From 4b781c20558f0f1694b737f1c3117e2779ccec96 Mon Sep 17 00:00:00 2001
From: eiery <19350831+eiery@users.noreply.github.com>
Date: Thu, 20 Apr 2023 17:04:31 -0400
Subject: [PATCH] set default n_batch to 512 when using BLAS

---
 Makefile          | 3 +++
 examples/common.h | 4 ++++
 2 files changed, 7 insertions(+)

diff --git a/Makefile b/Makefile
index f267d0864..f3434fbfe 100644
--- a/Makefile
+++ b/Makefile
@@ -93,15 +93,18 @@ ifndef LLAMA_NO_ACCELERATE
 	# `-framework Accelerate` works on Mac Intel as well, with negliable performance boost (as of the predict time).
 	ifeq ($(UNAME_S),Darwin)
 		CFLAGS  += -DGGML_USE_ACCELERATE
+		CXXFLAGS  += -DGGML_USE_ACCELERATE
 		LDFLAGS += -framework Accelerate
 	endif
 endif
 ifdef LLAMA_OPENBLAS
 	CFLAGS  += -DGGML_USE_OPENBLAS -I/usr/local/include/openblas
+	CXXFLAGS  += -DGGML_USE_OPENBLAS
 	LDFLAGS += -lopenblas
 endif
 ifdef LLAMA_CUBLAS
 	CFLAGS  += -DGGML_USE_CUBLAS -I/usr/local/cuda/include
+	CXXFLAGS  += -DGGML_USE_CUBLAS
 	LDFLAGS += -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64
 	OBJS	+= ggml-cuda.o
 ggml-cuda.o: ggml-cuda.cu ggml-cuda.h
diff --git a/examples/common.h b/examples/common.h
index cbbc2dfab..734bcd3fa 100644
--- a/examples/common.h
+++ b/examples/common.h
@@ -20,7 +20,11 @@ struct gpt_params {
     int32_t repeat_last_n = 64;   // last n tokens to penalize
     int32_t n_parts       = -1;   // amount of model parts (-1 = determine from model dimensions)
     int32_t n_ctx         = 512;  // context size
+#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS) || defined (GGML_USE_CUBLAS)
+    int32_t n_batch       = 512;    // batch size for prompt processing (must be >=32 to use BLAS)
+#else
     int32_t n_batch       = 8;    // batch size for prompt processing
+#endif
     int32_t n_keep        = 0;    // number of tokens to keep from initial prompt
 
     // sampling parameters