Layer mismatch

#3
by cpatonn - opened

Hi, thank you for sharing the model. I am trying to load the model using transformers and I see that there is a mismatch in the number of layers.

Some weights of the model checkpoint at zai-org/GLM-4.5-Air were not used when initializing Glm4MoeForCausalLM: ['model.layers.46.eh_proj.weight', 'model.layers.46.embed_tokens.weight', 'model.layers.46.enorm.weight', 'model.layers.46.hnorm.weight', 'model.layers.46.input_layernorm.weight', 'model.layers.46.mlp.experts.0.down_proj.weight', 'model.layers.46.mlp.experts.0.gate_proj.weight', 'model.layers.46.mlp.experts.0.up_
proj.weight', 'model.layers.46.mlp.experts.1.down_proj.weight', 'model.layers.46.mlp.experts.1.gate_proj.weight', 'model.layers.46.mlp.experts.1.up_proj.weight', 'model.layers.46.mlp.experts.10.down_proj.weight', 'model.layers.46.mlp.experts.10.gate_proj.weight', 'model.layers.46.mlp.experts.10.up_proj.weight', 'model.layers.46.mlp.experts.100.down_proj.weight', 'model.layers.46.mlp.experts.100.gate_proj.weight', 'model.la
yers.46.mlp.experts.100.up_proj.weight', 'model.layers.46.mlp.experts.101.down_proj.weight', 'model.layers.46.mlp.experts.101.gate_proj.weight', 'model.layers.46.mlp.experts.101.up_proj.weight', 'model.layers.46.mlp.experts.102.down_proj.weight', 'model.layers.46.mlp.experts.102.gate_proj.weight', 'model.layers.46.mlp.experts.102.up_proj.weight', 'model.layers.46.mlp.experts.103.down_proj.weight', 'model.layers.46.mlp.expe
rts.103.gate_proj.weight', 'model.layers.46.mlp.experts.103.up_proj.weight', 'model.layers.46.mlp.experts.104.down_proj.weight', 'model.layers.46.mlp.experts.104.gate_proj.weight', 'model.layers.46.mlp.experts.104.up_proj.weight', 'model.layers.46.mlp.experts.105.down_proj.weight', 'model.layers.46.mlp.experts.105.gate_proj.weight', 'model.layers.46.mlp.experts.105.up_proj.weight', 'model.layers.46.mlp.experts.106.down_pro
j.weight', 'model.layers.46.mlp.experts.106.gate_proj.weight', 'model.layers.46.mlp.experts.106.up_proj.weight', 'model.layers.46.mlp.experts.107.down_proj.weight', 'model.layers.46.mlp.experts.107.gate_proj.weight', 'model.layers.46.mlp.experts.107.up_proj.weight', 'model.layers.46.mlp.experts.108.down_proj.weight', 'model.layers.46.mlp.experts.108.gate_proj.weight', 'model.layers.46.mlp.experts.108.up_proj.weight', 'mode
l.layers.46.mlp.experts.109.down_proj.weight', 'model.layers.46.mlp.experts.109.gate_proj.weight', 'model.layers.46.mlp.experts.109.up_proj.weight', 'model.layers.46.mlp.experts.11.down_proj.weight', 'model.layers.46.mlp.experts.11.gate_proj.weight', 'model.layers.46.mlp.experts.11.up_proj.weight', 'model.layers.46.mlp.experts.110.down_proj.weight', 'model.layers.46.mlp.experts.110.gate_proj.weight', 'model.layers.46.mlp.e
xperts.110.up_proj.weight', 'model.layers.46.mlp.experts.111.down_proj.weight', 'model.layers.46.mlp.experts.111.gate_proj.weight', 'model.layers.46.mlp.experts.111.up_proj.weight', 'model.layers.46.mlp.experts.112.down_proj.weight', 'model.layers.46.mlp.experts.112.gate_proj.weight', 'model.layers.46.mlp.experts.112.up_proj.weight', 'model.layers.46.mlp.experts.113.down_proj.weight', 'model.layers.46.mlp.experts.113.gate_
