From 872c365a9176a011b13d31269bb3121fa89c37e1 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Sat, 22 Apr 2023 11:08:12 +0300 Subject: ggml : fix AVX build + update to new Q8_0 format --- llama.cpp | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'llama.cpp') diff --git a/llama.cpp b/llama.cpp index 00cce6e2..4e92f551 100644 --- a/llama.cpp +++ b/llama.cpp @@ -68,7 +68,7 @@ static const std::map & MEM_REQ_SCRATCH1() { MODEL_65B, 512ull * MB }, }; return _MEM_REQ_SCRATCH1; -}; +} // 2*n_embd*n_ctx*n_layer*sizeof(float16) static const std::map & MEM_REQ_KV_SELF() @@ -80,7 +80,7 @@ static const std::map & MEM_REQ_KV_SELF() { MODEL_65B, 5120ull * MB }, }; return _MEM_REQ_KV_SELF; -}; +} // this is mostly needed for temporary mul_mat buffers to dequantize the data // not actually needed if BLAS is disabled @@ -93,7 +93,7 @@ static const std::map & MEM_REQ_EVAL() { MODEL_65B, 1536ull * MB }, }; return _MEM_REQ_EVAL; -}; +} // default hparams (LLaMA 7B) struct llama_hparams { -- cgit v1.2.3