Remove KV cache scale boilerplate from model weight loading methods (#43167)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2026-06-05 05:19:04 -07:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 7fe7800fa4
commit 62215e72c6
56 changed files with 87 additions and 730 deletions
@@ -1541,6 +1541,11 @@ def maybe_remap_kv_scale_name(name: str, params_dict: dict) -> str | None:
if no remapping is needed.
None: If the remapped name is not found in params_dict.
"""
# Already in vLLM's expected form (e.g. weights pre-renamed by a
# `WeightsMapper` from the quant config). Skip the regex remap, which
# would otherwise double-apply the `.attn` prefix and drop the weight.
if name in params_dict:
return name
if name.endswith(".kv_scale"):
logger.warning_once(
"DEPRECATED. Found kv_scale in the checkpoint. "