llama: update to commit 71e90e88 (#10192)

llama/llama.cpp/src/llama-chat.h

@@ -38,6 +38,9 @@ enum llm_chat_template {
     LLM_CHAT_TEMPLATE_GRANITE,
     LLM_CHAT_TEMPLATE_GIGACHAT,
     LLM_CHAT_TEMPLATE_MEGREZ,
+    LLM_CHAT_TEMPLATE_YANDEX,
+    LLM_CHAT_TEMPLATE_BAILING,
+    LLM_CHAT_TEMPLATE_LLAMA4,
     LLM_CHAT_TEMPLATE_UNKNOWN,
 };
 

llama/llama.cpp/src/llama-cparams.h

@@ -30,6 +30,7 @@ struct llama_cparams {
     bool flash_attn;
     bool no_perf;
     bool cross_attn;
+    bool warmup;
 
     enum llama_pooling_type pooling_type;
 

llama/llama.cpp/src/llama-hparams.cpp

@@ -2,8 +2,6 @@
 
 #include "ggml.h"
 
-#include <algorithm>
-
 uint32_t llama_hparams::n_head(uint32_t il) const {
     if (il < n_layer) {
         return n_head_arr[il];
@@ -80,6 +78,14 @@ bool llama_hparams::n_bskcn(uint32_t n, uint32_t il) const {
     GGML_ABORT("fatal error");
 }
 
+bool llama_hparams::is_swa(uint32_t il) const {
+    if (il < n_layer) {
+        return n_swa > 0 && n_swa_pattern > 0 && il % n_swa_pattern < (n_swa_pattern - 1);
+    }
+
+    GGML_ABORT("fatal error");
+}
+
 bool llama_hparams::cross_attention_layers(uint32_t il) const {
     return std::find(cross_attn_layers.begin(), cross_attn_layers.end(), il) != cross_attn_layers.end();
 }

llama/llama.cpp/src/llama-memory.cpp

+#include "llama-memory.h"