Merge pull request #713 from mspronesti/master

fix(dev/cuda): memory leaks
karpathy · Jul 28, 2024 · 7379cd1 · 7379cd1
2 parents 362c6a8 + c6e74ea
commit 7379cd1
Show file tree

Hide file tree

Showing 5 changed files with 5 additions and 0 deletions.
diff --git a/dev/cuda/attention_backward.cu b/dev/cuda/attention_backward.cu
@@ -1137,6 +1137,7 @@ int main(int argc, char **argv) {
     free(dinp);
     free(dpreatt);
     free(datt);
+    free(h_dinp);
     cudaCheck(cudaFree(d_inp));
     cudaCheck(cudaFree(d_qkvr));
     cudaCheck(cudaFree(d_preatt));

diff --git a/dev/cuda/attention_forward.cu b/dev/cuda/attention_forward.cu
@@ -1377,6 +1377,7 @@ int main(int argc, char **argv) {
     cudaCheck(cudaFree(d_preatt));
     cudaCheck(cudaFree(d_att));
     cudaCheck(cudaFree(d_inp));
+    cudaCheck(cudaFree(d_stats));
     cublasDestroy(cublas_handle);
 
     #ifdef ENABLE_CUDNN

diff --git a/dev/cuda/classifier_fused.cu b/dev/cuda/classifier_fused.cu
@@ -766,6 +766,7 @@ int main(int argc, char **argv) {
     cudaCheck(cudaFree(d_logits));
     cudaCheck(cudaFree(d_dlosses));
     cudaCheck(cudaFree(d_targets));
+    cudaCheck(cudaFree(d_dlogits_no_pad));
 
     return 0;
 }
diff --git a/dev/cuda/nccl_all_reduce.cu b/dev/cuda/nccl_all_reduce.cu
@@ -193,5 +193,6 @@ int main(int argc, char **argv) {
 
   free(all_reduce_buffer_host);
   cudaCheck(cudaFree(all_reduce_buffer));
+  cudaCheck(cudaFree(all_reduce_buffer_recv));
   multi_gpu_config_free(&multi_gpu_config);
 }
diff --git a/dev/cuda/trimat_forward.cu b/dev/cuda/trimat_forward.cu
@@ -643,6 +643,7 @@ int main(int argc, char **argv) {
     free(inp);
     cudaCheck(cudaFree(d_out));
     cudaCheck(cudaFree(d_inp));
+    cudaCheck(cudaFree(d_qkvr));
     cublasDestroy(cublas_handle);
 
     return 0;