xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-09 20:04:27 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Model] fix glm4_moe_mtp load weights with GLM-4.6 checkpoint. (#27597)

Browse Source 
Signed-off-by: wuao.scotty <wuao.scotty@bytedance.com>
Co-authored-by: wuao.scotty <wuao.scotty@bytedance.com>
This commit is contained in:
wuyaoxuehun 2025-11-12 17:14:00 +07:00 committed by GitHub
parent 1761dea1a8
commit d3ade61e42
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 11 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
vllm/model_executor/models/glm4_moe_mtp.py
View File

@ -256,11 +256,18 @@ class Glm4MoeMTP(nn.Module, SupportsPP, Glm4MixtureOfExperts):
params_dict = dict(self.named_parameters())
loaded_params: set[str] = set()
spec_layer = self.model.mtp_start_layer_idx
for name, loaded_weight in weights:
spec_layer = get_spec_layer_idx_from_weight_name(self.config, name)
if spec_layer is None:
continue
name = self._rewrite_spec_layer_name(spec_layer, name)
if name == "lm_head.weight":
name = f"model.layers.{spec_layer}.shard_head.head.weight"
elif name == "model.embed_tokens.weight":
# This name is same with local model, rewriting is not needed.
pass
else:
spec_layer = get_spec_layer_idx_from_weight_name(self.config, name)
if spec_layer is None:
continue
name = self._rewrite_spec_layer_name(spec_layer, name)
for param_name, weight_name, shard_id in stacked_params_mapping:
# Skip non-stacked layers and experts (experts handled below).
if weight_name not in name: