Use minimal softmax mask values based on the datatype #966
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
… related to a hardcoded value withing the conversion script (3) fixed possible bugs in the conversion script wrt the mp sharding convention
megatron/model/gpt2_model.py
Outdated
| mask_value = torch.finfo(attention_scores.dtype).min | ||
| # Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`. | ||
| # Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device` | ||
| mask_value = torch.tensor(mask_value, dtype=attention_scores.dtype).to(attention_scores.device) |
There was a problem hiding this comment.
It's always best to initialize tensors directly on the target device. Using .to necessitates a copy first on the torch default device ('cpu' unless explicitly set).
Change to:
mask_value = torch.tensor(mask_value, dtype=attention_scores.dtype, device=attention_scores.device)
There was a problem hiding this comment.
Sounds good, fixed below
Quentin-Anthony
approved these changes
Jun 5, 2023
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The current GPT-NeoX implementation uses a hardcoded value of -10000.0 mask values in its softmax function. While this works fine most of the time, it can become problematic if unnormalized softmax values computed by the model become very small. Therefore, it is always best practice to take the smallest possible (based on the datatype) numerical value for softmax masking. This change does exactly this, following the implementation at https://github.com/huggingface/transformers/blob/539e2281cd97c35ef4122757f26c88f44115fa94/src/transformers/models/gpt_neox/modeling_gpt_neox.py#L226