Fix Llama4 attention flops #2030
Merged
+51
−1
Conversation
This file contains hidden or 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
gagika
requested review from
gobbleturk,
khatwanimohit,
bvandermoon,
vipannalla,
RissyRan,
richjames0,
shralex,
yangyuwei,
SurbhiJainUSC,
hengtaoguo,
A9isha and
aireenmei
as code owners
gobbleturk
reviewed
Jul 28, 2025
| # FLOPs for a single global attention layer (full attention, non-causal) | ||
| global_attention_flops_per_layer = 4 * config.per_device_batch_size * seq_len**2 * config.num_query_heads * config.head_dim | ||
|
|
||
| # FLOPs for a single chunked attention layer (non-causal) |
There was a problem hiding this comment.
Can we separate out chunked attention flops into its own method?
This was referenced Aug 1, 2025
Open
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Outdated suggestions cannot be applied.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.
Description
Fix Llama4 attention flops to only count for chunk attention window size for chunk attention layers.
Tests
Comparing
total_tflops, learnable_weight_tflops, attention_tflopsbefore and after the fix:https://diff.googleplex.com/#key=wo0bVmy9wbUc
learnable_weight_tflops are matching
attention_tflops are matching for context length <= 8192
For context length 2* 8192, old attention flops was increasing 4 x over 8192, now
it's increasing 10/4 x (3 chunk attention increasing 2x and 1 global attention increasing 4x).
1187* 10/4 = 2968
Checklist
Before submitting this PR, please make sure (put X in square brackets):