Skip to content

Improvements to inference using int8 compressed kv's #871

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
copybara-service wants to merge 1 commit into
base: dev
Choose a base branch
from
+667 −17
Open

Improvements to inference using int8 compressed kv's #871

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 13 additions & 1 deletion compression/compress-inl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -511,6 +511,17 @@ struct CompressTraits<int8_t> {
raw1 = hn::ConvertTo(df, vec_i32_1);
}

template <class DI16, HWY_IF_I16_D(DI16)>
static HWY_INLINE void Load2(DI16 di16,
const PackedSpan<const Packed>& packed,
const size_t packed_ofs, hn::Vec<DI16>& raw0,
hn::Vec<DI16>& raw1) {
const hn::Repartition<int8_t, DI16> di8;
const auto packed_vec = hn::LoadU(di8, packed.ptr + packed_ofs);
raw0 = hn::PromoteLowerTo(di16, packed_vec);
raw1 = hn::PromoteUpperTo(di16, packed_vec);
}

template <class DBF, HWY_IF_BF16_D(DBF)>
static HWY_INLINE void Load2(DBF dbf, const PackedSpan<const Packed>& packed,
const size_t packed_ofs, hn::Vec<DBF>& raw0,
Expand Down Expand Up @@ -745,7 +756,8 @@ HWY_INLINE void VerifyRawAndPackedForDecompress() {
using TRaw = hn::TFromD<DRaw>;
// We can decompress any Packed to f32 or BF16, or f32 to f64.
static_assert(hwy::IsSameEither<TRaw, float, BF16>() ||
(IsF32<Packed>() && hwy::IsSame<TRaw, double>()));
(IsF32<Packed>() && hwy::IsSame<TRaw, double>()) ||
(IsInt8<Packed>() && hwy::IsSame<TRaw, int16_t>()));
}

} // namespace detail
Expand Down
Loading
Loading