Signed-off-by: NickLucche <nicolo.lucchesi@mistral.ai> Co-authored-by: Yifan Qiao <17067717+ivanium@users.noreply.github.com>
This commit is contained in:
co-authored by
Yifan Qiao
parent
153242a314
commit
9488318c6a
@@ -1051,15 +1051,9 @@ class MooncakeConnectorWorker:
|
||||
self.finished_recving_reqs: set[ReqId] = set()
|
||||
# Written from the receiver loop, drained from the worker thread.
|
||||
self._invalid_block_ids: queue.Queue[set[int]] = queue.Queue()
|
||||
self._is_hma_required = (
|
||||
not vllm_config.scheduler_config.disable_hybrid_kv_cache_manager
|
||||
and any(
|
||||
not isinstance(g.kv_cache_spec, FullAttentionSpec)
|
||||
for g in kv_cache_config.transfer_groups
|
||||
)
|
||||
)
|
||||
# Block IDs are only unique within a group; with HMA the scheduler
|
||||
# tracks a single merged group, so failures are reported per request.
|
||||
self._is_hma_required = len(kv_cache_config.kv_cache_groups) > 1
|
||||
# Block IDs are only unique within a group, so with multiple groups
|
||||
# load failures are reported per request instead.
|
||||
self._failed_recv_reqs: queue.Queue[ReqId] = queue.Queue()
|
||||
|
||||
self.xfer_stats = MooncakeKVConnectorStats()
|
||||
|
||||
@@ -1461,13 +1461,7 @@ class MooncakeStoreWorker:
|
||||
and not self.can_put
|
||||
)
|
||||
self.cache_config = vllm_config.cache_config
|
||||
self._is_hma_required = (
|
||||
not vllm_config.scheduler_config.disable_hybrid_kv_cache_manager
|
||||
and any(
|
||||
not isinstance(g.kv_cache_spec, FullAttentionSpec)
|
||||
for g in kv_cache_config.transfer_groups
|
||||
)
|
||||
)
|
||||
self._is_hma_required = len(kv_cache_config.kv_cache_groups) > 1
|
||||
self.block_size, self.hash_block_size = resolve_kv_cache_block_sizes(
|
||||
kv_cache_config, vllm_config
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user