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
-16
View File
@@ -555,22 +555,6 @@ class MiMoV2Model(nn.Module):
if "mtp" in name:
continue
if self.quant_config is not None:
cache_scale_name = self.quant_config.get_cache_scale(name)
if cache_scale_name is not None and cache_scale_name in params_dict:
param = params_dict[cache_scale_name]
weight_loader = getattr(
param, "weight_loader", default_weight_loader
)
kv_scale = loaded_weight
if kv_scale.dim() > 0 and kv_scale.numel() > 1:
kv_scale = kv_scale.view(-1)[0]
weight_loader(param, kv_scale)
loaded_params.add(cache_scale_name)
continue
expert_matched = False
for param_name, weight_name, expert_id, shard_id in expert_params_mapping:
if weight_name not in name: