From d7787c8f6cfe58a4f1171c664f4ba1ae02cb663d Mon Sep 17 00:00:00 2001 From: RangerUFO Date: Fri, 21 Jun 2024 13:06:26 +0800 Subject: [PATCH] Fix KV cache size calculation error --- gemma/gemma.cc | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/gemma/gemma.cc b/gemma/gemma.cc index f32cde5..1c07c9e 100644 --- a/gemma/gemma.cc +++ b/gemma/gemma.cc @@ -114,7 +114,7 @@ struct CreateKVCache { KVCache operator()() const { KVCache kv_cache = {}; - const size_t size_cache_pos = CacheLayerSize()(); + const size_t size_cache_pos = CachePosSize()(); if (size_cache_pos != 0) { const size_t seq_len = TConfig::kSeqLen + kPrefillBatchSize; kv_cache.kv_cache = hwy::AllocateAligned(seq_len * size_cache_pos);