Skip to content

[Test] Fix 4 broken Qwen3-TTS async chunk unit tests #2351

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.

Sign up for GitHub

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

Merged
linyueqian merged 2 commits into from
Mar 31, 2026
Merged

[Test] Fix 4 broken Qwen3-TTS async chunk unit tests #2351

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
37828fa
[Test] Fix 4 broken Qwen3-TTS async chunk unit tests
linyueqian Mar 31, 2026
4ea0e8b
[Test] Strengthen IC=16 assertion for high-load new request
linyueqian Mar 31, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
43 changes: 30 additions & 13 deletions tests/model_executor/stage_input_processors/test_qwen3_tts_async_chunk.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,8 @@
talker2code2wav_async_chunk,
)

pytestmark = [pytest.mark.core_model, pytest.mark.cpu]

_FRAME = [1, 2, 3, 4]
_Q = len(_FRAME)

Expand Down Expand Up @@ -90,7 +92,7 @@ def test_flush_on_finish():
is_finished=True,
)
assert p is not None
assert p["finished"].item() is True
assert p["finished"] is True
assert len(p["code_predictor_codes"]) == _Q * 24


Expand Down Expand Up @@ -158,24 +160,29 @@ def test_dynamic_ic_adapts_to_load():


def test_ic_load_change_mid_request():
"""IC stateless: load spike mid-request shifts initial_coverage."""
"""IC is cached per request; a load spike only affects new requests."""
tm = _tm(chunk_frames=25, left_context=25, max_num_seqs=8)

# Low load -> IC=2 -> emit at frame 2
# Low load -> IC=2 (cached for "r"), emit at frame 2
p1 = _call(tm, "r", n_frames=2)
assert p1 is not None

# Spike load: 6 others -> IC=16 -> initial_coverage=16
# Spike load: 6 others running
for i in range(6):
tm.code_prompt_token_ids[f"other-{i}"] = [[0]] * 10

# adjusted=25-16=9, 9%25!=0 -> hold
# IC for "r" is still cached as 2.
# initial_coverage = ((25-1)//2)*2 = 24, first normal emit at 24+25=49
assert _call(tm, "r", n_frames=25) is None

# First normal emit at 16+25=41
p3 = _call(tm, "r", n_frames=41)
assert _call(tm, "r", n_frames=27) is None
p3 = _call(tm, "r", n_frames=49)
assert p3 is not None
assert p3["left_context_size"] == 16

# A *new* request under high load gets IC=16 (not IC=2).
# Frame 2 would emit under IC=2 but must hold under IC=16.
assert _call(tm, "new_req", n_frames=2) is None
p4 = _call(tm, "new_req", n_frames=16)
assert p4 is not None
Comment on lines +184 to +185

@chatgpt-codex-connector chatgpt-codex-connector Bot Mar 31, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

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

P2 Badge Validate IC=16 for high-load new request

The new check in test_ic_load_change_mid_request only asserts that p4 is emitted, but that does not prove the cached/dynamic IC value is 16 as the test comment claims. With chunk_frames=25, frame 16 also emits for IC values 2/4/8, so a regression in load-based IC selection could slip through while this test still passes. Please assert an IC-dependent field (for example left_context_size) so the high-load behavior is actually verified.

Useful? React with 👍 / 👎.



@pytest.mark.parametrize(
Expand Down Expand Up @@ -295,8 +302,13 @@ def test_non_async_processor_prepends_ref_code_and_sets_trim_context():
],
dtype=torch.long,
)
output = SimpleNamespace(multimodal_output={"audio_codes": audio_codes, "ref_code": ref_code})
stage = SimpleNamespace(engine_outputs=[SimpleNamespace(outputs=[output])])
output = SimpleNamespace(
multimodal_output={"audio_codes": audio_codes, "ref_code": ref_code},
token_ids=list(range(3)),
)
stage = SimpleNamespace(
engine_outputs=[SimpleNamespace(outputs=[output], finished=True)],
)

prompts = talker2code2wav(stage_list=[stage], engine_input_source=[0])

Expand Down Expand Up @@ -335,8 +347,13 @@ def test_non_async_processor_filters_out_of_range_codec_values():
],
dtype=torch.long,
)
output = SimpleNamespace(multimodal_output={"audio_codes": audio_codes, "ref_code": ref_code})
stage = SimpleNamespace(engine_outputs=[SimpleNamespace(outputs=[output])])
output = SimpleNamespace(
multimodal_output={"audio_codes": audio_codes, "ref_code": ref_code},
token_ids=list(range(4)),
)
stage = SimpleNamespace(
engine_outputs=[SimpleNamespace(outputs=[output], finished=True)],
)

prompts = talker2code2wav(stage_list=[stage], engine_input_source=[0])

Expand Down
Loading