From c879133a5a4ddba2815bb738172950e43b7d848c Mon Sep 17 00:00:00 2001 From: "The gemma.cpp Authors" Date: Fri, 12 Jul 2024 06:27:57 -0700 Subject: [PATCH] Increase the prefill batch size to 64. PiperOrigin-RevId: 651754772 --- gemma/common.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/gemma/common.h b/gemma/common.h index 70c950c..0ca7e5c 100644 --- a/gemma/common.h +++ b/gemma/common.h @@ -36,7 +36,7 @@ ByteStorageT AllocateSizeof() { return hwy::AllocateAligned(sizeof(T)); } -constexpr size_t kPrefillBatchSize = 16; +constexpr size_t kPrefillBatchSize = 64; constexpr size_t kDecodeBatchSize = 1; constexpr size_t kBatchedQueryBatchSize = 16; constexpr size_t kMinAdjustedPrefillBatchSize =