Skip to content

PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度

Fix omni model test CI issue (#1667)

Liang Lvbbda404699bd
diff --git a/test/test_cpu/models/test_omni_model.py b/test/test_cpu/models/test_omni_model.py
index 2c7968752..136f19b31 100644
--- a/test/test_cpu/models/test_omni_model.py
+++ b/test/test_cpu/models/test_omni_model.py
@@ -263,6 +263,7 @@ class TestQwen3OmniMoeReplacement:
"""Test that unfused weights match original fused weights."""
from auto_round.modeling.fused_moe.replace_modules import apply_replacements, materialize_model_
+ torch.manual_seed(42)
config = _make_tiny_qwen3_omni_moe_config()
model = Qwen3OmniMoeForConditionalGeneration(config)