (acestep.cpp gRPC server)
Screenshots
-
acestep-cpp-4.12.1~amd64+amdgpu_targets_gfx908 +amdgpu_targets_gfx90a +amdgpu_targets_gfx942 +amdgpu_targets_gfx950 +amdgpu_targets_gfx1030 +amdgpu_targets_gfx1100 +amdgpu_targets_gfx1101 +amdgpu_targets_gfx1102 +amdgpu_targets_gfx1103 +amdgpu_targets_gfx1150 +amdgpu_targets_gfx1151 +amdgpu_targets_gfx1152 +amdgpu_targets_gfx1153 +amdgpu_targets_gfx1200 +amdgpu_targets_gfx1201 amdgpu_targets_gfx803 amdgpu_targets_gfx900 amdgpu_targets_gfx906 amdgpu_targets_gfx90c amdgpu_targets_gfx1010 amdgpu_targets_gfx1011 amdgpu_targets_gfx1012 amdgpu_targets_gfx1031 amdgpu_targets_gfx1032 amdgpu_targets_gfx1033 amdgpu_targets_gfx1034 amdgpu_targets_gfx1035 amdgpu_targets_gfx1036 native openblas +openmp vulkan video_cards_amdgpu cuda rocm cpu_flags_x86_amx_bf16 cpu_flags_x86_amx_int8 cpu_flags_x86_amx_tile cpu_flags_x86_avx cpu_flags_x86_avx_vnni cpu_flags_x86_avx2 cpu_flags_x86_avx512_bf16 cpu_flags_x86_avx512bw cpu_flags_x86_avx512f cpu_flags_x86_avx512vbmi cpu_flags_x86_avx512_vnni cpu_flags_x86_bmi2 cpu_flags_x86_fma3 cpu_flags_x86_f16c cpu_flags_x86_sse4_2
View 
Download 
Browse License: MIT Overlay: local-ai
ChangeLog
commit 91db65b3d533387a49330d6560225412c42aea71
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Fri Oct 23 19:19:50 2026 +0000
localai-backend/acestep-cpp: patch renamed for 4.12.1
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit 8de6f6d5531dd0d87376d4f60dd9af63b03122ce
Author: gitea-actions <gitea-actions@noreply.ipnmod.org>
Date: Fri Oct 23 19:05:58 2026 +0000
Manifests for localai-v4.12.1
commit e1f1377fa4b9958fa6acafe2bf1e050b4bd89748
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Fri Oct 23 18:58:51 2026 +0000
LocalAI 4.12.1
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit 9198bd4b7dc3715c908d5fd083c7ba216abd8671
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Fri Oct 23 17:24:05 2026 +0000
LocalAI 4.12.0: server and C++/Go backends
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit 19aec51c84f086e4f5caaf899a6bc2d468edb671
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Thu Oct 22 15:03:41 2026 +0000
local-ai-ggml.eclass: build USE=rocm with ROCm clang, not hipcc
With hipcc as CXX, ggml takes its legacy whole-project hipcc path and
CMake cannot resolve the OpenMP runtime for the final link: on ROCm
10.1 (clang 23) ggml-cpu's omp_* symbols stay unresolved, fatally for
grpc-server and silently for the shared engine libraries. Plain clang
sends the .cu files through CMake's HIP language, so CXXFLAGS are
forwarded via CMAKE_HIP_FLAGS. Same move as sci-misc/llama-cpp::guru.
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit f91427abfd07674afc46e4b81bfcf99176c43364
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Wed Oct 21 19:46:26 2026 +0000
Manifests for the LLVM 23 rocm revbumps
EBUILD entries follow the 17 renames from ea36568; no distfile changed.
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit ea365680b3f989d34a9d3f32199bcb5c668c4a83
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Wed Oct 21 19:41:36 2026 +0000
local-ai-ggml.eclass: rewrite __bf16 WMMA vectors for LLVM 23, revbump rocm consumers
ROCm 10.1 compiles HIP with stock clang 23, whose AMDGPU bf16 builtins
reject the __bf16 vectors ggml's mma.cuh declares (llvm-project PR
176033). Apply the same element-type rewrite as llama.cpp PR 29536 and
sci-ml/ggml's llvm-23 patch from src_prepare under USE=rocm, so every
vendored ggml copy gets it regardless of line offsets.