proj.weight', 'model.layers.46.mlp.experts.113.up_proj.weight', 'model.layers.46.mlp.experts.114.down_proj.weight', 'model.layers.46.mlp.experts.114.gate_proj.weight', 'model.layers.46.mlp.experts.114.up_proj.weight', 'model.layers.46.mlp.experts.115.down_proj.weight', 'model.layers.46.mlp.experts.115.gate_proj.weight', 'model.layers.46.mlp.experts.115.up_proj.weight', 'model.layers.46.mlp.experts.116.down_proj.weight', 'm
odel.layers.46.mlp.experts.116.gate_proj.weight', 'model.layers.46.mlp.experts.116.up_proj.weight', 'model.layers.46.mlp.experts.117.down_proj.weight', 'model.layers.46.mlp.experts.117.gate_proj.weight', 'model.layers.46.mlp.experts.117.up_proj.weight', 'model.layers.46.mlp.experts.118.down_proj.weight', 'model.layers.46.mlp.experts.118.gate_proj.weight', 'model.layers.46.mlp.experts.118.up_proj.weight', 'model.layers.46.m
lp.experts.119.down_proj.weight', 'model.layers.46.mlp.experts.119.gate_proj.weight', 'model.layers.46.mlp.experts.119.up_proj.weight', 'model.layers.46.mlp.experts.12.down_proj.weight', 'model.layers.46.mlp.experts.12.gate_proj.weight', 'model.layers.46.mlp.experts.12.up_proj.weight', 'model.layers.46.mlp.experts.120.down_proj.weight', 'model.layers.46.mlp.experts.120.gate_proj.weight', 'model.layers.46.mlp.experts.120.up
_proj.weight', 'model.layers.46.mlp.experts.121.down_proj.weight', 'model.layers.46.mlp.experts.121.gate_proj.weight', 'model.layers.46.mlp.experts.121.up_proj.weight', 'model.layers.46.mlp.experts.122.down_proj.weight', 'model.layers.46.mlp.experts.122.gate_proj.weight', 'model.layers.46.mlp.experts.122.up_proj.weight', 'model.layers.46.mlp.experts.123.down_proj.weight', 'model.layers.46.mlp.experts.123.gate_proj.weight',
 'model.layers.46.mlp.experts.123.up_proj.weight', 'model.layers.46.mlp.experts.124.down_proj.weight', 'model.layers.46.mlp.experts.124.gate_proj.weight', 'model.layers.46.mlp.experts.124.up_proj.weight', 'model.layers.46.mlp.experts.125.down_proj.weight', 'model.layers.46.mlp.experts.125.gate_proj.weight', 'model.layers.46.mlp.experts.125.up_proj.weight', 'model.layers.46.mlp.experts.126.down_proj.weight', 'model.layers.4
6.mlp.experts.126.gate_proj.weight', 'model.layers.46.mlp.experts.126.up_proj.weight', 'model.layers.46.mlp.experts.127.down_proj.weight', 'model.layers.46.mlp.experts.127.gate_proj.weight', 'model.layers.46.mlp.experts.127.up_proj.weight', 'model.layers.46.mlp.experts.13.down_proj.weight', 'model.layers.46.mlp.experts.13.gate_proj.weight', 'model.layers.46.mlp.experts.13.up_proj.weight', 'model.layers.46.mlp.experts.14.do
wn_proj.weight', 'model.layers.46.mlp.experts.14.gate_proj.weight', 'model.layers.46.mlp.experts.14.up_proj.weight', 'model.layers.46.mlp.experts.15.down_proj.weight', 'model.layers.46.mlp.experts.15.gate_proj.weight', 'model.layers.46.mlp.experts.15.up_proj.weight', 'model.layers.46.mlp.experts.16.down_proj.weight', 'model.layers.46.mlp.experts.16.gate_proj.weight', 'model.layers.46.mlp.experts.16.up_proj.weight', 'model.
