happyz
/
gemma.cpp
mirror of https://github.com/google/gemma.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fix int8 

PiperOrigin-RevId: 882611833
This commit is contained in:
Krzysztof Rymski 2026-03-12 08:42:56 -07:00 committed by Copybara-Service
parent d6e836c651
commit 197c1a049c
2 changed files with 3 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
compression/compress-inl.h
View File

@ -472,7 +472,7 @@ struct CompressTraits<int8_t> {
const auto vi32_1 = hn::NearestInt(v1);
const auto vi16 = hn::OrderedDemote2To(di16, vi32_0, vi32_1);
const auto vi8 = hn::OrderedDemote2To(
di8_16, hn::UpperHalf(di16_16, vi16), hn::LowerHalf(di16_16, vi16));
di8_16, hn::LowerHalf(di16_16, vi16), hn::UpperHalf(di16_16, vi16));
hn::StoreU(vi8, di8_16, packed.ptr + packed_ofs + i);
}
}
@ -487,7 +487,7 @@ struct CompressTraits<int8_t> {
const auto vi32_1 = hn::NearestInt(v1);
const auto vi16 = hn::OrderedDemote2To(di16, vi32_0, vi32_1);
const auto vi8 = hn::OrderedDemote2To(
di8_16, hn::UpperHalf(di16_16, vi16), hn::LowerHalf(di16_16, vi16));
di8_16, hn::LowerHalf(di16_16, vi16), hn::UpperHalf(di16_16, vi16));
hn::StoreN(vi8, di8_16, packed.ptr + packed_ofs + i, remaining);
}
}

3
gemma/tiled_attention.cc
View File

@ -249,8 +249,7 @@ static HWY_INLINE void ComputeQKVTransposedTile(
v_cache_values = v_buf;
}
if (attention_impl == AttentionImpl::kFlashTransposedQsBF16 &&
!IsInt8<KV_T>()) {
if (attention_impl == AttentionImpl::kFlashTransposedQsBF16) {
const int in_tile_idx_mod_2 = in_tile_idx % 2;
for (int dim = 0; dim < qkv_dim; dim += 2) {
const int dim_mod_2 = dim % 2;