mirror of
https://github.com/kvcache-ai/ktransformers.git
synced 2026-05-02 21:51:30 +00:00
57d14d22bc
* refactor: move legacy code to archive/ directory - Moved ktransformers, csrc, third_party, merge_tensors to archive/ - Moved build scripts and configurations to archive/ - Kept kt-kernel, KT-SFT, doc, and README files in root - Preserved complete git history for all moved files * refactor: restructure repository to focus on kt-kernel and KT-SFT modules * fix README * fix README * fix README * fix README * docs: add performance benchmarks to kt-kernel section Add comprehensive performance data for kt-kernel to match KT-SFT's presentation: - AMX kernel optimization: 21.3 TFLOPS (3.9× faster than PyTorch) - Prefill phase: up to 20× speedup vs baseline - Decode phase: up to 4× speedup - NUMA optimization: up to 63% throughput improvement - Multi-GPU (8×L20): 227.85 tokens/s total throughput with DeepSeek-R1 FP8 Source: https://lmsys.org/blog/2025-10-22-KTransformers/ This provides users with concrete performance metrics for both core modules, making it easier to understand the capabilities of each component. * refactor: improve kt-kernel performance data with specific hardware and models Replace generic performance descriptions with concrete benchmarks: - Specify exact hardware: 8×L20 GPU + Xeon Gold 6454S, Single/Dual-socket Xeon + AMX - Include specific models: DeepSeek-R1-0528 (FP8), DeepSeek-V3 (671B) - Show detailed metrics: total throughput, output throughput, concurrency details - Match KT-SFT presentation style for consistency This provides users with actionable performance data they can use to evaluate hardware requirements and expected performance for their use cases. * fix README * docs: clean up performance table and improve formatting * add pic for README * refactor: simplify .gitmodules and backup legacy submodules - Remove 7 legacy submodules from root .gitmodules (archive/third_party/*) - Keep only 2 active submodules for kt-kernel (llama.cpp, pybind11) - Backup complete .gitmodules to archive/.gitmodules - Add documentation in archive/README.md for researchers who need legacy submodules This reduces initial clone size by ~500MB and avoids downloading unused dependencies. * refactor: move doc/ back to root directory Keep documentation in root for easier access and maintenance. * refactor: consolidate all images to doc/assets/ - Move kt-kernel/assets/heterogeneous_computing.png to doc/assets/ - Remove KT-SFT/assets/ (images already in doc/assets/) - Update KT-SFT/README.md image references to ../doc/assets/ - Eliminates ~7.9MB image duplication - Centralizes all documentation assets in one location * fix pic path for README
100 lines
2.6 KiB
CMake
100 lines
2.6 KiB
CMake
include(CheckCSourceRuns)
|
|
|
|
set(AVX_CODE "
|
|
#include <immintrin.h>
|
|
int main()
|
|
{
|
|
__m256 a;
|
|
a = _mm256_set1_ps(0);
|
|
return 0;
|
|
}
|
|
")
|
|
|
|
set(AVX512_CODE "
|
|
#include <immintrin.h>
|
|
int main()
|
|
{
|
|
__m512i a = _mm512_set_epi8(0, 0, 0, 0, 0, 0, 0, 0,
|
|
0, 0, 0, 0, 0, 0, 0, 0,
|
|
0, 0, 0, 0, 0, 0, 0, 0,
|
|
0, 0, 0, 0, 0, 0, 0, 0,
|
|
0, 0, 0, 0, 0, 0, 0, 0,
|
|
0, 0, 0, 0, 0, 0, 0, 0,
|
|
0, 0, 0, 0, 0, 0, 0, 0,
|
|
0, 0, 0, 0, 0, 0, 0, 0);
|
|
__m512i b = a;
|
|
__mmask64 equality_mask = _mm512_cmp_epi8_mask(a, b, _MM_CMPINT_EQ);
|
|
return 0;
|
|
}
|
|
")
|
|
|
|
set(AVX2_CODE "
|
|
#include <immintrin.h>
|
|
int main()
|
|
{
|
|
__m256i a = {0};
|
|
a = _mm256_abs_epi16(a);
|
|
__m256i x;
|
|
_mm256_extract_epi64(x, 0); // we rely on this in our AVX2 code
|
|
return 0;
|
|
}
|
|
")
|
|
|
|
set(FMA_CODE "
|
|
#include <immintrin.h>
|
|
int main()
|
|
{
|
|
__m256 acc = _mm256_setzero_ps();
|
|
const __m256 d = _mm256_setzero_ps();
|
|
const __m256 p = _mm256_setzero_ps();
|
|
acc = _mm256_fmadd_ps( d, p, acc );
|
|
return 0;
|
|
}
|
|
")
|
|
|
|
macro(check_sse type flags)
|
|
set(__FLAG_I 1)
|
|
set(CMAKE_REQUIRED_FLAGS_SAVE ${CMAKE_REQUIRED_FLAGS})
|
|
foreach (__FLAG ${flags})
|
|
if (NOT ${type}_FOUND)
|
|
set(CMAKE_REQUIRED_FLAGS ${__FLAG})
|
|
check_c_source_runs("${${type}_CODE}" HAS_${type}_${__FLAG_I})
|
|
if (HAS_${type}_${__FLAG_I})
|
|
set(${type}_FOUND TRUE CACHE BOOL "${type} support")
|
|
set(${type}_FLAGS "${__FLAG}" CACHE STRING "${type} flags")
|
|
endif()
|
|
math(EXPR __FLAG_I "${__FLAG_I}+1")
|
|
endif()
|
|
endforeach()
|
|
set(CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS_SAVE})
|
|
|
|
if (NOT ${type}_FOUND)
|
|
set(${type}_FOUND FALSE CACHE BOOL "${type} support")
|
|
set(${type}_FLAGS "" CACHE STRING "${type} flags")
|
|
endif()
|
|
|
|
mark_as_advanced(${type}_FOUND ${type}_FLAGS)
|
|
endmacro()
|
|
|
|
# flags are for MSVC only!
|
|
check_sse("AVX" " ;/arch:AVX")
|
|
if (NOT ${AVX_FOUND})
|
|
set(LLAMA_AVX OFF)
|
|
else()
|
|
set(LLAMA_AVX ON)
|
|
endif()
|
|
|
|
check_sse("AVX2" " ;/arch:AVX2")
|
|
check_sse("FMA" " ;/arch:AVX2")
|
|
if ((NOT ${AVX2_FOUND}) OR (NOT ${FMA_FOUND}))
|
|
set(LLAMA_AVX2 OFF)
|
|
else()
|
|
set(LLAMA_AVX2 ON)
|
|
endif()
|
|
|
|
check_sse("AVX512" " ;/arch:AVX512")
|
|
if (NOT ${AVX512_FOUND})
|
|
set(LLAMA_AVX512 OFF)
|
|
else()
|
|
set(LLAMA_AVX512 ON)
|
|
endif()
|