adds softmax_scale to flash attention #209
Merged
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
vchiley
approved these changes
Mar 2, 2023
bcui19
approved these changes
Mar 2, 2023
bmosaicml
reviewed
Mar 2, 2023
| self.d_model = cfg.d_model | ||
| self.n_heads = cfg.n_heads | ||
|
|
||
| if self.attn_qk_ln or self.clip_qkv: | ||
| if self.attn_qk_ln or self.clip_qkv or self.softmax_scale: | ||
| self.W_qkv = nn.Linear(self.d_model, | ||
| 3 * self.d_model, | ||
| bias=True, | ||
| device=device) | ||
| self.inner_attn = FlashAttention(attention_dropout=cfg.attn_pdrop, |
There was a problem hiding this comment.
I see we can now scale the attention by any custom value...what is the best value and how should we keep track of it?
There was a problem hiding this comment.
1/srqt(d/n_heads) is standard.
1/(d/n_heads) is the recommended muP mod.
Since its part of the model config, it'll get dumped into wandb if we ever need to check what we used.
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Outdated suggestions cannot be applied.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.
extends the option to scale softmax to flash attention in addition to triton flash attention