fix scale init in tests (#338)

Author: dsikka

tests/test_compressors/quantized_compressors/test_fp8_quant.py

@@ -61,8 +61,8 @@ def make_dummy_g_idx(columns: int, group_size: int) -> torch.Tensor:
         [
             QuantizationStrategy.GROUP,
             128,
-            torch.rand((512, 8, 1)) * 0.01,
-            torch.zeros((512, 8, 1), dtype=torch.int8),
+            torch.rand((512, 8)) * 0.01,
+            torch.zeros((512, 8), dtype=torch.int8),
         ],
         [
             QuantizationStrategy.CHANNEL,

tests/test_compressors/quantized_compressors/test_int_quant.py

@@ -53,8 +53,8 @@ def get_dummy_quant_config(strategy, group_size=None, symmetric=True):
             QuantizationStrategy.GROUP,
             True,
             128,
-            torch.rand((512, 8, 1)) * 0.01,
-            torch.zeros((512, 8, 1), dtype=torch.int8),
+            torch.rand((512, 8)) * 0.01,
+            torch.zeros((512, 8), dtype=torch.int8),
         ],
         [
             QuantizationStrategy.CHANNEL,

tests/test_quantization/lifecycle/test_forward.py

@@ -108,17 +108,17 @@ def test_forward_quantize(
             "int",
             QuantizationStrategy.GROUP,
             128,
-            torch.rand((512, 8, 1)) * 0.01,
-            torch.zeros((512, 8, 1)),
+            torch.rand((512, 8)) * 0.01,
+            torch.zeros((512, 8)),
             None,
         ),
         (
             4,
             "int",
             QuantizationStrategy.GROUP,
             128,
-            torch.rand((512, 8, 1)) * 0.01,
-            torch.zeros((512, 8, 1)),
+            torch.rand((512, 8)) * 0.01,
+            torch.zeros((512, 8)),
             make_dummy_g_idx(1024, 128),
         ),
         (
@@ -135,17 +135,17 @@ def test_forward_quantize(
             "float",
             QuantizationStrategy.GROUP,
             128,
-            torch.rand((512, 8, 1)) * 0.01,
-            torch.zeros((512, 8, 1)),
+            torch.rand((512, 8)) * 0.01,
+            torch.zeros((512, 8)),
             None,
         ),
         (
             8,
             "float",
             QuantizationStrategy.GROUP,
             128,
-            torch.rand((512, 8, 1)) * 0.01,
-            torch.zeros((512, 8, 1)),
+            torch.rand((512, 8)) * 0.01,
+            torch.zeros((512, 8)),
             make_dummy_g_idx(1024, 128),
         ),
     ],
@@ -174,17 +174,17 @@ def test_quantize(num_bits, type, strategy, group_size, scale, zero_point, g_idx
             "int",
             QuantizationStrategy.GROUP,
             128,
-            torch.rand((512, 8, 1)) * 0.01,
-            torch.zeros((512, 8, 1)),
+            torch.rand((512, 8)) * 0.01,
+            torch.zeros((512, 8)),
             None,
         ),
         (
             8,
             "int",
             QuantizationStrategy.GROUP,
             128,
-            torch.rand((512, 8, 1)) * 0.01,
-            torch.zeros((512, 8, 1)),
+            torch.rand((512, 8)) * 0.01,
+            torch.zeros((512, 8)),
             make_dummy_g_idx(1024, 128),
         ),
     ],