layers.46.mlp.experts.17.down_proj.weight', 'model.layers.46.mlp.experts.17.gate_proj.weight', 'model.layers.46.mlp.experts.17.up_proj.weight', 'model.layers.46.mlp.experts.18.down_proj.weight', 'model.layers.46.mlp.experts.18.gate_proj.weight', 'model.layers.46.mlp.experts.18.up_proj.weight', 'model.layers.46.mlp.experts.19.down_proj.weight', 'model.layers.46.mlp.experts.19.gate_proj.weight', 'model.layers.46.mlp.experts.
19.up_proj.weight', 'model.layers.46.mlp.experts.2.down_proj.weight', 'model.layers.46.mlp.experts.2.gate_proj.weight', 'model.layers.46.mlp.experts.2.up_proj.weight', 'model.layers.46.mlp.experts.20.down_proj.weight', 'model.layers.46.mlp.experts.20.gate_proj.weight', 'model.layers.46.mlp.experts.20.up_proj.weight', 'model.layers.46.mlp.experts.21.down_proj.weight', 'model.layers.46.mlp.experts.21.gate_proj.weight', 'mode
l.layers.46.mlp.experts.21.up_proj.weight', 'model.layers.46.mlp.experts.22.down_proj.weight', 'model.layers.46.mlp.experts.22.gate_proj.weight', 'model.layers.46.mlp.experts.22.up_proj.weight', 'model.layers.46.mlp.experts.23.down_proj.weight', 'model.layers.46.mlp.experts.23.gate_proj.weight', 'model.layers.46.mlp.experts.23.up_proj.weight', 'model.layers.46.mlp.experts.24.down_proj.weight', 'model.layers.46.mlp.experts.
24.gate_proj.weight', 'model.layers.46.mlp.experts.24.up_proj.weight', 'model.layers.46.mlp.experts.25.down_proj.weight', 'model.layers.46.mlp.experts.25.gate_proj.weight', 'model.layers.46.mlp.experts.25.up_proj.weight', 'model.layers.46.mlp.experts.26.down_proj.weight', 'model.layers.46.mlp.experts.26.gate_proj.weight', 'model.layers.46.mlp.experts.26.up_proj.weight', 'model.layers.46.mlp.experts.27.down_proj.weight', 'm
odel.layers.46.mlp.experts.27.gate_proj.weight', 'model.layers.46.mlp.experts.27.up_proj.weight', 'model.layers.46.mlp.experts.28.down_proj.weight', 'model.layers.46.mlp.experts.28.gate_proj.weight', 'model.layers.46.mlp.experts.28.up_proj.weight', 'model.layers.46.mlp.experts.29.down_proj.weight', 'model.layers.46.mlp.experts.29.gate_proj.weight', 'model.layers.46.mlp.experts.29.up_proj.weight', 'model.layers.46.mlp.exper
ts.3.down_proj.weight', 'model.layers.46.mlp.experts.3.gate_proj.weight', 'model.layers.46.mlp.experts.3.up_proj.weight', 'model.layers.46.mlp.experts.30.down_proj.weight', 'model.layers.46.mlp.experts.30.gate_proj.weight', 'model.layers.46.mlp.experts.30.up_proj.weight', 'model.layers.46.mlp.experts.31.down_proj.weight', 'model.layers.46.mlp.experts.31.gate_proj.weight', 'model.layers.46.mlp.experts.31.up_proj.weight', 'm
odel.layers.46.mlp.experts.32.down_proj.weight', 'model.layers.46.mlp.experts.32.gate_proj.weight', 'model.layers.46.mlp.experts.32.up_proj.weight', 'model.layers.46.mlp.experts.33.down_proj.weight', 'model.layers.46.mlp.experts.33.gate_proj.weight', 'model.layers.46.mlp.experts.33.up_proj.weight', 'model.layers.46.mlp.experts.34.down_proj.weight', 'model.layers.46.mlp.experts.34.gate_proj.weight', 'model.layers.46.mlp.exp
