Closed flxst closed 3 months ago
The getting started example has now been updated, adjusted to all the recent changes (CLI commands, config files etc.) and it's working again. However, the parameters in the training config file still need to be improved - they are currently not equivalent to those in the old training config file, and the text generated by the example model is very poor.
Parameters in the training and text generation config files have been improved. Text generated by the example model seems to be according to expectations.
The getting started example is updated as it does not reflect some recent changes in
modalities
. In particular, this PR aims to:examples/getting_started/README.md
a general updatetokenizer.json
) by a tokenizer instance (e.g. the variantPreTrainedHFTokenizer
)modalities data pack_encoded_data
examples/getting_started/example_config.yaml
modalities generate_text [..]
to work again (based on #102)EDIT: Note that the above changes only make the getting started example work again. Suggestions for additional changes to improve user friendliness can be found in #117.