[mem_cache] Settle extend kv_committed_len inside alloc_for_extend (#37085)
This commit is contained in:
@@ -53,14 +53,14 @@ _SS = "session/streaming_session.py"
|
||||
_OWNER_SITES = {
|
||||
# non-spec scheduler
|
||||
(_SB, "ScheduleBatch.prepare_for_decode", "decode_batch_idx"): 1,
|
||||
(_SB, "ScheduleBatch.prepare_for_decode", "kv_committed_len"): 1,
|
||||
(_SB, "ScheduleBatch.prepare_for_extend", "extend_batch_idx"): 1,
|
||||
(_SB, "ScheduleBatch.prepare_for_extend", "kv_committed_len"): 1,
|
||||
# kv_allocated_len is settled inside the owned-kv alloc functions (op28).
|
||||
("mem_cache/allocation.py", "alloc_for_extend", "evict"): 1,
|
||||
("mem_cache/allocation.py", "alloc_for_extend", "kv_allocated_len"): 1,
|
||||
("mem_cache/allocation.py", "alloc_for_extend", "kv_committed_len"): 1,
|
||||
("mem_cache/allocation.py", "alloc_for_decode", "evict"): 1,
|
||||
("mem_cache/allocation.py", "alloc_for_decode", "kv_allocated_len"): 1,
|
||||
("mem_cache/allocation.py", "alloc_for_decode", "kv_committed_len"): 1,
|
||||
# spec v2: no pre-claim; resolve commits the full accepted run uniformly.
|
||||
# kv_allocated_len for spec v2 draft decode (eagle + dflash) is settled
|
||||
# inside the owned-kv alloc_for_spec_decode function (op42).
|
||||
|
||||
Reference in New Issue
Block a user