A high-throughput and memory-efficient inference and serving engine for LLMs
[Model Runner V2] Enable GraniteMOE for MRv2 by default (#45461)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Jun 16, 2026 at 12:31 UTC
88a9cdd4397932aa189b9999b346f5cde427ef1e
2 files changed
+2
-1
tests/test_config.py
+1
-1
@@ -168,7 +168,7 @@ def test_v2_model_runner_env_tri_state(monkeypatch, env_value, expected):
168
is_moe=True,
169
is_quantized=False,
170
),
171
- False,
171
+ True,
172
),
173
(
174
SimpleNamespace(
vllm/config/vllm.py
+1
@@ -70,6 +70,7 @@ DEFAULT_V2_MODEL_RUNNER_ARCHITECTURES = frozenset(
70
"Qwen3ForCausalLM",
71
"DeepseekV2ForCausalLM",
72
"Qwen2MoeForCausalLM",
73
+ "GraniteMoeForCausalLM",
74
"LlamaForCausalLM",
75
"MistralForCausalLM",
76
}