update

Signed-off-by: ssbuild <[email protected]>
ssbuild · Aug 4, 2023 · f9e788e · f9e788e
1 parent 659b177
commit f9e788e
Show file tree

Hide file tree

Showing 3 changed files with 5 additions and 5 deletions.
diff --git a/setup.py b/setup.py
@@ -19,7 +19,7 @@
 ]
 setup(
     name='deep_training',
-    version='0.1.13.post1',
+    version='0.1.13.post2',
     description='an easy training architecture',
     long_description='torch_training: https://github.com/ssbuild/deep_training.git',
     license='Apache License 2.0',

diff --git a/src/deep_training/nlp/models/internlm/quantization.py b/src/deep_training/nlp/models/internlm/quantization.py
@@ -161,7 +161,7 @@ def quantize(model, bits, empty_init=False, device=None,**kwarg):
                     QuantizedLinear(
                         bits=bits,
                         weight=w.weight.to(torch.cuda.current_device()),
-                        bias=w.bias.to(torch.cuda.current_device()),
+                        bias=w.bias.to(torch.cuda.current_device()) if w.bias is not None else None,
                         empty_init=empty_init,
                         device=w.weight.device if device is None else device,
                         dtype=w.weight.dtype,
@@ -176,7 +176,7 @@ def quantize(model, bits, empty_init=False, device=None,**kwarg):
                     QuantizedLinear(
                         bits=bits,
                         weight=w.weight.to(torch.cuda.current_device()),
-                        bias=w.bias.to(torch.cuda.current_device()),
+                        bias=w.bias.to(torch.cuda.current_device()) if w.bias is not None else None,
                         empty_init=empty_init,
                         device=w.weight.device if device is None else device,
                         dtype=w.weight.dtype,

diff --git a/src/deep_training/nlp/models/qwen/quantization.py b/src/deep_training/nlp/models/qwen/quantization.py
@@ -160,7 +160,7 @@ def quantize(model, bits, empty_init=False, device=None,**kwarg):
                     QuantizedLinear(
                         bits=bits,
                         weight=w.weight.to(torch.cuda.current_device()),
-                        bias=w.bias.to(torch.cuda.current_device()),
+                        bias=w.bias.to(torch.cuda.current_device()) if w.bias is not None else None,
                         empty_init=empty_init,
                         device=w.weight.device if device is None else device,
                         dtype=w.weight.dtype,
@@ -175,7 +175,7 @@ def quantize(model, bits, empty_init=False, device=None,**kwarg):
                     QuantizedLinear(
                         bits=bits,
                         weight=w.weight.to(torch.cuda.current_device()),
-                        bias=w.bias.to(torch.cuda.current_device()),
+                        bias=w.bias.to(torch.cuda.current_device()) if w.bias is not None else None,
                         empty_init=empty_init,
                         device=w.weight.device if device is None else device,
                         dtype=w.weight.dtype,