Changed the conversion function name for better understanding

Signed-off-by: Hrishith Thadicherla <[email protected]>

Author: hthadicherla

modelopt/onnx/quantization/int4.py

@@ -98,7 +98,7 @@
 # supported and working
 CLIP_MIN = 1e-5
 
-def convert_ml_dtypes_int4_to_int8_format(tensor):
+def safe_cupy_array(tensor):
     """
     Convert ml_dtypes.int4 tensor to numpy.int8 for CuPy compatibility.
     

tests/gpu/onnx/test_quantize_onnx_torch_int4_awq.py

@@ -94,19 +94,19 @@ def _forward_loop(model, dataloader):
         scale_awq_lite = find_init(onnx_model_awq_lite, scale_names[i])
 
         if int4.has_cupy:
-            wq_onnx_awq_lite = int4.convert_ml_dtypes_int4_to_int8_format(wq_onnx_awq_lite)
-            scale_awq_lite = int4.convert_ml_dtypes_int4_to_int8_format(scale_awq_lite)
+            wq_onnx_awq_lite = int4.safe_cupy_array(wq_onnx_awq_lite)
+            scale_awq_lite = int4.safe_cupy_array(scale_awq_lite)
 
         wq_onnx_awq_lite = dq_tensor(wq_onnx_awq_lite, scale_awq_lite, block_size)
         wq_torch_awq_clip = model_torch_copy.net[i * 2].weight_quantizer(
             model_torch_copy.net[i * 2].weight
         )
         wq_onnx_awq_clip = find_init(onnx_model_awq_clip, wq_names[i])
         scale_awq_clip = find_init(onnx_model_awq_clip, scale_names[i])
-        
+
         if int4.has_cupy:
-            wq_onnx_awq_clip = int4.convert_ml_dtypes_int4_to_int8_format(wq_onnx_awq_clip)
-            scale_awq_clip = int4.convert_ml_dtypes_int4_to_int8_format(scale_awq_clip)
+            wq_onnx_awq_clip = int4.safe_cupy_array(wq_onnx_awq_clip)
+            scale_awq_clip = int4.safe_cupy_array(scale_awq_clip)
 
         wq_onnx_awq_clip = dq_tensor(wq_onnx_awq_clip, scale_awq_clip, block_size)