mixtral layers and shapes

gistfile1.txt

tok_embeddings.weight torch.Size([32000, 4096])
norm.weight torch.Size([4096])
output.weight torch.Size([32000, 4096])
layers.0.attention_norm.weight torch.Size([4096])
layers.0.attention.wq.weight torch.Size([4096, 4096])
layers.0.attention.wk.weight torch.Size([1024, 4096])
layers.0.attention.wv.weight torch.Size([1024, 4096])
layers.0.attention.wo.weight torch.Size([4096, 4096])
layers.0.feed_forward.gate.weight torch.Size([8, 4096])
layers.0.ffn_norm.weight torch.Size([4096])
layers.1.attention_norm.weight torch.Size([4096])
layers.1.attention.wq.weight torch.Size([4096, 4096])
layers.1.attention.wk.weight torch.Size([1024, 4096])
layers.1.attention.wv.weight torch.Size([1024, 4096])
layers.1.attention.wo.weight torch.Size([4096, 4096])
layers.1.feed_forward.gate.weight torch.Size([8, 4096])
layers.1.ffn_norm.weight torch.Size([4096])
layers.2.attention_norm.weight torch.Size([4096])
layers.2.attention.wq.weight torch.Size([4096, 4096])
layers.2.attention.wk.weight torch.Size([1024, 4096])
layers.2.attention.wv.weight torch.Size([1024, 4096])
layers.2.attention.wo.weight torch.Size([4096, 4096])
layers.2.feed_forward.gate.weight torch.Size([8, 4096])
layers.2.ffn_norm.weight torch.Size([4096])
layers.3.attention_norm.weight torch.Size([4096])
layers.3.attention.wq.weight torch.Size([4096, 4096])
layers.3.attention.wk.weight torch.Size([1024, 4096])
layers.3.attention.wv.weight torch.Size([1024, 4096])
layers.3.attention.wo.weight torch.Size([4096, 4096])
layers.3.feed_forward.gate.weight torch.Size([8, 4096])
layers.3.ffn_norm.weight torch.Size([4096])
layers.4.attention_norm.weight torch.Size([4096])
layers.4.attention.wq.weight torch.Size([4096, 4096])
layers.4.attention.wk.weight torch.Size([1024, 4096])
layers.4.attention.wv.weight torch.Size([1024, 4096])
layers.4.attention.wo.weight torch.Size([4096, 4096])
layers.4.feed_forward.gate.weight torch.Size([8, 4096])
layers.4.ffn_norm.weight torch.Size([4096])
layers.5.attention_norm.weight torch.Size([4096])
layers.5.attention.wq.weight torch.Size([4096, 4096])
layers.5.attention.wk.weight torch.Size([1024, 4096])
layers.5.attention.wv.weight torch.Size([1024, 4096])
layers.5.attention.wo.weight torch.Size([4096, 4096])
layers.5.feed_forward.gate.weight torch.Size([8, 4096])
layers.5.ffn_norm.weight torch.Size([4096])
layers.6.attention_norm.weight torch.Size([4096])
layers.6.attention.wq.weight torch.Size([4096, 4096])
layers.6.attention.wk.weight torch.Size([1024, 4096])
layers.6.attention.wv.weight torch.Size([1024, 4096])
layers.6.attention.wo.weight torch.Size([4096, 4096])
layers.6.feed_forward.gate.weight torch.Size([8, 4096])
layers.6.ffn_norm.weight torch.Size([4096])
layers.7.attention_norm.weight torch.Size([4096])
layers.7.attention.wq.weight torch.Size([4096, 4096])
layers.7.attention.wk.weight torch.Size([1024, 4096])
layers.7.attention.wv.weight torch.Size([1024, 4096])
layers.7.attention.wo.weight torch.Size([4096, 4096])
layers.7.feed_forward.gate.weight torch.Size([8, 4096])
layers.7.ffn_norm.weight torch.Size([4096])
layers.8.attention_norm.weight torch.Size([4096])
layers.8.attention.wq.weight torch.Size([4096, 4096])
layers.8.attention.wk.weight torch.Size([1024, 4096])
layers.8.attention.wv.weight torch.Size([1024, 4096])
layers.8.attention.wo.weight torch.Size([4096, 4096])
layers.8.feed_forward.gate.weight torch.Size([8, 4096])
layers.8.ffn_norm.weight torch.Size([4096])
layers.9.attention_norm.weight torch.Size([4096])
layers.9.attention.wq.weight torch.Size([4096, 4096])
layers.9.attention.wk.weight torch.Size([1024, 4096])
layers.9.attention.wv.weight torch.Size([1024, 4096])
layers.9.attention.wo.weight torch.Size([4096, 4096])
layers.9.feed_forward.gate.weight torch.Size([8, 4096])
layers.9.ffn_norm.weight torch.Size([4096])
layers.10.attention_norm.weight torch.Size([4096])
layers.10.attention.wq.weight torch.Size([4096, 4096])
layers.10.attention.wk.weight torch.Size([1024, 4096])
layers.10.attention.wv.weight torch.Size([1024, 4096])
layers.10.attention.wo.weight torch.Size([4096, 4096])
layers.10.feed_forward.gate.weight torch.Size([8, 4096])
layers.10.ffn_norm.weight torch.Size([4096])
layers.11.attention_norm.weight torch.Size([4096])
layers.11.attention.wq.weight torch.Size([4096, 4096])
layers.11.attention.wk.weight torch.Size([1024, 4096])
layers.11.attention.wv.weight torch.Size([1024, 4096])
layers.11.attention.wo.weight torch.Size([4096, 4096])
layers.11.feed_forward.gate.weight torch.Size([8, 4096])
layers.11.ffn_norm.weight torch.Size([4096])
layers.12.attention_norm.weight torch.Size([4096])
layers.12.attention.wq.weight torch.Size([4096, 4096])
layers.12.attention.wk.weight torch.Size([1024, 4096])
layers.12.attention.wv.weight torch.Size([1024, 4096])
layers.12.attention.wo.weight torch.Size([4096, 4096])
layers.12.feed_forward.gate.weight torch.Size([8, 4096])
layers.12.ffn_norm.weight torch.Size([4096])
layers.13.attention_norm.weight torch.Size([4096])
layers.13.attention.wq.weight torch.Size([4096, 4096])
layers.13.attention.wk.weight torch.Size([1024, 4096])
layers.13.attention.wv.weight torch.Size([1024, 4096])
layers.13.attention.wo.weight torch.Size([4096, 4096])
layers.13.feed_forward.gate.weight torch.Size([8, 4096])
layers.13.ffn_norm.weight torch.Size([4096])
layers.14.attention_norm.weight torch.Size([4096])
layers.14.attention.wq.weight torch.Size([4096, 4096])
layers.14.attention.wk.weight torch.Size([1024, 4096])
layers.14.attention.wv.weight torch.Size([1024, 4096])
layers.14.attention.wo.weight torch.Size([4096, 4096])
layers.14.feed_forward.gate.weight torch.Size([8, 4096])
layers.14.ffn_norm.weight torch.Size([4096])
layers.15.attention_norm.weight torch.Size([4096])
layers.15.attention.wq.weight torch.Size([4096, 4096])
layers.15.attention.wk.weight torch.Size([1024, 4096])
layers.15.attention.wv.weight torch.Size([1024, 4096])
layers.15.attention.wo.weight torch.Size([4096, 4096])
layers.15.feed_forward.gate.weight torch.Size([8, 4096])
layers.15.ffn_norm.weight torch.Size([4096])
layers.16.attention_norm.weight torch.Size([4096])
layers.16.attention.wq.weight torch.Size([4096, 4096])
layers.16.attention.wk.weight torch.Size([1024, 4096])
layers.16.attention.wv.weight torch.Size([1024, 4096])
layers.16.attention.wo.weight torch.Size([4096, 4096])
layers.16.feed_forward.gate.weight torch.Size([8, 4096])
layers.16.ffn_norm.weight torch.Size([4096])
layers.17.attention_norm.weight torch.Size([4096])
layers.17.attention.wq.weight torch.Size([4096, 4096])
layers.17.attention.wk.weight torch.Size([1024, 4096])
layers.17.attention.wv.weight torch.Size([1024, 4096])
layers.17.attention.wo.weight torch.Size([4096, 4096])
layers.17.feed_forward.gate.weight torch.Size([8, 4096])
layers.17.ffn_norm.weight torch.Size([4096])
layers.18.attention_norm.weight torch.Size([4096])
layers.18.attention.wq.weight torch.Size([4096, 4096])
layers.18.attention.wk.weight torch.Size([1024, 4096])
layers.18.attention.wv.weight torch.Size([1024, 4096])
layers.18.attention.wo.weight torch.Size([4096, 4096])
layers.18.feed_forward.gate.weight torch.Size([8, 4096])
layers.18.ffn_norm.weight torch.Size([4096])
layers.19.attention_norm.weight torch.Size([4096])
layers.19.attention.wq.weight torch.Size([4096, 4096])
layers.19.attention.wk.weight torch.Size([1024, 4096])
