### What this PR does / why we need it? 1. Replace manual memory cleanup with passing parameter. 2. FusedMoEPrepareAndFinalizeWithMC2 inherits All2All avoid duplicated code. 3. Fix MC2 bug introduced in https://github.com/vllm-project/vllm-ascend/pull/3365 4. Unify aclgraph & eager in W8A8_dynamic. ### Does this PR introduce _any_ user-facing change? No ### How was this patch tested? e2e & ut - vLLM version: v0.11.0rc3 - vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0 Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>
This commit is contained in:
@@ -141,6 +141,7 @@ def test_models_distributed_Qwen3_W4A8DYNAMIC_new_version(model):
|
||||
|
||||
@pytest.mark.parametrize("model", DEEPSEEK_W4A8_MODELS)
|
||||
@patch.dict(os.environ, {"VLLM_ASCEND_MLA_PA": "1"})
|
||||
@patch.dict(os.environ, {"HCCL_BUFFSIZE": "1024"})
|
||||
def test_models_distributed_DeepSeek_W4A8DYNAMIC(model):
|
||||
prompts = [
|
||||
"Hello, my name is",
|
||||
|
||||
Reference in New Issue
Block a user