Rewards model
Reward Modelling is an essential part of making high-quality models.
Now, you will have to install the chaiverse package to submit your first model
import chaiverse as chai
from chaiverse.formatters import PromptFormatter
How it works
Rewards Model ExplainedGreat You're Ready to Submit your first rewards model
model_url = "ChaiML/phase2_winner_13b2"
reward_url = "ChaiML/reward_models_100_170000000_cp_498032"
submission_parameters = {
"model_repo": model_url,
"reward_repo": reward_url,
"generation_params": {
"temperature": 0.99,
"top_p": 0.2,
"top_k": 40,
"stopping_words": ['\n'],
"presence_penalty": 0.,
"frequency_penalty": 0.,
"max_input_tokens": 1024,
"best_of": 4
},
'model_name': 'My-First-RwardModel',
}
Simply submit your model using the chai package
submitter = chai.ModelSubmitter(verbose=True)
submission_id = submitter.submit(submission_parameters)
Last updated