mirror of
https://github.com/Niko1221/Strata.git
synced 2026-11-07 14:21:06 -10:00
Niko1221
ad9eb89079
metal/ is its own CMake project (macOS only), nothing in src/ or include/ changes. MSL is compiled at run time (the Command Line Tools have no offline metal compiler) with ggml-common.h prepended for the i-quant grids. - src/mtl_rt.*: a C++ face over Metal (buffers, pipelines with function constants, serial/concurrent encoders, indirect dispatch, indirect command buffers, GPU timing). - src/ref_quant.*: CPU reference dequantizers for Q2_0, Q4_0, Q5_0, Q8_0, Q3_K..Q6_K, IQ2_XXS/XS/S, IQ3_XXS/S, IQ4_NL/XS, BF16, F16; checked bitwise against gguf-py (tests/ref_dump + tools/check_ref.py: 17/17 equal). - kernels/quant.metal: the formats as per-lane units (decode once, dot with every token of the window). - kernels/moe.metal: moe_plan (top-k -> groups by expert, deterministic, writes the indirect grids), moe_gu (gate+up+SwiGLU, grouped), moe_down, moe_combine; tests/expert_parity: parity vs double and the bench. - tests/rt_bench: M5 Max streaming read ~585 GB/s, ~0.85 us per dependent dispatch, ~170 us commit+wait.
31 lines
1.4 KiB
C++
31 lines
1.4 KiB
C++
// metal/tests/ref_dump.cpp - writes random blocks of every type ref_quant knows and their CPU-reference dequant, so
|
|
// that the reference itself can be checked against gguf-py (metal/tools/check_ref.py):
|
|
// ref_dump <dir> -> <dir>/<TYPE>.bin (raw blocks, 16 rows x 2560) and <dir>/<TYPE>.f32
|
|
#include "ref_quant.h"
|
|
|
|
#include <cstdio>
|
|
#include <string>
|
|
#include <vector>
|
|
|
|
int main(int argc, char** argv) {
|
|
const std::string dir = argc > 1 ? argv[1] : ".";
|
|
const int rows = 16, cols = 2560;
|
|
for (int t : {ref::Q2_0, ref::Q4_0, ref::Q5_0, ref::Q8_0, ref::Q3_K, ref::Q4_K, ref::Q5_K, ref::Q6_K, ref::IQ2_XXS,
|
|
ref::IQ2_XS, ref::IQ2_S, ref::IQ3_XXS, ref::IQ3_S, ref::IQ4_NL, ref::IQ4_XS, ref::BF16, ref::F16}) {
|
|
const size_t rb = ref::row_bytes(t, cols);
|
|
std::vector<uint8_t> raw(rb * rows);
|
|
ref::random_blocks(t, raw.data(), rows, cols, 77 + t);
|
|
std::vector<float> f((size_t)rows * cols);
|
|
ref::dequant(t, raw.data(), f.data(), f.size());
|
|
const std::string n = dir + "/" + ref::type_name(t);
|
|
FILE* a = std::fopen((n + ".bin").c_str(), "wb");
|
|
FILE* b = std::fopen((n + ".f32").c_str(), "wb");
|
|
std::fwrite(raw.data(), 1, raw.size(), a);
|
|
std::fwrite(f.data(), 4, f.size(), b);
|
|
std::fclose(a);
|
|
std::fclose(b);
|
|
}
|
|
std::printf("wrote %s\n", dir.c_str());
|
|
return 0;
|
|
}
|