layers.19.attention.wv.weight torch.Size([1024, 4096])
layers.19.attention.wo.weight torch.Size([4096, 4096])
layers.19.feed_forward.gate.weight torch.Size([8, 4096])
layers.19.ffn_norm.weight torch.Size([4096])
layers.20.attention_norm.weight torch.Size([4096])
layers.20.attention.wq.weight torch.Size([4096, 4096])
layers.20.attention.wk.weight torch.Size([1024, 4096])
layers.20.attention.wv.weight torch.Size([1024, 4096])
layers.20.attention.wo.weight torch.Size([4096, 4096])
layers.20.feed_forward.gate.weight torch.Size([8, 4096])
layers.20.ffn_norm.weight torch.Size([4096])
layers.21.attention_norm.weight torch.Size([4096])
layers.21.attention.wq.weight torch.Size([4096, 4096])
layers.21.attention.wk.weight torch.Size([1024, 4096])
layers.21.attention.wv.weight torch.Size([1024, 4096])
layers.21.attention.wo.weight torch.Size([4096, 4096])
layers.21.feed_forward.gate.weight torch.Size([8, 4096])
layers.21.ffn_norm.weight torch.Size([4096])
layers.22.attention_norm.weight torch.Size([4096])
layers.22.attention.wq.weight torch.Size([4096, 4096])
layers.22.attention.wk.weight torch.Size([1024, 4096])
layers.22.attention.wv.weight torch.Size([1024, 4096])
layers.22.attention.wo.weight torch.Size([4096, 4096])
layers.22.feed_forward.gate.weight torch.Size([8, 4096])
layers.22.ffn_norm.weight torch.Size([4096])
layers.23.attention_norm.weight torch.Size([4096])
layers.23.attention.wq.weight torch.Size([4096, 4096])
layers.23.attention.wk.weight torch.Size([1024, 4096])
layers.23.attention.wv.weight torch.Size([1024, 4096])
layers.23.attention.wo.weight torch.Size([4096, 4096])
layers.23.feed_forward.gate.weight torch.Size([8, 4096])
layers.23.ffn_norm.weight torch.Size([4096])
layers.24.attention_norm.weight torch.Size([4096])
layers.24.attention.wq.weight torch.Size([4096, 4096])
layers.24.attention.wk.weight torch.Size([1024, 4096])
layers.24.attention.wv.weight torch.Size([1024, 4096])
layers.24.attention.wo.weight torch.Size([4096, 4096])
layers.24.feed_forward.gate.weight torch.Size([8, 4096])
layers.24.ffn_norm.weight torch.Size([4096])
layers.25.attention_norm.weight torch.Size([4096])
layers.25.attention.wq.weight torch.Size([4096, 4096])
layers.25.attention.wk.weight torch.Size([1024, 4096])
layers.25.attention.wv.weight torch.Size([1024, 4096])
layers.25.attention.wo.weight torch.Size([4096, 4096])
layers.25.feed_forward.gate.weight torch.Size([8, 4096])
layers.25.ffn_norm.weight torch.Size([4096])
layers.26.attention_norm.weight torch.Size([4096])
layers.26.attention.wq.weight torch.Size([4096, 4096])
layers.26.attention.wk.weight torch.Size([1024, 4096])
layers.26.attention.wv.weight torch.Size([1024, 4096])
layers.26.attention.wo.weight torch.Size([4096, 4096])
layers.26.feed_forward.gate.weight torch.Size([8, 4096])
layers.26.ffn_norm.weight torch.Size([4096])
layers.27.attention_norm.weight torch.Size([4096])
layers.27.attention.wq.weight torch.Size([4096, 4096])
layers.27.attention.wk.weight torch.Size([1024, 4096])
layers.27.attention.wv.weight torch.Size([1024, 4096])
layers.27.attention.wo.weight torch.Size([4096, 4096])
layers.27.feed_forward.gate.weight torch.Size([8, 4096])
layers.27.ffn_norm.weight torch.Size([4096])
layers.28.attention_norm.weight torch.Size([4096])
layers.28.attention.wq.weight torch.Size([4096, 4096])
layers.28.attention.wk.weight torch.Size([1024, 4096])
layers.28.attention.wv.weight torch.Size([1024, 4096])
layers.28.attention.wo.weight torch.Size([4096, 4096])
layers.28.feed_forward.gate.weight torch.Size([8, 4096])
layers.28.ffn_norm.weight torch.Size([4096])
layers.29.attention_norm.weight torch.Size([4096])
layers.29.attention.wq.weight torch.Size([4096, 4096])
layers.29.attention.wk.weight torch.Size([1024, 4096])
layers.29.attention.wv.weight torch.Size([1024, 4096])
layers.29.attention.wo.weight torch.Size([4096, 4096])
layers.29.feed_forward.gate.weight torch.Size([8, 4096])
layers.29.ffn_norm.weight torch.Size([4096])
layers.30.attention_norm.weight torch.Size([4096])
layers.30.attention.wq.weight torch.Size([4096, 4096])
layers.30.attention.wk.weight torch.Size([1024, 4096])
layers.30.attention.wv.weight torch.Size([1024, 4096])
layers.30.attention.wo.weight torch.Size([4096, 4096])
layers.30.feed_forward.gate.weight torch.Size([8, 4096])
layers.30.ffn_norm.weight torch.Size([4096])
layers.31.attention_norm.weight torch.Size([4096])
layers.31.attention.wq.weight torch.Size([4096, 4096])
layers.31.attention.wk.weight torch.Size([1024, 4096])
layers.31.attention.wv.weight torch.Size([1024, 4096])
layers.31.attention.wo.weight torch.Size([4096, 4096])
layers.31.feed_forward.gate.weight torch.Size([8, 4096])
layers.31.ffn_norm.weight torch.Size([4096])
layers.0.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.0.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.0.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.0.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.0.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.0.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.0.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.0.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.0.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.0.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.1.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.1.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.1.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.1.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.1.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.1.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.1.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.1.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.1.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.2.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.2.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.2.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.2.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.2.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.2.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.2.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.2.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.2.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.3.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.3.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.3.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.3.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.3.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.3.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.3.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.3.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.3.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.4.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.4.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.4.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.4.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.4.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.4.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.4.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.4.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.4.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.5.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.5.