Fixed import
4a7707c
PeteBleackley
commited on
Spaces:
Build error
Build error
Commit History
Fixed import
e8324a1
PeteBleackley
commited on
Factorized the weight matrix in the GlobalAttentionPoolingHead, thus reducing the number of parameters in this layer by a factor of 48
a1e9f64
PeteBleackley
commited on
Using torch.nn.CosineSimilarity to simplify code
798488e
PeteBleackley
commited on
Fix Einstein summation notation
7fe1144
PeteBleackley
commited on
Use keepdim option when normalising vectors
738b546
PeteBleackley
commited on
Make EPSILON a tensor
cf5f935
PeteBleackley
commited on
torch.maximum, not torch.max
98ad67d
PeteBleackley
commited on
Unsqueeze attention mask
bc77ce5
PeteBleackley
commited on
Converted GlobalAttentionPoolingHead to use PyTorch
32df2f1
PeteBleackley
commited on
Trainable => trainable
a8c528d
PeteBleackley
commited on
Ensure weights are trainable
e556cb6
PeteBleackley
commited on
Making sure RoBERTa layers have all required arguments
b2593fa
PeteBleackley
commited on
Final dot product
8ec9bd9
PeteBleackley
commited on
dot_prod needs to unpack arguments from tuple
8d80339
PeteBleackley
commited on
Removed unnecessary complication
c284c9a
PeteBleackley
commited on
Only inner function needs decorator
210f1cb
PeteBleackley
commited on
More vectorized_map weirdness
eecf608
PeteBleackley
commited on
tensorflow.vectorized_map might not like getting function arguments in a tuple
3f78694
PeteBleackley
commited on
Broadcasting dot products
f2bd224
PeteBleackley
commited on
Broadcasting dot products
825e41b
PeteBleackley
commited on
Broadcasting dot products
948988c
PeteBleackley
commited on
Error in invocation of tensordot
1670b0e
PeteBleackley
commited on
More work on models
f16a715
PeteBleackley
commited on
GlobalAttentionPoolingHead layer
8f1745b
PeteBleackley
commited on