erts.34.up_proj.weight', 'model.layers.46.mlp.experts.35.down_proj.weight', 'model.layers.46.mlp.experts.35.gate_proj.weight', 'model.layers.46.mlp.experts.35.up_proj.weight', 'model.layers.46.mlp.experts.36.down_proj.weight', 'model.layers.46.mlp.experts.36.gate_proj.weight', 'model.layers.46.mlp.experts.36.up_proj.weight', 'model.layers.46.mlp.experts.37.down_proj.weight', 'model.layers.46.mlp.experts.37.gate_proj.weight
', 'model.layers.46.mlp.experts.37.up_proj.weight', 'model.layers.46.mlp.experts.38.down_proj.weight', 'model.layers.46.mlp.experts.38.gate_proj.weight', 'model.layers.46.mlp.experts.38.up_proj.weight', 'model.layers.46.mlp.experts.39.down_proj.weight', 'model.layers.46.mlp.experts.39.gate_proj.weight', 'model.layers.46.mlp.experts.39.up_proj.weight', 'model.layers.46.mlp.experts.4.down_proj.weight', 'model.layers.46.mlp.e
xperts.4.gate_proj.weight', 'model.layers.46.mlp.experts.4.up_proj.weight', 'model.layers.46.mlp.experts.40.down_proj.weight', 'model.layers.46.mlp.experts.40.gate_proj.weight', 'model.layers.46.mlp.experts.40.up_proj.weight', 'model.layers.46.mlp.experts.41.down_proj.weight', 'model.layers.46.mlp.experts.41.gate_proj.weight', 'model.layers.46.mlp.experts.41.up_proj.weight', 'model.layers.46.mlp.experts.42.down_proj.weight
', 'model.layers.46.mlp.experts.42.gate_proj.weight', 'model.layers.46.mlp.experts.42.up_proj.weight', 'model.layers.46.mlp.experts.43.down_proj.weight', 'model.layers.46.mlp.experts.43.gate_proj.weight', 'model.layers.46.mlp.experts.43.up_proj.weight', 'model.layers.46.mlp.experts.44.down_proj.weight', 'model.layers.46.mlp.experts.44.gate_proj.weight', 'model.layers.46.mlp.experts.44.up_proj.weight', 'model.layers.46.mlp.
experts.45.down_proj.weight', 'model.layers.46.mlp.experts.45.gate_proj.weight', 'model.layers.46.mlp.experts.45.up_proj.weight', 'model.layers.46.mlp.experts.46.down_proj.weight', 'model.layers.46.mlp.experts.46.gate_proj.weight', 'model.layers.46.mlp.experts.46.up_proj.weight', 'model.layers.46.mlp.experts.47.down_proj.weight', 'model.layers.46.mlp.experts.47.gate_proj.weight', 'model.layers.46.mlp.experts.47.up_proj.wei
ght', 'model.layers.46.mlp.experts.48.down_proj.weight', 'model.layers.46.mlp.experts.48.gate_proj.weight', 'model.layers.46.mlp.experts.48.up_proj.weight', 'model.layers.46.mlp.experts.49.down_proj.weight', 'model.layers.46.mlp.experts.49.gate_proj.weight', 'model.layers.46.mlp.experts.49.up_proj.weight', 'model.layers.46.mlp.experts.5.down_proj.weight', 'model.layers.46.mlp.experts.5.gate_proj.weight', 'model.layers.46.m
lp.experts.5.up_proj.weight', 'model.layers.46.mlp.experts.50.down_proj.weight', 'model.layers.46.mlp.experts.50.gate_proj.weight', 'model.layers.46.mlp.experts.50.up_proj.weight', 'model.layers.46.mlp.experts.51.down_proj.weight', 'model.layers.46.mlp.experts.51.gate_proj.weight', 'model.layers.46.mlp.experts.51.up_proj.weight', 'model.layers.46.mlp.experts.52.down_proj.weight', 'model.layers.46.mlp.experts.52.gate_proj.w
