Support for nanotron #11
Conversation
thomwolf
changed the title
| dataset = Dataset.from_list( | ||
| [{k: str(v) for k, v in asdict(detail).items()} for detail in task_details] | ||
| ) | ||
| # try: |
There was a problem hiding this comment.
If you remove the high level try catch, please add other try catches to prevent the other possible failures
There was a problem hiding this comment.
are we sure we want to silently catch mistake or should we not rather let the run fail?
There was a problem hiding this comment.
No because we still want the results to be saved locally. That way we can upload them by hand instead of having to redo the whole eval.
|
|
||
|
|
||
| class GenerativeTaskDatasetNanotron(DynamicBatchDataset): | ||
| def __getitem__(self, index) -> Request: |
There was a problem hiding this comment.
Why do you need your own class? (Is it only to return the index with the item?)
There was a problem hiding this comment.
base_model does not use the index for each sample, that means that we need to accommodate the dataset to nanotron.
There was a problem hiding this comment.
Yes but I'm unsure why we need to grab the index for brr
| config_cls: Type = Config, | ||
| model_config_cls: Optional[Type] = None, | ||
| model_cls: Optional[Type] = None, | ||
| ): |
Support for Nanotron models