diff --git a/examples/llm_ptq/example_utils.py b/examples/llm_ptq/example_utils.py index 9c692e5b7aa..6dbf894e9b7 100755 --- a/examples/llm_ptq/example_utils.py +++ b/examples/llm_ptq/example_utils.py @@ -55,11 +55,7 @@ SPECULATIVE_MODEL_LIST = ["Eagle", "Medusa"] # TODO: Refactor into the config system. -_QWEN36_AUTOQ_DISABLED_LAYERS = ( - "*shared_expert_gate*", - "*linear_attn.in_proj_a*", - "*linear_attn.in_proj_b*", -) +_QWEN36_AUTOQ_DISABLED_LAYERS = ("*shared_expert_gate*",) _VLM_AUTOQ_DISABLED_LAYERS = ("*visual*", "*mtp*", "*vision_tower*") diff --git a/tests/examples/llm_ptq/test_hf_ptq_args.py b/tests/examples/llm_ptq/test_hf_ptq_args.py index 94f0b079df7..b8ee9f1a76b 100644 --- a/tests/examples/llm_ptq/test_hf_ptq_args.py +++ b/tests/examples/llm_ptq/test_hf_ptq_args.py @@ -97,8 +97,6 @@ def test_qwen_autoquant_disabled_layers_are_scoped_to_qwen_models(monkeypatch): llama_model = SimpleNamespace(config=SimpleNamespace(model_type="llama")) qwen_only_patterns = { "*shared_expert_gate*", - "*linear_attn.in_proj_a*", - "*linear_attn.in_proj_b*", } monkeypatch.setattr(example_utils, "is_multimodal_model", lambda model: False)