Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
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
19 changes: 19 additions & 0 deletions gemma/configs.cc
Original file line number Diff line number Diff line change
Expand Up @@ -736,4 +736,23 @@ AttentionImpl GetAttentionImpl(const std::string& impl_name) {
return AttentionImpl::kFlash;
}

std::string KVEncodingToString(KVEncoding encoding) {
switch (encoding) {
case KVEncoding::kF32:
return "F32";
case KVEncoding::kBF16:
return "BF16";
case KVEncoding::kF32TwoTranspositions:
return "F32TwoTranspositions";
case KVEncoding::kBF16TwoTranspositions:
return "BF16TwoTranspositions";
case KVEncoding::kInt8:
return "Int8";
case KVEncoding::kInt8TwoTranspositions:
return "Int8TwoTranspositions";
default:
return "Unknown";
}
}

} // namespace gcpp
7 changes: 7 additions & 0 deletions gemma/configs.h
Original file line number Diff line number Diff line change
Expand Up @@ -92,6 +92,13 @@ enum class KVEncoding {
kInt8TwoTranspositions = 6,
};

// Returns a string representation of the KVEncoding.
// This representation is should not change and can be used for
// serialization or logging.
// Note that no reverse function exists to convert a string back to a
// KVEncoding.
std::string KVEncodingToString(KVEncoding encoding);

enum class AttentionImpl {
kFlash = 0, // Flash Attention (default)
kFlashTransposedQs,
Expand Down
Loading