[Bugfix][Core] Log KV cache capacity after block-size resolution (#50462)
Signed-off-by: Tanmay Dixit <22117787+tandixit95@users.noreply.github.com> Signed-off-by: Michael Goin <mgoin64@gmail.com> Signed-off-by: Matthew Bonanni <mbonanni@redhat.com> Co-authored-by: Tanmay Dixit <22117787+tandixit95@users.noreply.github.com> Co-authored-by: Michael Goin <mgoin64@gmail.com> Co-authored-by: Matthew Bonanni <mbonanni@redhat.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
co-authored by
Tanmay Dixit
Michael Goin
Matthew Bonanni
Claude Opus 5
parent
5789897aa4
commit
5f2ee2fa8c
@@ -1853,6 +1853,23 @@ def get_kv_cache_capacity(
|
||||
return int(max_concurrency * max_model_len), max_concurrency
|
||||
|
||||
|
||||
def update_kv_cache_capacity(
|
||||
vllm_config: VllmConfig, kv_cache_config: KVCacheConfig
|
||||
) -> None:
|
||||
"""Store and log the resolved KV cache capacity."""
|
||||
num_tokens, max_concurrency = get_kv_cache_capacity(vllm_config, kv_cache_config)
|
||||
vllm_config.cache_config.kv_cache_size_tokens = num_tokens
|
||||
vllm_config.cache_config.kv_cache_max_concurrency = max_concurrency
|
||||
max_model_len = vllm_config.model_config.max_model_len
|
||||
logger.info_once(
|
||||
"GPU KV cache size: %s tokens, "
|
||||
"Maximum concurrency for %s tokens per request: %.2fx",
|
||||
f"{num_tokens:,}",
|
||||
f"{max_model_len:,}",
|
||||
max_concurrency,
|
||||
)
|
||||
|
||||
|
||||
def _max_memory_usage_bytes_from_groups(
|
||||
vllm_config: VllmConfig,
|
||||
kv_cache_groups: list[KVCacheGroupSpec],
|
||||
@@ -2188,23 +2205,6 @@ def get_kv_cache_configs(
|
||||
assert tensor.size % num_blocks_old == 0
|
||||
tensor.size = tensor.size // num_blocks_old * min_num_blocks
|
||||
|
||||
if len(kv_cache_config.kv_cache_groups) > 0:
|
||||
max_model_len = vllm_config.model_config.max_model_len
|
||||
# GPU KV cache size in tokens = max_concurrency * max_model_len:
|
||||
# the total tokens of context the pool can hold at peak
|
||||
# utilization. Sourcing this from the concurrency calculation
|
||||
# handles hybrid layouts correctly.
|
||||
num_tokens, max_concurrency = get_kv_cache_capacity(
|
||||
vllm_config, kv_cache_config
|
||||
)
|
||||
|
||||
logger.info_once("GPU KV cache size: %s tokens", f"{num_tokens:,}")
|
||||
logger.info_once(
|
||||
"Maximum concurrency for %s tokens per request: %.2fx",
|
||||
f"{max_model_len:,}",
|
||||
max_concurrency,
|
||||
)
|
||||
|
||||
return kv_cache_configs
|
||||
|
||||
|
||||
|
||||
@@ -46,11 +46,11 @@ from vllm.utils.system_utils import decorate_logs, set_process_title
|
||||
from vllm.v1.core.kv_cache_utils import (
|
||||
BlockHash,
|
||||
generate_scheduler_kv_cache_config,
|
||||
get_kv_cache_capacity,
|
||||
get_kv_cache_configs,
|
||||
get_request_block_hasher,
|
||||
init_none_hash,
|
||||
resolve_kv_cache_block_sizes,
|
||||
update_kv_cache_capacity,
|
||||
)
|
||||
from vllm.v1.core.sched.interface import PauseState, SchedulerInterface
|
||||
from vllm.v1.core.sched.output import SchedulerOutput
|
||||
@@ -319,11 +319,7 @@ class EngineCore:
|
||||
vllm_config.cache_config.block_size = min(
|
||||
g.kv_cache_spec.block_size for g in kv_cache_groups
|
||||
)
|
||||
num_tokens, max_concurrency = get_kv_cache_capacity(
|
||||
vllm_config, scheduler_kv_cache_config
|
||||
)
|
||||
vllm_config.cache_config.kv_cache_size_tokens = num_tokens
|
||||
vllm_config.cache_config.kv_cache_max_concurrency = max_concurrency
|
||||
update_kv_cache_capacity(vllm_config, scheduler_kv_cache_config)
|
||||
|
||||
vllm_config.validate_block_size()
|
||||
|
||||
|
||||
Reference in New Issue
Block a user