diff options
Diffstat (limited to 'README.md')
| -rw-r--r-- | README.md | 34 |
1 files changed, 33 insertions, 1 deletions
@@ -28,8 +28,40 @@ make download make generate ``` +## Train + + +Use, modify, or create a new experiment found at `training/conf/experiment/`. +To run an experiment we first need to enter the virtual env by running: + +```sh +poetry shell +``` + +Then we can train a new model by running: + +```sh +python main.py +experiment=conv_transformer_paragraphs +``` + +## Network + +Create a picture of the network and place it here + +## Graveyard + +Ideas of mine that did not work unfortunately: + +* Use VQVAE to create pre-train a good latent representation + - Tests with various compressions did not show any performance increase compared to training directly e2e, more like decrease to be honest + - This is very unfortunate as I really hoped that this idea would work :( + - I still really like this idea, and I might not have given up just yet... + + +* Axial Transformer Encoder + - Added a lot of extra parameters with no gain in performance + - Cool idea, but on a single GPU, nah... not worth it! -## TODO ## Todo - [ ] remove einops |