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.5.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.5.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.5.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.5.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.5.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.5.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.5.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.6.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.6.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.6.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.6.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.6.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.6.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.6.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.6.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.6.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.7.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.7.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.7.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.7.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.7.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.7.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.7.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.7.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.7.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.8.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.8.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.8.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.8.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.8.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.8.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.8.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.8.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.8.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.9.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.9.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.9.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.9.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.9.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.9.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.9.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.9.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.9.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.10.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.10.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.10.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.10.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.10.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.10.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.10.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.10.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.10.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.11.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.11.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.11.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.11.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.11.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.11.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.11.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.11.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.11.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.12.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.12.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.12.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.12.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.12.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.12.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.12.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.12.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.12.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.13.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.13.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.13.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.13.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.13.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.13.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.13.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.13.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.13.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.14.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.14.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.14.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.14.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.14.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.14.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.14.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.14.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.14.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.15.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.15.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.15.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.15.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.15.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.15.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.15.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.15.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.15.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.16.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.16.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.16.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.16.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.16.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.16.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.16.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.16.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.16.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.17.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.17.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.17.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.17.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.17.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.17.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.17.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.17.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.17.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.18.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.18.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.18.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.18.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.18.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.18.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.18.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.18.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.18.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.19.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.19.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.19.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.19.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.19.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.19.