[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 5
parent
1085b64425
commit
651a88c09a
@@ -1372,8 +1372,7 @@ def _patch_first_rank_pp_group(monkeypatch):
|
||||
|
||||
def test_deepseek_v4_mhc_broadcast_finalize_sums_hc_streams(monkeypatch):
|
||||
"""First finalize (at the end of load_weights) allocates
|
||||
hc_attn_fn_broadcast as hc_attn_fn summed over hc streams.
|
||||
"""
|
||||
hc_attn_fn_broadcast as hc_attn_fn summed over hc streams."""
|
||||
_patch_first_rank_pp_group(monkeypatch)
|
||||
layer = _make_mhc_decoder_layer(hc_mult=2, hidden_size=8)
|
||||
model = SimpleNamespace(start_layer=0, end_layer=1, layers=[layer])
|
||||
@@ -1388,8 +1387,7 @@ def test_deepseek_v4_mhc_broadcast_finalize_sums_hc_streams(monkeypatch):
|
||||
def test_deepseek_v4_mhc_broadcast_refit_refreshes_in_place(monkeypatch):
|
||||
"""Re-finalizing after a weight refit must copy into the existing
|
||||
broadcast tensor so its address stays stable for captured CUDA graphs,
|
||||
while picking up the new hc_attn_fn values.
|
||||
"""
|
||||
while picking up the new hc_attn_fn values."""
|
||||
_patch_first_rank_pp_group(monkeypatch)
|
||||
layer = _make_mhc_decoder_layer(hc_mult=2, hidden_size=8)
|
||||
model = SimpleNamespace(start_layer=0, end_layer=1, layers=[layer])
|
||||
|
||||
Reference in New Issue
Block a user