ncnn/tools/quantize at master - ncnn

mirror of https://github.com/Tencent/ncnn.git synced 2026-10-16 20:32:33 +08:00

Files

T

History

風的彷徨 ab3790faca fix MHA singleton mask broadcast across backends (#6869 )

Keep persistent KV caches on the blob allocator, broadcast single-channel attention masks across heads on CPU and Vulkan paths, and add long multi-token cache regressions.

2026-10-10 19:47:13 +08:00

CMakeLists.txt

gemm block quantization infrastructure for llm decoder style (#6814 )

2026-07-13 10:13:39 +08:00

imreadwrite.cpp

use spdx license header, copyright Tencent (#6152 )

2025-07-08 11:01:29 +08:00

imreadwrite.h

use spdx license header, copyright Tencent (#6152 )

2025-07-08 11:01:29 +08:00

ncnn2int8.cpp

Add percentile PTQ method in ncnn2table (#6854 )

2026-10-09 10:33:34 +08:00

ncnn2table.cpp

Add percentile PTQ method in ncnn2table (#6854 )

2026-10-09 10:33:34 +08:00

ncnnllm2int.cpp

gemm block quantization infrastructure for llm decoder style (#6814 )

2026-07-13 10:13:39 +08:00

ncnnllm2table.cpp

fix MHA singleton mask broadcast across backends (#6869 )

2026-10-10 19:47:13 +08:00

ncnnllm_quant.h

gemm block quantization infrastructure for llm decoder style (#6814 )

2026-07-13 10:13:39 +08:00

npy.hpp

add read npy for ncnn2table (#5930 )

2025-03-18 16:50:36 +08:00

README.md

update readme of quantize (#3283 )

2021-10-15 16:42:06 +08:00

README.md

see quantized-int8-inference