compressed-tensors-nightly 0.7.1.20241023__tar.gz → 0.7.1.20241024__tar.gz

{compressed-tensors-nightly-0.7.1.20241023/src/compressed_tensors_nightly.egg-info → compressed-tensors-nightly-0.7.1.20241024}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: compressed-tensors-nightly
-Version: 0.7.1.20241023
+Version: 0.7.1.20241024
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.

{compressed-tensors-nightly-0.7.1.20241023 → compressed-tensors-nightly-0.7.1.20241024}/src/compressed_tensors/quantization/lifecycle/calibration.py

@@ -64,14 +64,12 @@ def set_module_for_calibration(module: Module, quantize_weights_upfront: bool =
                 quantization_args=module.quantization_scheme.weights,
             )
 
-        observer = module.weight_observer
-
-        g_idx = getattr(module, "weight_g_idx", None)
-
         offloaded = is_module_offloaded(module)
         if offloaded:
             module._hf_hook.pre_forward(module)
 
+        observer = module.weight_observer
+        g_idx = getattr(module, "weight_g_idx", None)
         scale, zero_point = observer(module.weight, g_idx=g_idx)
         update_parameter_data(module, scale, "weight_scale")
         update_parameter_data(module, zero_point, "weight_zero_point")

{compressed-tensors-nightly-0.7.1.20241023 → compressed-tensors-nightly-0.7.1.20241024}/src/compressed_tensors/quantization/lifecycle/frozen.py

@@ -14,7 +14,6 @@
 
 
 from compressed_tensors.quantization.quant_config import QuantizationStatus
-from compressed_tensors.quantization.utils import is_kv_cache_quant_scheme
 from torch.nn import Module
 
 

{compressed-tensors-nightly-0.7.1.20241023 → compressed-tensors-nightly-0.7.1.20241024}/src/compressed_tensors/quantization/observers/helpers.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 from collections import Counter
-from typing import Optional, Tuple
+from typing import Tuple
 
 import torch
 from compressed_tensors.quantization.quant_args import (

{compressed-tensors-nightly-0.7.1.20241023 → compressed-tensors-nightly-0.7.1.20241024}/src/compressed_tensors/quantization/observers/mse.py

@@ -70,7 +70,9 @@ class MovingAverageMSEObserver(Observer):
             absolute_min_val = torch.amin(observed, dim=reduce_dims, keepdims=True)
             absolute_max_val = torch.amax(observed, dim=reduce_dims, keepdims=True)
 
-        best = torch.full_like(absolute_min_val, torch.finfo(absolute_min_val.dtype).max)
+        best = torch.full_like(
+            absolute_min_val, torch.finfo(absolute_min_val.dtype).max
+        )
         min_val = torch.ones_like(absolute_min_val)
         max_val = torch.zeros_like(absolute_max_val)
         for i in range(int(self.maxshrink * self.grid)):

{compressed-tensors-nightly-0.7.1.20241023 → compressed-tensors-nightly-0.7.1.20241024/src/compressed_tensors_nightly.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: compressed-tensors-nightly
-Version: 0.7.1.20241023
+Version: 0.7.1.20241024
 Summary: Library for utilization of compressed safetensors of neural network models
 Home-page: https://github.com/neuralmagic/compressed-tensors
 Author: Neuralmagic, Inc.