Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Added

Add max_len option in fit method #21
Set default model path as a command line argument for try CLI #24

Fixed

Add max_len option in Dataset to restrict the max length of a tensor inputting to a model #20
Revise code structure #22
Fix two issues in #23
- Responses include EOS at the end of the utterance.
- Tokenizer adds special tokens if the tokenizer adds them as default.

[v0.3.0]

Added

output_path and save_best_model parameters to fit method. #13
- output_path: when given, model is saved while training in the fit method. Default is None (will not be saved)
- save_best_model: when given with output_path, fit will only save the best model in output_path
min_new_tokens option to generate method. #16
eval method in model and cli. #14
print_config option to fit cli command. #18

Changed

streamlit UI by using side-by-side columns to show conversation. #15
Change CLI command name from run_streamlit to try. #17

[v0.2.0]

Added

fit method in ConversatonModel. #8
fit CLI #10

Changed

device transfer code in fit method. #9
document to use MkDocs. #11

Removed

Obsoleted modules for bert and gpt2_lm. https://github.com/colorfulscoop/convmodel/pull/7/files

[v0.1.1] - 2021-08-28

Added

build_data_loader method in ConversationDataset class. #5

[v0.1.0] - 2021-08-28

Added

ConversationModel and its usage.