bayesianflow-for-chem 1.2.5__py3-none-any.whl → 1.2.6__py3-none-any.whl

bayesianflow_for_chem/__init__.py

@@ -7,5 +7,5 @@ from . import data, tool, train, scorer
 from .model import ChemBFN, MLP
 
 __all__ = ["data", "tool", "train", "scorer", "ChemBFN", "MLP"]
-__version__ = "1.2.5"
+__version__ = "1.2.6"
 __author__ = "Nianze A. Tao (Omozawa Sueno)"

bayesianflow_for_chem/tool.py

@@ -492,15 +492,16 @@ def inpaint(
 
 def quantise_model(model: ChemBFN) -> nn.Module:
     """
-    Dynamic quantisation of the trained model.
+    Dynamic quantisation of the trained model to `torch.qint8` data type.
 
     :param model: trained ChemBFN model
     :type model: bayesianflow_for_chem.model.ChemBFN
     :return: quantised model
     :rtype: torch.nn.Module
     """
-    from torch.ao.nn.quantized.modules.utils import _quantize_weight
     from torch.ao.nn.quantized import dynamic
+    from torch.ao.nn.quantized.modules.utils import _quantize_weight
+    from torch.ao.quantization.qconfig import default_dynamic_qconfig
 
     class QuantisedLinear(dynamic.Linear):
         # Modified from https://github.com/pytorch/pytorch/blob/main/torch/ao/nn/quantized/dynamic/modules/linear.py
@@ -543,7 +544,22 @@ def quantise_model(model: ChemBFN) -> nn.Module:
             self._packed_params.requires_grad_(False)
 
         def forward(self, x: Tensor) -> Tensor:
-            result = dynamic.Linear.forward(self, x)
+            if self._packed_params.dtype == torch.qint8:
+                if self.version is None or self.version < 4:
+                    Y = torch.ops.quantized.linear_dynamic(
+                        x, self._packed_params._packed_params
+                    )
+                else:
+                    Y = torch.ops.quantized.linear_dynamic(
+                        x, self._packed_params._packed_params, reduce_range=True
+                    )
+            elif self._packed_params.dtype == torch.float16:
+                Y = torch.ops.quantized.linear_dynamic_fp16(
+                    x, self._packed_params._packed_params
+                )
+            else:
+                raise RuntimeError("Unsupported dtype on dynamic quantized linear!")
+            result = Y.to(x.dtype)
             if self.lora_enabled and isinstance(self.lora_dropout, float):
                 result += (
                     nn.functional.dropout(x, self.lora_dropout, self.training)
@@ -562,11 +578,6 @@ def quantise_model(model: ChemBFN) -> nn.Module:
             if mod.qconfig is not None and mod.qconfig.weight is not None:
                 weight_observer = mod.qconfig.weight()
             else:
-                # We have the circular import issues if we import the qconfig in the beginning of this file:
-                # https://github.com/pytorch/pytorch/pull/24231. The current workaround is to postpone the
-                # import until we need it.
-                from torch.ao.quantization.qconfig import default_dynamic_qconfig
-
                 weight_observer = default_dynamic_qconfig.weight()
             dtype = weight_observer.dtype
             assert dtype in [torch.qint8, torch.float16], (

{bayesianflow_for_chem-1.2.5.dist-info → bayesianflow_for_chem-1.2.6.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: bayesianflow_for_chem
-Version: 1.2.5
+Version: 1.2.6
 Summary: Bayesian flow network framework for Chemistry
 Home-page: https://augus1999.github.io/bayesian-flow-network-for-chemistry/
 Author: Nianze A. Tao

{bayesianflow_for_chem-1.2.5.dist-info → bayesianflow_for_chem-1.2.6.dist-info}/RECORD

@@ -1,12 +1,12 @@
-bayesianflow_for_chem/__init__.py,sha256=GMGe5nU963qFL6vJ9OZSfqfSyEImC_P2zyUS0cyP3Mg,293
+bayesianflow_for_chem/__init__.py,sha256=sdyCK-Zd32-FNOcjuSB02ABx8vn53phorQeVqyWMWk4,293
 bayesianflow_for_chem/data.py,sha256=9tpRba40lxwrB6aPSJMkxUglEVC3VEQC9wWxhDuz3Q8,7760
 bayesianflow_for_chem/model.py,sha256=HvEvW_xRbkv4eSv5lhd72BJMZkg-ZACEi1DAW3p5Q1Y,35918
 bayesianflow_for_chem/scorer.py,sha256=mV1vX8aBGFra2BE7N8WHihVIo3dXmUdPQIGfSaiuNdk,4084
-bayesianflow_for_chem/tool.py,sha256=tJjb8q3_orNkj2BYJwz5VxqeaOv55dvqO93_uigLJIk,23221
+bayesianflow_for_chem/tool.py,sha256=VuEqbT7Qraa4vnKMHbToyAYIiRoQI7gEPLKEBCWGmVg,23706
 bayesianflow_for_chem/train.py,sha256=kj6icGqymUUYopDtpre1oE_wpvpeNilbpzgffBsd1tk,9589
 bayesianflow_for_chem/vocab.txt,sha256=HgtAZmpWYk4y8PqEVC4vqut1vE75DfRKE_10s2UW0rU,790
-bayesianflow_for_chem-1.2.5.dist-info/LICENSE,sha256=hIahDEOTzuHCU5J2nd07LWwkLW7Hko4UFO__ffsvB-8,34523
-bayesianflow_for_chem-1.2.5.dist-info/METADATA,sha256=hwEEDW6ipmHpjRjQDKxWk5zqI9jwjsl-yxBpvYn93HQ,5890
-bayesianflow_for_chem-1.2.5.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-bayesianflow_for_chem-1.2.5.dist-info/top_level.txt,sha256=KHsanI3BMCt8D9Qpze2ycrF6nMa3PyojgO6eS1c8kco,22
-bayesianflow_for_chem-1.2.5.dist-info/RECORD,,
+bayesianflow_for_chem-1.2.6.dist-info/LICENSE,sha256=hIahDEOTzuHCU5J2nd07LWwkLW7Hko4UFO__ffsvB-8,34523
+bayesianflow_for_chem-1.2.6.dist-info/METADATA,sha256=Akoh5dQW_0jeYuGC4ZKKYHS1WJn0xRwGDr7ut-Q-5sc,5890
+bayesianflow_for_chem-1.2.6.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+bayesianflow_for_chem-1.2.6.dist-info/top_level.txt,sha256=KHsanI3BMCt8D9Qpze2ycrF6nMa3PyojgO6eS1c8kco,22
+bayesianflow_for_chem-1.2.6.dist-info/RECORD,,