[Feature] Beam search support (#31626)

Co-authored-by: cswuyg <cswuyg@gmail.com>
Co-authored-by: cswuyg <496090217@qq.com>
Co-authored-by: Vedant Jhaveri <vedantjh2@gmail.com>
Co-authored-by: Vedant Jhaveri <vjhaveri@linkedin.com>
This commit is contained in:
Liangsheng Yin
2026-08-26 16:56:15 -07:00
committed by GitHub
co-authored by cswuyg cswuyg Vedant Jhaveri Vedant Jhaveri
parent e5a1c5a423
commit ec4bdbfa4a
39 changed files with 3066 additions and 33 deletions
@@ -91,6 +91,19 @@ _OWNER_SITES = {
"alloc_for_decode_prealloc_hisparse",
"kv_allocated_len",
): 1,
# Beam member rows alias the leader's decode region, so releasing them
# rewinds the leader's watermarks to keep its own per-Req release from
# freeing that region twice.
(
"beam_search/fork.py",
"free_member_rows",
"kv_committed_len",
): 1,
(
"beam_search/fork.py",
"free_member_rows",
"kv_allocated_len",
): 1,
# streaming session slot save/restore and tail trimming
(_SS, "SessionSlot.save_from_req", "kv_committed_len"): 1,
(_SS, "SessionSlot.restore_to_req", "kv_committed_len"): 1,