aparler/c-llm
1
0
Fork 0
mirror of https://github.com/karpathy/llm.c.git synced 2026-07-27 20:25:09 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

fixed up test and profile targets

Browse source
This commit is contained in:
Erik Schultheis 2024-05-04 11:59:59 +03:00
parent 67a82a611e
commit aa5bb258b6
3 changed files with 9 additions and 19 deletions
Download patch file Download diff file Expand all files Collapse all files

8
Makefile
View file

@ -203,14 +203,14 @@ train_gpt2cu: train_gpt2.cu cudnn_att.o
train_gpt2fp32cu: train_gpt2_fp32.cu
$(NVCC) $(NVCC_FLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE)
test_gpt2cu: test_gpt2.cu
$(NVCC) $(NVCC_FLAGS) $(PFLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE)
test_gpt2cu: test_gpt2.cu cudnn_att.o
$(NVCC) $(NVCC_FLAGS) $(PFLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE) cudnn_att.o
test_gpt2fp32cu: test_gpt2_fp32.cu
$(NVCC) $(NVCC_FLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE)
profile_gpt2cu: profile_gpt2.cu
$(NVCC) $(NVCC_FLAGS) $(PFLAGS) -lineinfo $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE)
profile_gpt2cu: profile_gpt2.cu cudnn_att.o
$(NVCC) $(NVCC_FLAGS) $(PFLAGS) -lineinfo $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE) cudnn_att.o
clean:
$(REMOVE_FILES) $(TARGETS)

9
profile_gpt2.cu
View file

@ -49,9 +49,7 @@ int main() {
cublasCheck(cublasSetMathMode(cublas_handle, cublas_math_mode));
// setup the (global) cuBLASLt workspace
cudaCheck(cudaMalloc(&cublaslt_workspace, cublaslt_workspace_size));
#ifdef ENABLE_CUDNN
checkCudnnErr(cudnnCreate(&cudnn_handle));
#endif
create_cudnn();
// build the GPT-2 model from a checkpoint
GPT2 model;
@ -81,10 +79,7 @@ int main() {
// free
gpt2_free(&model);
#ifdef ENABLE_CUDNN
if (cudnn_workspace != NULL) { cudaCheck(cudaFree(cudnn_workspace)); }
checkCudnnErr(cudnnDestroy(cudnn_handle));
#endif
destroy_cudnn();
cudaCheck(cudaFree(cublaslt_workspace));
cublasCheck(cublasDestroy(cublas_handle));
cublasCheck(cublasLtDestroy(cublaslt_handle));

11
test_gpt2.cu
View file

@ -106,10 +106,8 @@ int main(int argc, char *argv[]) {
cublasMath_t cublas_math_mode = enable_tf32 ? CUBLAS_TF32_TENSOR_OP_MATH : CUBLAS_DEFAULT_MATH;
cublasCheck(cublasSetMathMode(cublas_handle, cublas_math_mode));
cudaCheck(cudaMalloc(&cublaslt_workspace, cublaslt_workspace_size));
#ifdef ENABLE_CUDNN
checkCudnnErr(cudnnCreate(&cudnn_handle));
#endif
// set up cuDNN (noop if not available)
create_cudnn();
// build the GPT-2 model from a checkpoint
GPT2 model;
@ -326,10 +324,7 @@ int main(int argc, char *argv[]) {
free(grads_memory_cpu);
free(grads_memory_cpu_float);
gpt2_free(&model);
#ifdef ENABLE_CUDNN
if (cudnn_workspace != NULL) { cudaCheck(cudaFree(cudnn_workspace)); }
checkCudnnErr(cudnnDestroy(cudnn_handle));
#endif
destroy_cudnn();
cudaCheck(cudaFree(cublaslt_workspace));
cublasCheck(cublasDestroy(cublas_handle));
cublasCheck(cublasLtDestroy(cublaslt_handle));