Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[GEMM] [Tuning] Option to try different initialization strategies #486

Merged
merged 9 commits into from
Jan 27, 2024

Conversation

vgokhale
Copy link
Collaborator

Added init strategies:

  1. trig_float: element at col i in row j gets value sin(i)
  2. hpl: uniform random in [-0.5, 0.5)
  3. randn: default, uniform random in [0, 1)
  4. zeros: all 0.0s

@vgokhale vgokhale self-assigned this Jan 25, 2024
Copy link

@zhanglx13 zhanglx13 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just a small nitpick. But it shouldn't block us from merging this PR.

@@ -673,7 +690,7 @@ def main():
print(f"Tuning starts at: {start_time}", flush=True)
f_results = open(tuning_output_file, 'w')

for (M, N, K, col_a, col_b, myConfig) in mnks:
for (M, N, K, col_a, col_b, init_type, myConfig) in mnks:

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we can remove init_type from mnks. All gemm sizes in a single tuning/benchmarking session can shared the same init_type

@zhanglx13 zhanglx13 changed the title Option to try different initialization strategies [GEMM] [Tuning] Option to try different initialization strategies Jan 27, 2024
@zhanglx13 zhanglx13 merged commit 49f6c3a into triton-mlir Jan 27, 2024
2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants