yoyodyne
yoyodyne copied to clipboard
CUNY-CL
Small-vocabulary sequence-to-sequence generation with optional feature conditioning
[copied from CUNY-CL/abstractness/issues/52] The transducer uses random choice for tutoring. Alternatively, Boltzmann exploration could be used: https://arxiv.org/abs/1705.10257
We can just silence the warning by agreeing not to use the nested tensors backend. Closes #214. Mutually exclusive with #224.
(Lowest of the low priorities) SSMs have been making the rounds but people have only cared about them for 'major' tasks. (NMT models, speech, LLM). Since they're special LSTMs and...
LightningCLI supports YAML configs. However we're going to need to add some additional OmegaConf and Hydra support to our existing scripts to best utilize this. This issue will work on...
…and when updating inference.
For some reason, someone added an `embeddings` argument to `BaseModule` but didn't update the other modules to pass it. Why are we passing embeddings as an argument? Those should be...
This allows us to use the experimental, but supposedly faster, nested tensor API: https://pytorch.org/docs/stable/nested.html As the documentation indicates, this apparently is particularly helpful with padding. Closes #214. Mutually exclusive with...
The transformer logs this on creation: ```UserWarning: enable_nested_tensor is True, but self.use_nested_tensor is False because encoder_layer.norm_first was True``` There are two things we can do about it: 1. To silence...
#202 pinned numpy to < 2 because there was a conflict with PyTorch. This serves a reminder to remove this pin: most likely we can set numpy > 2.x.y (tbd)...
Current LSTM outputs create ugly typing as mentioned in https://github.com/CUNY-CL/yoyodyne/pull/195. Replace with a dataclass wrapper to make eyes not bleed.