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit 431a8b440e3cf7e5da3e16140cbec3be336aa77c
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Sat Oct 17 08:49:45 2026 +0000
localai-backend/acestep-cpp, vllm-cpp: -r1 for the GCC 16 / CUDA 13 fixes
The fixes changed the built sources in place; a revision makes every
install, binhost included, pick them up.
commit c6f2a3819c3bd494fe7c4bd55e37c5ac57b024af
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Sat Oct 17 07:17:15 2026 +0000
localai-backend/acestep-cpp: build with CUDA 13's CCCL
CCCL 3.1 stopped providing <cuda/iterator> through <cub/cub.cuh>; backport
the two llama.cpp one-liners, already on the ggml fork's master.
commit 3bb869d602ba74cce86a0fef3eb89fa89c336fa8
Author: gitea-actions <gitea-actions@noreply.ipnmod.org>
Date: Sun Oct 4 08:31:58 2026 +0000
Manifests for localai-v4.11.0
commit c10ccda015a84b3237ba732353f1aa4c6acd72d0
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Sun Oct 4 07:37:06 2026 +0000
reorganize: www-apps/localai-server, sci-ml/localai meta, localai-backend/*
The world atom becomes the meta via chained pkgmoves; see the news item.
Assisted-by: Claude:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Fri Oct 23 19:19:50 2026 +0000
localai-backend/acestep-cpp: patch renamed for 4.12.1
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit 8de6f6d5531dd0d87376d4f60dd9af63b03122ce
Author: gitea-actions <gitea-actions@noreply.ipnmod.org>
Date: Fri Oct 23 19:05:58 2026 +0000
Manifests for localai-v4.12.1
commit e1f1377fa4b9958fa6acafe2bf1e050b4bd89748
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Fri Oct 23 18:58:51 2026 +0000
LocalAI 4.12.1
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit 9198bd4b7dc3715c908d5fd083c7ba216abd8671
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Fri Oct 23 17:24:05 2026 +0000
LocalAI 4.12.0: server and C++/Go backends
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit 19aec51c84f086e4f5caaf899a6bc2d468edb671
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Thu Oct 22 15:03:41 2026 +0000
local-ai-ggml.eclass: build USE=rocm with ROCm clang, not hipcc
With hipcc as CXX, ggml takes its legacy whole-project hipcc path and
CMake cannot resolve the OpenMP runtime for the final link: on ROCm
10.1 (clang 23) ggml-cpu's omp_* symbols stay unresolved, fatally for
grpc-server and silently for the shared engine libraries. Plain clang
sends the .cu files through CMake's HIP language, so CXXFLAGS are
forwarded via CMAKE_HIP_FLAGS. Same move as sci-misc/llama-cpp::guru.
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit f91427abfd07674afc46e4b81bfcf99176c43364
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Wed Oct 21 19:46:26 2026 +0000
Manifests for the LLVM 23 rocm revbumps
EBUILD entries follow the 17 renames from ea36568; no distfile changed.
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit ea365680b3f989d34a9d3f32199bcb5c668c4a83
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Wed Oct 21 19:41:36 2026 +0000
local-ai-ggml.eclass: rewrite __bf16 WMMA vectors for LLVM 23, revbump rocm consumers
ROCm 10.1 compiles HIP with stock clang 23, whose AMDGPU bf16 builtins
reject the __bf16 vectors ggml's mma.cuh declares (llvm-project PR
176033). Apply the same element-type rewrite as llama.cpp PR 29536 and
sci-ml/ggml's llvm-23 patch from src_prepare under USE=rocm, so every
vendored ggml copy gets it regardless of line offsets.
Assisted-by: Claude Code:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>
commit 431a8b440e3cf7e5da3e16140cbec3be336aa77c
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Sat Oct 17 08:49:45 2026 +0000
localai-backend/acestep-cpp, vllm-cpp: -r1 for the GCC 16 / CUDA 13 fixes
The fixes changed the built sources in place; a revision makes every
install, binhost included, pick them up.
commit c6f2a3819c3bd494fe7c4bd55e37c5ac57b024af
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Sat Oct 17 07:17:15 2026 +0000
localai-backend/acestep-cpp: build with CUDA 13's CCCL
CCCL 3.1 stopped providing <cuda/iterator> through <cub/cub.cuh>; backport
the two llama.cpp one-liners, already on the ggml fork's master.
commit 3bb869d602ba74cce86a0fef3eb89fa89c336fa8
Author: gitea-actions <gitea-actions@noreply.ipnmod.org>
Date: Sun Oct 4 08:31:58 2026 +0000
Manifests for localai-v4.11.0
commit c10ccda015a84b3237ba732353f1aa4c6acd72d0
Author: Plamen K. Kosseff <plamen@ipnmod.org>
Date: Sun Oct 4 07:37:06 2026 +0000
reorganize: www-apps/localai-server, sci-ml/localai meta, localai-backend/*
The world atom becomes the meta via chained pkgmoves; see the news item.
Assisted-by: Claude:claude-fable-5
Signed-off-by: Plamen K. Kosseff <plamen@ipnmod.org>