Upload convert_to_linear.py with huggingface_hub
Browse files- convert_to_linear.py +4 -0
convert_to_linear.py
CHANGED
@@ -94,3 +94,7 @@ class NewGptOssExperts(nn.Module):
|
|
94 |
mixed = (outs * rw).sum(dim=0)
|
95 |
return mixed.view(batch_size, -1, self.hidden_size)
|
96 |
|
|
|
|
|
|
|
|
|
|
94 |
mixed = (outs * rw).sum(dim=0)
|
95 |
return mixed.view(batch_size, -1, self.hidden_size)
|
96 |
|
97 |
+
|
98 |
+
# monkey patch to linear
|
99 |
+
from transformers.models.gpt_oss import modeling_gpt_oss
|
100 |
+
modeling_gpt_oss.GptOssExperts = NewGptOssExperts
|