Merge branch 'master' into py-tqdm

Makefile

@@ -10,16 +10,16 @@ INCLUDES =
 # e.g. on MacOS: brew install libomp
 # e.g. on Ubuntu: sudo apt-get install libomp-dev
 # later, run the program by prepending the number of threads, e.g.: OMP_NUM_THREADS=8 ./gpt2
-ifeq ($(shell echo | $(CC) -Xpreprocessor -fopenmp -x c -E - > /dev/null 2>&1; echo $$?), 0)
+ifeq ($(shell echo | $(CC) -fopenmp -x c -E - > /dev/null 2>&1; echo $$?), 0)
   ifeq ($(shell uname), Darwin)
     # macOS with Homebrew
-    CFLAGS += -Xclang -fopenmp
+    CFLAGS += -Xclang -fopenmp -DOMP
     LDFLAGS += -L/opt/homebrew/opt/libomp/lib
     LDLIBS += -lomp
     INCLUDES += -I/opt/homebrew/opt/libomp/include
   else
     # Ubuntu or other Linux distributions
-    CFLAGS += -fopenmp
+    CFLAGS += -fopenmp -DOMP
     LDLIBS += -lgomp
   endif
   $(info NICE Compiling with OpenMP support)

requirements.txt

@@ -1,4 +1,5 @@
+tqdm
 numpy
 torch
 tiktoken
-tqdm
+transformers

train_gpt2.c

@@ -14,7 +14,9 @@ There will be other versions of this code that specialize it and make it fast.
 #include <time.h>
 #include <string.h>
 #include <unistd.h>
+#ifdef OMP
 #include <omp.h>
+#endif
 
 // ----------------------------------------------------------------------------
 // all the individual layers' forward and backward passes
@@ -1121,4 +1123,4 @@ int main() {
     gpt2_free(&model);
     return 0;
 }
-#endif
+#endif