fix and cleanup after merge

tests/kernels/test_moe.py

@@ -257,7 +257,7 @@ def test_fused_marlin_moe(
 @pytest.mark.parametrize("group_size", [-1, 32, 64, 128])
 @pytest.mark.parametrize("act_order", [True, False])
 @pytest.mark.parametrize("num_bits", [4, 8])
-def test_marlin_moe_mmm(
+def test_single_marlin_moe_multiply(
     m: int,
     n: int,
     k: int,

vllm/model_executor/layers/quantization/gptq_marlin.py

@@ -611,4 +611,5 @@ def apply(
             topk_ids,
             w1_scale=layer.w13_scales,
             w2_scale=layer.w2_scales,
+            num_bits=self.quant_config.quant_type.size_bits,
         ).to(orig_dtype)

vllm/model_executor/model_loader/utils.py

@@ -23,13 +23,7 @@ def get_model_architecture(
     architectures = getattr(model_config.hf_config, "architectures", [])
     # Special handling for quantized Mixtral.
     # FIXME(woosuk): This is a temporary hack.
-    mixtral_supported = ["fp8", "compressed-tensors"]
-    # for gptq_marlin, only run fused MoE for int4
-    if model_config.quantization == "gptq_marlin":
-        hf_quant_config = getattr(model_config.hf_config,
-                                  "quantization_config", None)
-        if hf_quant_config and hf_quant_config.get("bits") == 4:
-            mixtral_supported.append("gptq_marlin")
+    mixtral_supported = ["fp8", "compressed-tensors", "gptq_marlin"]
 
     if (model_config.quantization is not None
             and model_config.quantization not in mixtral_supported