[MRv2] Default to MRv1 when a connector is present (#42955)
Signed-off-by: NickLucche <nlucches@redhat.com>
This commit is contained in:
@@ -148,6 +148,13 @@ def test_is_default_v2_model_runner_model(model_config, expected):
|
|||||||
assert VllmConfig._is_default_v2_model_runner_model(config) is expected
|
assert VllmConfig._is_default_v2_model_runner_model(config) is expected
|
||||||
|
|
||||||
|
|
||||||
|
def test_use_v2_model_runner_defaults_to_v1_when_kv_connector_present():
|
||||||
|
config = SimpleNamespace(kv_transfer_config=object())
|
||||||
|
with patch.object(envs, "VLLM_USE_V2_MODEL_RUNNER", None):
|
||||||
|
result = VllmConfig.use_v2_model_runner.fget(config)
|
||||||
|
assert result is False
|
||||||
|
|
||||||
|
|
||||||
@pytest.mark.skip_global_cleanup
|
@pytest.mark.skip_global_cleanup
|
||||||
def test_with_hf_config_populates_missing_architectures_from_causal_lm_mapping(
|
def test_with_hf_config_populates_missing_architectures_from_causal_lm_mapping(
|
||||||
monkeypatch,
|
monkeypatch,
|
||||||
|
|||||||
@@ -498,6 +498,10 @@ class VllmConfig:
|
|||||||
if use_v2_model_runner is not None:
|
if use_v2_model_runner is not None:
|
||||||
return use_v2_model_runner
|
return use_v2_model_runner
|
||||||
|
|
||||||
|
# KVCache layout changes are breaking, let's stick with v1 for now (see #42846)
|
||||||
|
if self.kv_transfer_config is not None:
|
||||||
|
return False
|
||||||
|
|
||||||
if not self._is_default_v2_model_runner_model():
|
if not self._is_default_v2_model_runner_model():
|
||||||
return False
|
return False
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user