From: Georgi Gerganov <redacted>
Date: Thu, 17 Jul 2025 17:52:33 +0000 (+0300)
Subject: kv-cache : fix k-shift for multiple streams (#14742)
X-Git-Tag: upstream/0.0.6073~150
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=d6fb3f6b49b27ef1c0f4cf5128e041f7e7dc03af;p=pkg%2Fggml%2Fsources%2Fllama.cpp

kv-cache : fix k-shift for multiple streams (#14742)

ggml-ci
---

diff --git a/src/llama-kv-cache-unified.cpp b/src/llama-kv-cache-unified.cpp
index 98c01ea7..321dc79f 100644
--- a/src/llama-kv-cache-unified.cpp
+++ b/src/llama-kv-cache-unified.cpp
@@ -1261,7 +1261,7 @@ void llama_kv_cache_unified::set_input_k_shift(ggml_tensor * dst) const {
         const auto & cells = v_cells[s];
 
         for (uint32_t i = 0; i < cells.size(); ++i) {
-            data[i] = cells.is_empty(i) ? 0 : cells.get_shift(i);
+            data[s*cells.size() + i] = cells.is_empty(i) ? 0 : cells.get_shift(i);
         }
     }
 }