eight', 'model.layers.46.mlp.experts.52.up_proj.weight', 'model.layers.46.mlp.experts.53.down_proj.weight', 'model.layers.46.mlp.experts.53.gate_proj.weight', 'model.layers.46.mlp.experts.53.up_proj.weight', 'model.layers.46.mlp.experts.54.down_proj.weight', 'model.layers.46.mlp.experts.54.gate_proj.weight', 'model.layers.46.mlp.experts.54.up_proj.weight', 'model.layers.46.mlp.experts.55.down_proj.weight', 'model.layers.46
.mlp.experts.55.gate_proj.weight', 'model.layers.46.mlp.experts.55.up_proj.weight', 'model.layers.46.mlp.experts.56.down_proj.weight', 'model.layers.46.mlp.experts.56.gate_proj.weight', 'model.layers.46.mlp.experts.56.up_proj.weight', 'model.layers.46.mlp.experts.57.down_proj.weight', 'model.layers.46.mlp.experts.57.gate_proj.weight', 'model.layers.46.mlp.experts.57.up_proj.weight', 'model.layers.46.mlp.experts.58.down_pro
j.weight', 'model.layers.46.mlp.experts.58.gate_proj.weight', 'model.layers.46.mlp.experts.58.up_proj.weight', 'model.layers.46.mlp.experts.59.down_proj.weight', 'model.layers.46.mlp.experts.59.gate_proj.weight', 'model.layers.46.mlp.experts.59.up_proj.weight', 'model.layers.46.mlp.experts.6.down_proj.weight', 'model.layers.46.mlp.experts.6.gate_proj.weight', 'model.layers.46.mlp.experts.6.up_proj.weight', 'model.layers.46
.mlp.experts.60.down_proj.weight', 'model.layers.46.mlp.experts.60.gate_proj.weight', 'model.layers.46.mlp.experts.60.up_proj.weight', 'model.layers.46.mlp.experts.61.down_proj.weight', 'model.layers.46.mlp.experts.61.gate_proj.weight', 'model.layers.46.mlp.experts.61.up_proj.weight', 'model.layers.46.mlp.experts.62.down_proj.weight', 'model.layers.46.mlp.experts.62.gate_proj.weight', 'model.layers.46.mlp.experts.62.up_pro
j.weight', 'model.layers.46.mlp.experts.63.down_proj.weight', 'model.layers.46.mlp.experts.63.gate_proj.weight', 'model.layers.46.mlp.experts.63.up_proj.weight', 'model.layers.46.mlp.experts.64.down_proj.weight', 'model.layers.46.mlp.experts.64.gate_proj.weight', 'model.layers.46.mlp.experts.64.up_proj.weight', 'model.layers.46.mlp.experts.65.down_proj.weight', 'model.layers.46.mlp.experts.65.gate_proj.weight', 'model.laye
rs.46.mlp.experts.65.up_proj.weight', 'model.layers.46.mlp.experts.66.down_proj.weight', 'model.layers.46.mlp.experts.66.gate_proj.weight', 'model.layers.46.mlp.experts.66.up_proj.weight', 'model.layers.46.mlp.experts.67.down_proj.weight', 'model.layers.46.mlp.experts.67.gate_proj.weight', 'model.layers.46.mlp.experts.67.up_proj.weight', 'model.layers.46.mlp.experts.68.down_proj.weight', 'model.layers.46.mlp.experts.68.gat
e_proj.weight', 'model.layers.46.mlp.experts.68.up_proj.weight', 'model.layers.46.mlp.experts.69.down_proj.weight', 'model.layers.46.mlp.experts.69.gate_proj.weight', 'model.layers.46.mlp.experts.69.up_proj.weight', 'model.layers.46.mlp.experts.7.down_proj.weight', 'model.layers.46.mlp.experts.7.gate_proj.weight', 'model.layers.46.mlp.experts.7.up_proj.weight', 'model.layers.46.mlp.experts.70.down_proj.weight', 'model.laye
