Skip to content

fix: consider kv cache 64 align to estimate kv cache memory size #417

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

Open
rebel-wonsubkim wants to merge 1 commit into
base: dev
Choose a base branch
from
Open

fix: consider kv cache 64 align to estimate kv cache memory size #417

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
10 changes: 9 additions & 1 deletion vllm_rbln/v1/worker/rbln_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@
"""A RBLN worker class."""

import copy
import math
import os
from types import NoneType
from typing import TYPE_CHECKING
Expand Down Expand Up @@ -279,6 +280,7 @@ def determine_available_memory(self) -> int:
# NOTE - model parallel(tp, dp, ep, pp)
# already applied into model params
n_model_params = n_model_attentions + n_model_experts
head_size = self.model_config.get_head_size()

available_memory_estimate = estimate_available_memory(
model_config=self.model_config,
Expand All @@ -293,7 +295,13 @@ def determine_available_memory(self) -> int:
logger.info(
"available_memory_estimate = %.2f GB", available_memory_estimate / 10**9
)

head_align_ratio = math.ceil(head_size / 64) * 64 / head_size
logger.info("head size align ratio = %s", head_align_ratio)
available_memory_estimate /= head_align_ratio
logger.info(
"available_memory_estimate considering 64B align = %.2f GB",
available_memory_estimate / 10**9
)
return available_memory_estimate

def get_kv_cache_spec(self) -> dict[str, KVCacheSpec]:
Expand Down
Loading