MLA layer eliminates redundant index operators #993
+21
−13
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
MengqingCao
reviewed
May 29, 2025
vllm_ascend/attention/attention.py
Outdated
|
|
||
| q_pe = self.rope_single(q_pe, cos, sin) | ||
| k_pe, k_nope = self.exec_kv(hidden_states_or_kv_c_normed, cos, sin, | ||
| if self.layer_idx == 0 or self.cos is None or self.sin is None: |
There was a problem hiding this comment.
Could you make more comments on why updating self.cos and self.sin only when layer_idx == 0?
There was a problem hiding this comment.
During the autoregressive decoding process, the cos and sin values are exactly the same for each layer(such as 61 layers). Therefore, they only need to be calculated in the first layer, and subsequent layers can directly reuse them.
There was a problem hiding this comment.
Thanks for the explaination, let's add this comments into code
vllm_ascend/models/deepseek_v2.py
Outdated
| @@ -392,6 +392,7 @@ def __init__( | |||
| kv_a_layernorm=self.kv_a_layernorm, | |||
| kv_b_proj=self.kv_b_proj, | |||
| o_proj=self.o_proj, | |||
| ascend_prefix=prefix, | |||
There was a problem hiding this comment.
I recommand to pass by debug_layer_idx instead of ascend_prefix
huiyingCCCC
force-pushed
the
main
branch
3 times, most recently
from
742a3ee to
8cd3c81
Compare
huiyingCCCC
changed the title
Signed-off-by: huiying <chenhuiying4@huawei.com>
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Outdated suggestions cannot be applied.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.
What this PR does / why we need it?
During the autoregressive decoding process, the cos and sin values are exactly the same for each layer(such as 61 layers). Therefore, they only need to be calculated in the first layer, and subsequent layers can directly reuse them.
Does this PR introduce any user-facing change?
How was this patch tested?