From c71bfd736ee99a56e697697b39240f2ee06ed26d Mon Sep 17 00:00:00 2001
From: slaren <slarengh@gmail.com>
Date: Thu, 18 Apr 2024 09:04:47 +0200
Subject: [PATCH] llama : fix compatibility with old 2 expert models (#6735)

---
 llama.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/llama.cpp b/llama.cpp
index f4f4063cf..8c1446296 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -4592,7 +4592,7 @@ static bool llm_load_tensors(
     size_t ctx_size = ggml_tensor_overhead()*(ml.n_tensors + 1); // +1 for models where tok_embd is duplicated as output
 
     // for moe merged tensors
-    ctx_size += ggml_tensor_overhead()*hparams.n_expert*n_layer;
+    ctx_size += ggml_tensor_overhead()*n_layer*3;
 
     std::map<ggml_backend_buffer_type_t, ggml_context *> ctx_map;
     for (auto & it : buft_layer_count) {