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.19.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.19.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.19.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.20.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.20.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.20.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.20.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.20.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.20.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.20.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.20.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.20.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.21.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.21.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.21.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.21.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.21.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.21.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.21.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.21.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.21.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.22.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.22.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.22.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.22.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.22.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.22.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.22.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.22.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.22.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.23.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.23.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.23.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.23.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.23.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.23.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.23.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.23.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.23.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.24.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.24.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.24.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.24.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.24.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.24.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.24.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.24.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.24.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.25.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.25.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.25.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.25.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.25.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.25.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.25.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.25.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.25.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.26.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.26.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.26.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.26.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.26.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.26.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.26.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.26.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.26.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.27.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.27.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.27.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.27.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.27.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.27.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.27.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.27.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.27.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.28.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.28.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.28.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.28.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.28.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.28.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.28.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.28.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.28.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.29.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.29.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.29.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.29.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.29.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.29.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.29.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.29.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.29.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.30.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.30.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.30.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.30.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.30.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.30.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.30.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.30.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.30.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.0.w1.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.1.w1.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.2.w1.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.3.w1.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.4.w1.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.5.w1.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.6.w1.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.7.w1.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.0.w2.weight torch.Size([4096, 14336])
layers.31.feed_forward.experts.1.w2.weight torch.Size([4096, 14336])
layers.31.feed_forward.experts.2.w2.weight torch.Size([4096, 14336])
layers.31.feed_forward.experts.3.w2.weight torch.Size([4096, 14336])
layers.31.feed_forward.experts.4.w2.weight torch.Size([4096, 14336])
layers.31.feed_forward.experts.5.w2.weight torch.Size([4096, 14336])
layers.31.feed_forward.experts.6.w2.weight torch.Size([4096, 14336])
layers.31.feed_forward.experts.7.w2.weight torch.Size([4096, 14336])
layers.31.feed_forward.experts.0.w3.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.1.w3.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.2.w3.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.3.w3.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.4.w3.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.5.w3.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.6.w3.weight torch.Size([14336, 4096])
layers.31.feed_forward.experts.7.w3.weight torch.Size([14336, 4096])