From d65a8361fed8be97389776fb94217e937ec6b03c Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Wed, 17 Jul 2024 10:32:59 +0300
Subject: [PATCH] llama : disable context-shift for DeepSeek v2 (#8501)

---
 src/llama.cpp | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/src/llama.cpp b/src/llama.cpp
index 4a9903cc3..89b5763dc 100644
--- a/src/llama.cpp
+++ b/src/llama.cpp
@@ -15133,6 +15133,10 @@ static void llama_kv_cache_update_internal(struct llama_context & lctx) {
 
     // apply K-shift if needed
     if (lctx.model.hparams.rope_type != LLAMA_ROPE_TYPE_NONE && lctx.kv_self.has_shift) {
+        if (lctx.model.arch == LLM_ARCH_DEEPSEEK2) { // not supported due to MLA
+            GGML_ASSERT(false && "Deepseek2 does not support K-shift");
+        }
+
         {
             ggml_backend_sched_reset(lctx.sched);