expands run_features() signature to include the model (in order to extract its tokenizer)
adds if case for reward_num_tokens feature to calculate the number of tokens for each given phrase
adds hardcoded tokenizer to common BaseRewardModel to ensure that the mock models have a tokenizer
adds subtensor and metagraph config to dummy_config.yml in order to not break this config execution
adds reward_num_tokens feature to be created and plotted (using the dummy_config.yml)
Note: The reward and text column being used are the same as the other projections (e.g. _questionlength, _avg_wordlength), meaning that the reward it's the first value of the reward tensor and the high level feature column being derived is the column message
run_features()
signature to include the model (in order to extract its tokenizer)reward_num_tokens
feature to calculate the number of tokens for each given phrasedummy_config.yml
in order to not break this config executionreward_num_tokens
feature to be created and plotted (using thedummy_config.yml
)Note: The reward and text column being used are the same as the other projections (e.g. _questionlength, _avg_wordlength), meaning that the reward it's the first value of the reward tensor and the high level feature column being derived is the column
message
Closes #21