Selective merged prefill #643
Open
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
xuechendi
requested review from
kzawora-intel,
madamczykhabana,
michalkuligowski and
mgawarkiewicz
as code owners
xuechendi
force-pushed
the
mlperf_features
branch
from
da935cf to
d6bdc90
Compare
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
xuechendi
force-pushed
the
selective_merged_prefill
branch
from
6ddfcac to
f6c0c84
Compare
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi.Xue <[email protected]> Signed-off-by: Chendi Xue <[email protected]>
xuechendi
force-pushed
the
selective_merged_prefill
branch
from
b6f6961 to
2d6ceb9
Compare
yangw1234
reviewed
Jan 6, 2025
vllm/attention/backends/hpu_attn.py
Outdated
| key_cache, value_cache = HPUPagedAttention.split_kv_cache( | ||
| kv_cache, self.num_kv_heads, self.head_size) | ||
|
|
||
| key_cache = self.k_cache(padded_key_tensor, key_cache, block_indices, |
There was a problem hiding this comment.
It seems that when decoding, padded_key_tensor is not defined. Would this be a problem?
There was a problem hiding this comment.
I think you're right, but it didn't trigger any error, I'll look into it.
There was a problem hiding this comment.
@yangw1234 , after checking with the codes, since enable_merged_prefill is only enabled in prefill_fwd, I'll clean up the codes to make it more readable
yangw1234
reviewed
Jan 6, 2025
vllm/attention/backends/hpu_attn.py
Outdated
Comment on lines
279
to
286
| max_len=attn_metadata.slot_mapping.size(1) | ||
| seq_lens_tensor_list = attn_metadata.seq_lens_tensor.tolist() | ||
| # we need to copy the key and value tensors to the padded tensors | ||
| # shape is [bacth_size, entire_seq_len, num_kv_heads, head_size] | ||
| padded_key_tensor = split_and_pad_to_length(key, max_len, seq_lens_tensor_list) | ||
| padded_value_tensor = split_and_pad_to_length(value, max_len, seq_lens_tensor_list) | ||
| padded_key_tensor = padded_key_tensor.flatten(0, 1).unflatten(0, (block_indices.size(0), -1)) | ||
| padded_value_tensor = padded_value_tensor.flatten(0, 1).unflatten(0, (block_indices.size(0), -1)) |
There was a problem hiding this comment.
Is it possible to get rid of these line if we prepare block_indices and block_offsets in a way that excludes the padded tokens?
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi.Xue <[email protected]>
xuechendi
force-pushed
the
selective_merged_prefill
branch
from
d18df2d to
92bf903
Compare
Signed-off-by: Chendi Xue <[email protected]> Signed-off-by: Chendi.Xue <[email protected]>
xuechendi
force-pushed
the
selective_merged_prefill
branch
2 times, most recently
from
b7d0931 to
ce48860
Compare
Signed-off-by: Chendi.Xue <[email protected]>
xuechendi
force-pushed
the
selective_merged_prefill
branch
from
ce48860 to
a3602f2
Compare
Signed-off-by: Chendi.Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]>
Signed-off-by: Chendi Xue <[email protected]> Signed-off-by: Chendi.Xue <[email protected]>
Signed-off-by: Chendi.Xue <[email protected]>
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.
No description provided.