diff --git a/Makefile b/Makefile
index ff7c533ba..9a731e36c 100644
--- a/Makefile
+++ b/Makefile
@@ -42,6 +42,7 @@ endif
 CFLAGS   = -I.            -I./include -I./include/CL -I./otherarch -I./otherarch/tools -I./include/vulkan -O3 -DNDEBUG -std=c11   -fPIC -DLOG_DISABLE_LOGS -D_GNU_SOURCE
 CXXFLAGS = -I. -I./common -I./include -I./include/CL -I./otherarch -I./otherarch/tools -I./include/vulkan -O3 -DNDEBUG -std=c++11 -fPIC -DLOG_DISABLE_LOGS -D_GNU_SOURCE
 LDFLAGS  =
+FASTCFLAGS = $(subst -O3,-Ofast,$(CFLAGS))
 
 # these are used on windows, to build some libraries with extra old device compatibility
 SIMPLECFLAGS =
@@ -380,23 +381,23 @@ $(info )
 #
 
 ggml.o: ggml.c ggml.h ggml-cuda.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) -c $< -o $@
 ggml_v4_openblas.o: ggml.c ggml.h ggml-cuda.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) $(OPENBLAS_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(OPENBLAS_FLAGS) -c $< -o $@
 ggml_v4_failsafe.o: ggml.c ggml.h ggml-cuda.h
-	$(CC)  $(CFLAGS) $(NONECFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(NONECFLAGS) -c $< -o $@
 ggml_v4_noavx2.o: ggml.c ggml.h ggml-cuda.h
-	$(CC)  $(CFLAGS) $(SIMPLECFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(SIMPLECFLAGS) -c $< -o $@
 ggml_v4_clblast.o: ggml.c ggml.h ggml-cuda.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
 ggml_v4_cublas.o: ggml.c ggml.h ggml-cuda.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) $(CUBLAS_FLAGS) $(HIPFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(CUBLAS_FLAGS) $(HIPFLAGS) -c $< -o $@
 ggml_v4_clblast_noavx2.o: ggml.c ggml.h ggml-cuda.h
-	$(CC)  $(CFLAGS) $(SIMPLECFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(SIMPLECFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
 ggml_v4_vulkan.o: ggml.c ggml.h ggml-cuda.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) $(VULKAN_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(VULKAN_FLAGS) -c $< -o $@
 ggml_v4_vulkan_noavx2.o: ggml.c ggml.h ggml-cuda.h
-	$(CC)  $(CFLAGS) $(SIMPLECFLAGS) $(VULKAN_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(SIMPLECFLAGS) $(VULKAN_FLAGS) -c $< -o $@
 
 #quants
 ggml-quants.o: ggml-quants.c ggml.h ggml-quants.h ggml-cuda.h
@@ -415,41 +416,41 @@ ggml-backend.o: ggml-backend.c ggml.h ggml-backend.h
 
 #version 3 libs
 ggml_v3.o: otherarch/ggml_v3.c otherarch/ggml_v3.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) -c $< -o $@
 ggml_v3_openblas.o: otherarch/ggml_v3.c otherarch/ggml_v3.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) $(OPENBLAS_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(OPENBLAS_FLAGS) -c $< -o $@
 ggml_v3_failsafe.o: otherarch/ggml_v3.c otherarch/ggml_v3.h
-	$(CC)  $(CFLAGS) $(NONECFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(NONECFLAGS) -c $< -o $@
 ggml_v3_noavx2.o: otherarch/ggml_v3.c otherarch/ggml_v3.h
-	$(CC)  $(CFLAGS) $(SIMPLECFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(SIMPLECFLAGS) -c $< -o $@
 ggml_v3_clblast.o: otherarch/ggml_v3.c otherarch/ggml_v3.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
 ggml_v3_cublas.o: otherarch/ggml_v3.c otherarch/ggml_v3.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) $(CUBLAS_FLAGS) $(HIPFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(CUBLAS_FLAGS) $(HIPFLAGS) -c $< -o $@
 ggml_v3_clblast_noavx2.o: otherarch/ggml_v3.c otherarch/ggml_v3.h
-	$(CC)  $(CFLAGS) $(SIMPLECFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(SIMPLECFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
 
