[Quantization] enable compressed-tensors marlin support for turing (2) (#31008)

Signed-off-by: Jinzhen Lin <jinzhen.ljz@antgroup.com>

vllm/model_executor/layers/quantization/utils/marlin_utils.py

@@ -48,7 +48,7 @@ def query_marlin_supported_quant_types(
             -1 if capability_tuple is None else capability_tuple.to_int()
         )
 
-    if device_capability < 80:
+    if device_capability < 75:
         return []
 
     # - has_zp is True: return quant_types that has zero points

vllm/model_executor/layers/quantization/utils/marlin_utils_fp4.py

@@ -23,7 +23,7 @@ logger = init_logger(__name__)
 
 
 def is_fp4_marlin_supported():
-    return current_platform.has_device_capability(80)
+    return current_platform.has_device_capability(75)
 
 
 def nvfp4_marlin_process_scales(marlin_scales):

vllm/model_executor/layers/quantization/utils/marlin_utils_fp8.py

@@ -22,7 +22,7 @@ logger = init_logger(__name__)
 
 
 def is_fp8_marlin_supported():
-    return current_platform.has_device_capability(80)
+    return current_platform.has_device_capability(75)
 
 
 def fp8_fused_exponent_bias_into_scales(scales):