rs.46.mlp.experts.70.gate_proj.weight', 'model.layers.46.mlp.experts.70.up_proj.weight', 'model.layers.46.mlp.experts.71.down_proj.weight', 'model.layers.46.mlp.experts.71.gate_proj.weight', 'model.layers.46.mlp.experts.71.up_proj.weight', 'model.layers.46.mlp.experts.72.down_proj.weight', 'model.layers.46.mlp.experts.72.gate_proj.weight', 'model.layers.46.mlp.experts.72.up_proj.weight', 'model.layers.46.mlp.experts.73.dow
n_proj.weight', 'model.layers.46.mlp.experts.73.gate_proj.weight', 'model.layers.46.mlp.experts.73.up_proj.weight', 'model.layers.46.mlp.experts.74.down_proj.weight', 'model.layers.46.mlp.experts.74.gate_proj.weight', 'model.layers.46.mlp.experts.74.up_proj.weight', 'model.layers.46.mlp.experts.75.down_proj.weight', 'model.layers.46.mlp.experts.75.gate_proj.weight', 'model.layers.46.mlp.experts.75.up_proj.weight', 'model.l
ayers.46.mlp.experts.76.down_proj.weight', 'model.layers.46.mlp.experts.76.gate_proj.weight', 'model.layers.46.mlp.experts.76.up_proj.weight', 'model.layers.46.mlp.experts.77.down_proj.weight', 'model.layers.46.mlp.experts.77.gate_proj.weight', 'model.layers.46.mlp.experts.77.up_proj.weight', 'model.layers.46.mlp.experts.78.down_proj.weight', 'model.layers.46.mlp.experts.78.gate_proj.weight', 'model.layers.46.mlp.experts.7
8.up_proj.weight', 'model.layers.46.mlp.experts.79.down_proj.weight', 'model.layers.46.mlp.experts.79.gate_proj.weight', 'model.layers.46.mlp.experts.79.up_proj.weight', 'model.layers.46.mlp.experts.8.down_proj.weight', 'model.layers.46.mlp.experts.8.gate_proj.weight', 'model.layers.46.mlp.experts.8.up_proj.weight', 'model.layers.46.mlp.experts.80.down_proj.weight', 'model.layers.46.mlp.experts.80.gate_proj.weight', 'model
.layers.46.mlp.experts.80.up_proj.weight', 'model.layers.46.mlp.experts.81.down_proj.weight', 'model.layers.46.mlp.experts.81.gate_proj.weight', 'model.layers.46.mlp.experts.81.up_proj.weight', 'model.layers.46.mlp.experts.82.down_proj.weight', 'model.layers.46.mlp.experts.82.gate_proj.weight', 'model.layers.46.mlp.experts.82.up_proj.weight', 'model.layers.46.mlp.experts.83.down_proj.weight', 'model.layers.46.mlp.experts.8
3.gate_proj.weight', 'model.layers.46.mlp.experts.83.up_proj.weight', 'model.layers.46.mlp.experts.84.down_proj.weight', 'model.layers.46.mlp.experts.84.gate_proj.weight', 'model.layers.46.mlp.experts.84.up_proj.weight', 'model.layers.46.mlp.experts.85.down_proj.weight', 'model.layers.46.mlp.experts.85.gate_proj.weight', 'model.layers.46.mlp.experts.85.up_proj.weight', 'model.layers.46.mlp.experts.86.down_proj.weight', 'mo
del.layers.46.mlp.experts.86.gate_proj.weight', 'model.layers.46.mlp.experts.86.up_proj.weight', 'model.layers.46.mlp.experts.87.down_proj.weight', 'model.layers.46.mlp.experts.87.gate_proj.weight', 'model.layers.46.mlp.experts.87.up_proj.weight', 'model.layers.46.mlp.experts.88.down_proj.weight', 'model.layers.46.mlp.experts.88.gate_proj.weight', 'model.layers.46.mlp.experts.88.up_proj.weight', 'model.layers.46.mlp.expert
