 Alexander Borzunov
|
c365b2ec9f
Tweak settings for the upcoming demo (#2)
|
3 éve |
 Aleksandr Borzunov
|
64dee420da
Upgrade to using hivemind.optim.experimental
|
3 éve |
|
Aleksandr Borzunov
|
e97e7b8811
Try removing OffloadOptimizer
|
3 éve |
|
Aleksandr Borzunov
|
259e6b9009
Set per_device_batch_sizes = 3
|
3 éve |
|
Aleksandr Borzunov
|
75e68ed8c0
Fix hivemind commit in requirements.txt
|
3 éve |
|
Aleksandr Borzunov
|
3666555ee5
Add learning-at-home repos to requirements.txt
|
3 éve |
|
Aleksandr Borzunov
|
3b184c57da
Don't download VQGAN weights
|
3 éve |
|
Aleksandr Borzunov
|
c61c61b20d
Use t5-small tokenizer
|
3 éve |
|
Aleksandr Borzunov
|
144d35ebce
Set share_input_output_emb=True
|
3 éve |
|
Aleksandr Borzunov
|
380a5f3231
Set num threads to <= 4
|
3 éve |
|
Aleksandr Borzunov
|
df54ab6da5
Log number of params
|
3 éve |
|
Aleksandr Borzunov
|
c5493b9baf
Update per_device_batch_sizes and shuffle_buffer_size
|
3 éve |
|
Aleksandr Borzunov
|
17da8639c2
Fix and improve preprocessing
|
3 éve |
|
Aleksandr Borzunov
|
d0985de540
Enable rotary embeddings
|
3 éve |
|
Aleksandr Borzunov
|
f621362466
Make logging less verbose
|
3 éve |
|
Aleksandr Borzunov
|
31215c85d3
Handle None captions
|
3 éve |
|
Aleksandr Borzunov
|
1c30209eba
Increase timeouts
|
3 éve |
|
Aleksandr Borzunov
|
20e2a3aab2
Use dalle-pytorch instead of LeanAlbert
|
3 éve |
 Max Ryabinin
|
72fc0bcdb7
Initial commit (ru-max branch without private code)
|
4 éve |
