[Misc] Update GPTQ to use vLLMParameters (#7976)

tests/weight_loading/test_weight_loading.py

@@ -1,5 +1,7 @@
 import os
 
+import torch
+
 MAX_MODEL_LEN = 1024
 MODEL_NAME = os.environ.get("MODEL_NAME",
                             "robertgshaw2/zephyr-7b-beta-channelwise-gptq")
@@ -8,9 +10,12 @@ QUANTIZATION = os.environ.get("QUANTIZATION", "gptq_marlin")
 
 
 def test_weight_loading(vllm_runner):
+    """
+    Test parameter weight loading with tp>1.
+    """
     with vllm_runner(model_name=MODEL_NAME,
                      revision=REVISION,
-                     dtype="auto",
+                     dtype=torch.half if QUANTIZATION == "gptq" else "auto",
                      quantization=QUANTIZATION,
                      max_model_len=MAX_MODEL_LEN,
                      tensor_parallel_size=2) as model: