[ROCm][CI] Fix sparse MLA metadata sync fixture (#49128)
Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com> Co-authored-by: OpenAI Codex <noreply@openai.com>
This commit is contained in:
co-authored by
OpenAI Codex
parent
2730b657c4
commit
752bd10647
@@ -58,6 +58,7 @@ def _make_builder():
|
||||
max_num_batched_tokens + 1, dtype=torch.int32, device="cpu"
|
||||
)
|
||||
builder._num_attention_heads = 16
|
||||
builder._num_compute_units = current_platform.num_compute_units()
|
||||
builder._mla_work_meta_data = torch.empty(1, dtype=torch.int32, device="cpu")
|
||||
builder._mla_work_indptr = torch.empty(1, dtype=torch.int32, device="cpu")
|
||||
builder._mla_work_info_set = torch.empty(1, dtype=torch.int32, device="cpu")
|
||||
@@ -116,6 +117,7 @@ def test_sparse_persistent_metadata_syncs_only_after_recompute(monkeypatch):
|
||||
|
||||
assert events == ["metadata", "sync"]
|
||||
assert fake_get_mla_metadata_v1_mock.call_count == 1
|
||||
assert fake_get_mla_metadata_v1_mock.call_args.kwargs["max_split_per_batch"] == 1
|
||||
|
||||
events.clear()
|
||||
|
||||
|
||||
Reference in New Issue
Block a user