 #version 2 libs
 ggml_v2.o: otherarch/ggml_v2.c otherarch/ggml_v2.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) -c $< -o $@
 ggml_v2_openblas.o: otherarch/ggml_v2.c otherarch/ggml_v2.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) $(OPENBLAS_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(OPENBLAS_FLAGS) -c $< -o $@
 ggml_v2_failsafe.o: otherarch/ggml_v2.c otherarch/ggml_v2.h
-	$(CC)  $(CFLAGS) $(NONECFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(NONECFLAGS) -c $< -o $@
 ggml_v2_noavx2.o: otherarch/ggml_v2.c otherarch/ggml_v2.h
-	$(CC)  $(CFLAGS) $(SIMPLECFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(SIMPLECFLAGS) -c $< -o $@
 ggml_v2_clblast.o: otherarch/ggml_v2.c otherarch/ggml_v2.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
 ggml_v2_cublas.o: otherarch/ggml_v2.c otherarch/ggml_v2.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) $(CUBLAS_FLAGS) $(HIPFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) $(CUBLAS_FLAGS) $(HIPFLAGS) -c $< -o $@
 ggml_v2_clblast_noavx2.o: otherarch/ggml_v2.c otherarch/ggml_v2.h
-	$(CC)  $(CFLAGS) $(SIMPLECFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(SIMPLECFLAGS) $(CLBLAST_FLAGS) -c $< -o $@
 
 #extreme old version compat
 ggml_v1.o: otherarch/ggml_v1.c otherarch/ggml_v1.h
-	$(CC)  $(CFLAGS) $(FULLCFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(FULLCFLAGS) -c $< -o $@
 ggml_v1_failsafe.o: otherarch/ggml_v1.c otherarch/ggml_v1.h
-	$(CC)  $(CFLAGS) $(NONECFLAGS) -c $< -o $@
+	$(CC)  $(FASTCFLAGS) $(NONECFLAGS) -c $< -o $@
 
 #opencl
 ggml-opencl.o: ggml-opencl.cpp ggml-opencl.h
diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
index 95b72d69d..02c7d56b0 100644
--- a/gpttype_adapter.cpp
+++ b/gpttype_adapter.cpp
@@ -683,7 +683,14 @@ void PurgeMissingTokens(llama_context * ctx, std::vector<int> &current_context_t
 
 static int GetBatchSize(int desiredBlasBatchSize,FileFormat in_file_format)
 {
-    if(desiredBlasBatchSize<=0)
+    //check if approved to use BLAS
+    bool approved_format = !(file_format == FileFormat::BADFORMAT ||
+                            file_format == FileFormat::GPT2_1 ||
+                            file_format == FileFormat::GPTJ_1 ||
+                            file_format == FileFormat::GPTJ_2 ||
+                            file_format == FileFormat::RWKV_1 ||
+                            file_format==FileFormat::RWKV_2);
+    if(!approved_format || desiredBlasBatchSize<=0)
     {
         desiredBlasBatchSize = 16;
     }
@@ -1684,14 +1691,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs, generation_o
         }
     }
 
-    //if using BLAS and prompt is big enough, switch to single thread and use a huge batch
-    bool approved_format = !(file_format == FileFormat::BADFORMAT ||
-                            file_format == FileFormat::GPT2_1 ||
-                            file_format == FileFormat::GPTJ_1 ||
-                            file_format == FileFormat::GPTJ_2 ||
-                            file_format == FileFormat::RWKV_1 ||
-                            file_format==FileFormat::RWKV_2);
-    bool blasmode = (approved_format && embd_inp.size() >= 32 && ggml_cpu_has_blas() && kcpp_params->n_batch>=32);
+    bool blasmode = (embd_inp.size() >= 32 && ggml_cpu_has_blas() && kcpp_params->n_batch>=32);
 
     current_context_tokens.resize(n_past);