Skip to content

[SW-242794] Fix not warmed up decode buckets #484

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
jbyczkow wants to merge 1 commit into
base: v0.10.2_next
Choose a base branch
from
Draft

[SW-242794] Fix not warmed up decode buckets #484

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 9 additions & 3 deletions vllm_gaudi/extension/bucketing/common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -271,12 +271,15 @@ def ctx_not_over_max_ctx_for_merged_prefill(bs, query, ctx):
return ctx <= max_num_prefill_seqs * math.ceil(
(max_model_len - math.floor(query / max_num_prefill_seqs)) // block_size)

def aligned_max_model_len(bs):
return bs * math.ceil(max_model_len / block_size)

# decode
def block_not_greater_than_max_model_len(bs, query, ctx):
if not ctx <= bs * math.ceil(max_model_len / block_size):
if not ctx <= aligned_max_model_len(bs):
omitted_buckets.add(("condition: ctx <= bs * math.ceil(max_model_len / block_size)", "-> bs, quesry, ctx: ",
bs, query, ctx))
return ctx <= bs * math.ceil(max_model_len / block_size)
return ctx <= aligned_max_model_len(bs)

def batch_size_smaller_than_blocks(bs, query, ctx):
if not bs <= ctx:
Expand Down Expand Up @@ -313,7 +316,10 @@ def get_filters(is_prompt, use_merged_prefill, use_contiguous_pa):
expand_to_neighbor_buckets(bs_idx, bs_range, query_idx, query_range, max_num_batched_tokens))

for bs, query in buckets_2d:
for ctx in ctx_range:
add_ctx_range = []
if not use_contiguous_pa and aligned_max_model_len(bs) not in ctx_range:
add_ctx_range.append(aligned_max_model_len(bs))
for ctx in ctx_range + add_ctx_range:
if all(bucket_filter(bs, query, ctx) for bucket_filter in filters):
buckets.add((bs, query, ctx))
if not buckets:
Expand Down
Loading