Addressed review comments

Explicit precision set for tests

.github/workflows/ci_tests.yml

@@ -21,10 +21,10 @@ jobs:
       - name: test the dataloader without / with sanitize address
         run: |
           cd dev/test
-          make test_dataloader
+          make PRECISION=BF16 test_dataloader
           ./test_dataloader   
           make clean       
-          make test_dataloader TEST_CFLAGS="-fsanitize=address -fno-omit-frame-pointer"
+          make PRECISION=BF16 TEST_CFLAGS="-fsanitize=address -fno-omit-frame-pointer" test_dataloader 
           ./test_dataloader          
 
   ptx_and_sass_files:

dev/test/Makefile

@@ -19,7 +19,6 @@ NVCC_FLAGS = -O3 -t=0 --use_fast_math -std=c++17
 NVCC_LDFLAGS = -lcublas -lcublasLt
 NVCC_INCLUDES =
 NVCC_LDLIBS =
-NCLL_INCUDES =
 NVCC_CUDNN =
 # By default we don't build with cudnn because it blows up compile time from a few seconds to ~minute
 USE_CUDNN ?= 0
@@ -114,7 +113,10 @@ else
 endif
 
 # Precision settings, default to bf16 but ability to override
-PRECISION ?= BF16
+ifeq ($(MAKECMDGOALS), clean)
+  PRECISION=BF16 
+endif
+
 VALID_PRECISIONS := FP32 FP16 BF16
 ifeq ($(filter $(PRECISION),$(VALID_PRECISIONS)),)
   $(error Invalid precision $(PRECISION), valid precisions are $(VALID_PRECISIONS))