huggingface/transformers

Pull Requests Commits

thomasw21 committed 3 years ago

53e8738c

thomasw21 committed 3 years ago

784073a1

Maybe this needs to be duplicated

thomasw21 committed 3 years ago

13b9ce18

Fix cat dimension

thomasw21 committed 3 years ago

c127d24c

Fix in-place operation

thomasw21 committed 3 years ago

0e1a758e

thomasw21 committed 3 years ago

34f7e342

Reduce the number of concatenation

thomasw21 committed 3 years ago

5167346a

Revert "Add back layer_number normalization"

thomasw21 committed 3 years ago

8f4d6035

Add back layer_number normalization

thomasw21 committed 3 years ago

50e1f2f9

Add comment about support for torchScript v1.11

thomasw21 committed 3 years ago

7c399e6a

Apply suggestions from code review

thomasw21 committed 3 years ago

Verified bd1ae60d

Bypass torch.baddbmm

thomasw21 committed 3 years ago

fd117da1

Try passing a tuple

thomasw21 committed 3 years ago

b58a6f72

Does this help?

thomasw21 committed 3 years ago

790e2382

Revert "Fix FX test"

thomasw21 committed 3 years ago

1e4ccaf3

thomasw21 committed 3 years ago

a3d50c08

Merge remote-tracking branch 'origin/main' into thomas/bloom_clean_code

thomasw21 committed 3 years ago

87861b2e

[FX] Symbolic trace for Bloom (#18356)

michaelbenayoun committed 3 years ago

Verified 4e2f4a92

We don't actually need layer_number to be passed anymore

thomasw21 committed 3 years ago

88814bf4

Actually i think the attention casting only makes sense when we use torch.float16

thomasw21 committed 3 years ago

49aff18f

thomasw21 committed 3 years ago

5fcc118a

thomasw21 committed 3 years ago

1c93638e

Add comment that explains that some methods return views

thomasw21 committed 3 years ago

98fdf998

Improve type hinting

thomasw21 committed 3 years ago

8699509c

Rename with long version of variables

thomasw21 committed 3 years ago

2d58b7e3

Fix some doctests (#18359)

ydshieh committed 3 years ago

Verified 1763770b

Replace `as_target` context managers by direct calls (#18325)

sgugger committed 3 years ago

Verified 986526a0

No need to manually cast the values to a specific device

thomasw21 committed 3 years ago

323c0731

Optimize the device allocation in case of hidden_states in multiple devices

thomasw21 committed 3 years ago

4db82d46

thomasw21 committed 3 years ago

b4346e14

Older