Cheng Li 9caa74e577 Autotuning (#1554) | 2 years ago | |
---|---|---|
.. | ||
README.md | 9caa74e577 Autotuning (#1554) | 2 years ago |
__init__.py | 9caa74e577 Autotuning (#1554) | 2 years ago |
base_tuner.py | 9caa74e577 Autotuning (#1554) | 2 years ago |
cost_model.py | 9caa74e577 Autotuning (#1554) | 2 years ago |
index_based_tuner.py | 9caa74e577 Autotuning (#1554) | 2 years ago |
model_based_tuner.py | 9caa74e577 Autotuning (#1554) | 2 years ago |
utils.py | 9caa74e577 Autotuning (#1554) | 2 years ago |
exps
is a list of experiment descriptions (dictionarys).
An experimentation description has a ds_config
field that stores the DeepSpeed configuration to be used in the experiment.
A tuner is based on BaseTuner and at least implements the next_batch
method. It can implement a different tune
method from the BaseTuner's.
class NewTuner(BaseTuner):
def __init__(self, exps: list, resource_manager):
super(NewTuner, self).__init__(exps, resource_manager)
def next_batch(self, sample_size=1):
pass
def tune(self): # if it differs from BaseTuner
pass