Skip to content

Commit

Permalink
[XLA:GPU] Turn --xla_gpu_triton_gemm_any on by default.
Browse files Browse the repository at this point in the history
PiperOrigin-RevId: 698442142
  • Loading branch information
bchetioui authored and Google-ML-Automation committed Feb 10, 2025
1 parent 14f0aaa commit c43b9c5
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion xla/debug_options_flags.cc
Original file line number Diff line number Diff line change
Expand Up @@ -208,7 +208,7 @@ DebugOptions DefaultDebugOptionsIgnoringFlags() {
opts.set_xla_gpu_enable_triton_gemm(true);
opts.set_xla_gpu_unsupported_enable_generic_triton_emitter_for_gemms(false);
opts.set_xla_gpu_enable_cudnn_int8x32_convolution_reordering(true);
opts.set_xla_gpu_triton_gemm_any(false);
opts.set_xla_gpu_triton_gemm_any(true);
opts.set_xla_gpu_verify_triton_fusion_numerics(false);

// Moving reduce-scatter out of while loops can increase memory footprint, so
Expand Down

0 comments on commit c43b9c5

Please sign in to comment.