llama : bump max seq limit from 64 to 256 (#15916)

ggml-ci

diff --git a/src/llama-cparams.h b/src/llama-cparams.h
index dbbaba9f6274cb0a5a96113e1f42396344c897e4..eae7b839f4857da2df56be6cd8d4d9a5fe362b7a 100644 (file)
--- a/src/llama-cparams.h
+++ b/src/llama-cparams.h
@@ -4,7 +4,7 @@
 
 #include <cstdint>
 
-#define LLAMA_MAX_SEQ 64
+#define LLAMA_MAX_SEQ 256
 
 struct llama_cparams {
     uint32_t n_ctx;           // context size used during inference