s.89.down_proj.weight', 'model.layers.46.mlp.experts.89.gate_proj.weight', 'model.layers.46.mlp.experts.89.up_proj.weight', 'model.layers.46.mlp.experts.9.down_proj.weight', 'model.layers.46.mlp.experts.9.gate_proj.weight', 'model.layers.46.mlp.experts.9.up_proj.weight', 'model.layers.46.mlp.experts.90.down_proj.weight', 'model.layers.46.mlp.experts.90.gate_proj.weight', 'model.layers.46.mlp.experts.90.up_proj.weight', 'mo
del.layers.46.mlp.experts.91.down_proj.weight', 'model.layers.46.mlp.experts.91.gate_proj.weight', 'model.layers.46.mlp.experts.91.up_proj.weight', 'model.layers.46.mlp.experts.92.down_proj.weight', 'model.layers.46.mlp.experts.92.gate_proj.weight', 'model.layers.46.mlp.experts.92.up_proj.weight', 'model.layers.46.mlp.experts.93.down_proj.weight', 'model.layers.46.mlp.experts.93.gate_proj.weight', 'model.layers.46.mlp.expe
rts.93.up_proj.weight', 'model.layers.46.mlp.experts.94.down_proj.weight', 'model.layers.46.mlp.experts.94.gate_proj.weight', 'model.layers.46.mlp.experts.94.up_proj.weight', 'model.layers.46.mlp.experts.95.down_proj.weight', 'model.layers.46.mlp.experts.95.gate_proj.weight', 'model.layers.46.mlp.experts.95.up_proj.weight', 'model.layers.46.mlp.experts.96.down_proj.weight', 'model.layers.46.mlp.experts.96.gate_proj.weight'
, 'model.layers.46.mlp.experts.96.up_proj.weight', 'model.layers.46.mlp.experts.97.down_proj.weight', 'model.layers.46.mlp.experts.97.gate_proj.weight', 'model.layers.46.mlp.experts.97.up_proj.weight', 'model.layers.46.mlp.experts.98.down_proj.weight', 'model.layers.46.mlp.experts.98.gate_proj.weight', 'model.layers.46.mlp.experts.98.up_proj.weight', 'model.layers.46.mlp.experts.99.down_proj.weight', 'model.layers.46.mlp.e
xperts.99.gate_proj.weight', 'model.layers.46.mlp.experts.99.up_proj.weight', 'model.layers.46.mlp.gate.e_score_correction_bias', 'model.layers.46.mlp.gate.weight', 'model.layers.46.mlp.shared_experts.down_proj.weight', 'model.layers.46.mlp.shared_experts.gate_proj.weight', 'model.layers.46.mlp.shared_experts.up_proj.weight', 'model.layers.46.post_attention_layernorm.weight', 'model.layers.46.self_attn.k_proj.bias', 'model
.layers.46.self_attn.k_proj.weight', 'model.layers.46.self_attn.o_proj.weight', 'model.layers.46.self_attn.q_proj.bias', 'model.layers.46.self_attn.q_proj.weight', 'model.layers.46.self_attn.v_proj.bias', 'model.layers.46.self_attn.v_proj.weight', 'model.layers.46.shared_head.head.weight', 'model.layers.46.shared_head.norm.weight']
- This IS expected if you are initializing Glm4MoeForCausalLM from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
- This IS NOT expected if you are initializing Glm4MoeForCausalLM from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).

Based on model.safetensors.index.json, there are 47 layers, but in config.json, there are only 46 layers. Is this intentional?

Thanks.

This is the MTP layer, which does not affect the model output. The transformers repository does not implement MTP layers.

If you use vLLM or sglang, there won't be this problem, because MTP is implemented

cpatonn changed discussion status to closed

Sign up or log in to comment