mirror of https://github.com/google/gemma.cpp.git
parent
af8eb2fde3
commit
e527e7662e
|
|
@ -35,7 +35,6 @@ constexpr size_t kDecodeBatchSize = 1;
|
||||||
constexpr size_t kBatchedQueryBatchSize = 16;
|
constexpr size_t kBatchedQueryBatchSize = 16;
|
||||||
constexpr size_t kMinAdjustedPrefillBatchSize =
|
constexpr size_t kMinAdjustedPrefillBatchSize =
|
||||||
HWY_MAX((size_t)1, kPrefillBatchSize / kBatchedQueryBatchSize);
|
HWY_MAX((size_t)1, kPrefillBatchSize / kBatchedQueryBatchSize);
|
||||||
constexpr bool kSystemPrompt = false;
|
|
||||||
|
|
||||||
struct KVCache {
|
struct KVCache {
|
||||||
hwy::AlignedFreeUniquePtr<float[]>
|
hwy::AlignedFreeUniquePtr<float[]>
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue