mshr-h commented on code in PR #18524:
URL: https://github.com/apache/tvm/pull/18524#discussion_r2572965042
##########
tests/python/relax/test_frontend_from_exported_program.py:
##########
@@ -4255,6 +4257,20 @@ def main(
run_ep_decomposition=True,
)
+ # Test 2D input (seq_len, head_dim) - bug fix for #18441
+ class Attention2D(Module):
+ def forward(self, x):
+ return torch.nn.functional.scaled_dot_product_attention(x, x, x,
is_causal=False)
+
+ # For 2D input, we just verify that conversion succeeds without error
+ # The expected IR is complex due to reshape operations, so we use
verify_model_numerically
+ verify_model_numerically(
Review Comment:
Use structual equality `verify_model()` instead. E2E test should be avoided
due to computational cost.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
