[CI/Build] Split long weekly TTS jobs and route slow full-model cases to nightly (#8239)

Signed-off-by: wangyu <410167048@qq.com>
Co-authored-by: Alicia <115451386+congw729@users.noreply.github.com>
This commit is contained in:
wangyuandAlicia authored and GitHub committed 2026-09-30 09:08:24 +00:00
1 parent 004a5711dc
commit d0f073686a
6 files changed
+136 -21

No files matched your search

@@ -89,7 +89,6 @@ _DEPLOY_CONFIG: dict[str, Any] = {
"trust_remote_code": True,
"devices": "0,1,2,3", # set dynamically by _write_deploy_config
"vae_use_slicing": False,
"moe_backend": "flashinfer_cutlass",
"vae_use_tiling": False,
"parallel_config": {
"pipeline_parallel_size": 1,
@@ -16,7 +16,7 @@ from tests.e2e.online_serving.helpers.minicpmo_4_5_duplex import (
from tests.e2e.online_serving.helpers.minicpmo_window_e2e import run_window_turn
from tests.helpers.mark import hardware_test
pytestmark = [pytest.mark.omni, pytest.mark.advanced_model]
pytestmark = [pytest.mark.omni]
def _assert_complete(result, *, require_audio=False):
@@ -30,6 +30,7 @@ def _assert_complete(result, *, require_audio=False):
@hardware_test(res={"cuda": "H100", "npu": "A3"}, num_cards=1)
@pytest.mark.advanced_model
@pytest.mark.parametrize("omni_server", SERVER_PARAMS, indirect=True)
@pytest.mark.parametrize("mode,reference", [("basic", False), ("context", False), ("context", True)])
def test_window_rebuild_and_next_session(omni_server, mode, reference):
@@ -47,7 +48,8 @@ def test_window_rebuild_and_next_session(omni_server, mode, reference):
_assert_complete(result, require_audio=reference)
@hardware_test(res={"cuda": "H100", "npu": "A3"}, num_cards=1)
@hardware_test(res={"cuda": ["H100", "B200"], "npu": "A3"}, num_cards=1)
@pytest.mark.full_model
@pytest.mark.parametrize("omni_server", SERVER_PARAMS, indirect=True)
@pytest.mark.parametrize("mode,camera", [("basic", False), ("context", False), ("context", True)])
def test_window_continuous_input(omni_server, tmp_path, mode, camera):
@@ -71,6 +73,7 @@ def test_window_continuous_input(omni_server, tmp_path, mode, camera):
@hardware_test(res={"cuda": "H100", "npu": "A3"}, num_cards=1)
@pytest.mark.advanced_model
@pytest.mark.parametrize("omni_server", SERVER_PARAMS, indirect=True)
@pytest.mark.parametrize("mode", ["off", "basic", "context"])
def test_window_buffered_flush(omni_server, mode):
@@ -11,7 +11,7 @@ websocket (``/v1/realtime/robot/openpi``) — the same wire path a robot uses
The ``pi0_openpi_*`` helpers are protocol-level (handshake, send obs, receive
chunk) and not π0-specific, so π0.5 reuses them; only the deploy config and the
expected metadata differ. This file is selected explicitly by the nightly H100
expected metadata differ. This file is selected explicitly by the weekly H100
robot-policy job.
The in-process LeRobot parity oracle lives separately in
@@ -59,7 +59,7 @@ test_params = [
]
@pytest.mark.full_model
@pytest.mark.slow
@pytest.mark.diffusion
@hardware_test(res={"cuda": "H100"})
@pytest.mark.parametrize("omni_server", test_params, indirect=True)