gptneox-rm-static / index.json
Dahoas's picture
Upload index.json
5e05c58
raw
history blame
95.7 kB
{
"model.gpt_neox.embed_in.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.input_layernorm.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.input_layernorm.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.post_attention_layernorm.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.post_attention_layernorm.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.attention.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.attention.masked_bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.attention.rotary_emb.inv_freq": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.attention.query_key_value.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.attention.query_key_value.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.attention.dense.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.attention.dense.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.mlp.dense_h_to_4h.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.mlp.dense_h_to_4h.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.mlp.dense_4h_to_h.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.0.mlp.dense_4h_to_h.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.input_layernorm.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.input_layernorm.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.post_attention_layernorm.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.post_attention_layernorm.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.attention.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.attention.masked_bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.attention.rotary_emb.inv_freq": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.attention.query_key_value.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.attention.query_key_value.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.attention.dense.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.attention.dense.bias": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.mlp.dense_h_to_4h.weight": "hf_ckpt_0.pt",
"model.gpt_neox.layers.1.mlp.dense_h_to_4h.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.1.mlp.dense_4h_to_h.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.1.mlp.dense_4h_to_h.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.input_layernorm.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.input_layernorm.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.post_attention_layernorm.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.post_attention_layernorm.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.attention.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.attention.masked_bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.attention.rotary_emb.inv_freq": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.attention.query_key_value.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.attention.query_key_value.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.attention.dense.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.attention.dense.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.mlp.dense_h_to_4h.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.mlp.dense_h_to_4h.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.mlp.dense_4h_to_h.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.2.mlp.dense_4h_to_h.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.input_layernorm.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.input_layernorm.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.post_attention_layernorm.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.post_attention_layernorm.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.attention.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.attention.masked_bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.attention.rotary_emb.inv_freq": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.attention.query_key_value.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.attention.query_key_value.bias": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.attention.dense.weight": "hf_ckpt_1.pt",
"model.gpt_neox.layers.3.attention.dense.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.3.mlp.dense_h_to_4h.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.3.mlp.dense_h_to_4h.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.3.mlp.dense_4h_to_h.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.3.mlp.dense_4h_to_h.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.input_layernorm.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.input_layernorm.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.post_attention_layernorm.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.post_attention_layernorm.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.attention.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.attention.masked_bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.attention.rotary_emb.inv_freq": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.attention.query_key_value.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.attention.query_key_value.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.attention.dense.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.attention.dense.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.mlp.dense_h_to_4h.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.mlp.dense_h_to_4h.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.mlp.dense_4h_to_h.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.4.mlp.dense_4h_to_h.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.5.input_layernorm.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.5.input_layernorm.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.5.post_attention_layernorm.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.5.post_attention_layernorm.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.5.attention.bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.5.attention.masked_bias": "hf_ckpt_2.pt",
"model.gpt_neox.layers.5.attention.rotary_emb.inv_freq": "hf_ckpt_2.pt",
"model.gpt_neox.layers.5.attention.query_key_value.weight": "hf_ckpt_2.pt",
"model.gpt_neox.layers.5.attention.query_key_value.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.5.attention.dense.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.5.attention.dense.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.5.mlp.dense_h_to_4h.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.5.mlp.dense_h_to_4h.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.5.mlp.dense_4h_to_h.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.5.mlp.dense_4h_to_h.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.input_layernorm.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.input_layernorm.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.post_attention_layernorm.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.post_attention_layernorm.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.attention.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.attention.masked_bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.attention.rotary_emb.inv_freq": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.attention.query_key_value.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.attention.query_key_value.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.attention.dense.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.attention.dense.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.mlp.dense_h_to_4h.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.mlp.dense_h_to_4h.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.mlp.dense_4h_to_h.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.6.mlp.dense_4h_to_h.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.7.input_layernorm.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.7.input_layernorm.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.7.post_attention_layernorm.weight": "hf_ckpt_3.pt",
"model.gpt_neox.layers.7.post_attention_layernorm.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.7.attention.bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.7.attention.masked_bias": "hf_ckpt_3.pt",
"model.gpt_neox.layers.7.attention.rotary_emb.inv_freq": "hf_ckpt_4.pt",
"model.gpt_neox.layers.7.attention.query_key_value.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.7.attention.query_key_value.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.7.attention.dense.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.7.attention.dense.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.7.mlp.dense_h_to_4h.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.7.mlp.dense_h_to_4h.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.7.mlp.dense_4h_to_h.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.7.mlp.dense_4h_to_h.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.input_layernorm.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.input_layernorm.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.post_attention_layernorm.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.post_attention_layernorm.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.attention.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.attention.masked_bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.attention.rotary_emb.inv_freq": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.attention.query_key_value.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.attention.query_key_value.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.attention.dense.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.attention.dense.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.mlp.dense_h_to_4h.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.mlp.dense_h_to_4h.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.mlp.dense_4h_to_h.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.8.mlp.dense_4h_to_h.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.9.input_layernorm.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.9.input_layernorm.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.9.post_attention_layernorm.weight": "hf_ckpt_4.pt",
"model.gpt_neox.layers.9.post_attention_layernorm.bias": "hf_ckpt_4.pt",
"model.gpt_neox.layers.9.attention.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.9.attention.masked_bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.9.attention.rotary_emb.inv_freq": "hf_ckpt_5.pt",
"model.gpt_neox.layers.9.attention.query_key_value.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.9.attention.query_key_value.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.9.attention.dense.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.9.attention.dense.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.9.mlp.dense_h_to_4h.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.9.mlp.dense_h_to_4h.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.9.mlp.dense_4h_to_h.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.9.mlp.dense_4h_to_h.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.input_layernorm.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.input_layernorm.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.post_attention_layernorm.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.post_attention_layernorm.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.attention.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.attention.masked_bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.attention.rotary_emb.inv_freq": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.attention.query_key_value.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.attention.query_key_value.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.attention.dense.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.attention.dense.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.mlp.dense_h_to_4h.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.mlp.dense_h_to_4h.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.mlp.dense_4h_to_h.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.10.mlp.dense_4h_to_h.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.11.input_layernorm.weight": "hf_ckpt_5.pt",
"model.gpt_neox.layers.11.input_layernorm.bias": "hf_ckpt_5.pt",
"model.gpt_neox.layers.11.post_attention_layernorm.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.post_attention_layernorm.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.attention.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.attention.masked_bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.attention.rotary_emb.inv_freq": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.attention.query_key_value.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.attention.query_key_value.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.attention.dense.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.attention.dense.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.mlp.dense_h_to_4h.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.mlp.dense_h_to_4h.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.mlp.dense_4h_to_h.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.11.mlp.dense_4h_to_h.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.input_layernorm.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.input_layernorm.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.post_attention_layernorm.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.post_attention_layernorm.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.attention.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.attention.masked_bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.attention.rotary_emb.inv_freq": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.attention.query_key_value.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.attention.query_key_value.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.attention.dense.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.attention.dense.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.mlp.dense_h_to_4h.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.mlp.dense_h_to_4h.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.mlp.dense_4h_to_h.weight": "hf_ckpt_6.pt",
"model.gpt_neox.layers.12.mlp.dense_4h_to_h.bias": "hf_ckpt_6.pt",
"model.gpt_neox.layers.13.input_layernorm.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.input_layernorm.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.post_attention_layernorm.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.post_attention_layernorm.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.attention.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.attention.masked_bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.attention.rotary_emb.inv_freq": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.attention.query_key_value.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.attention.query_key_value.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.attention.dense.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.attention.dense.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.mlp.dense_h_to_4h.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.mlp.dense_h_to_4h.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.mlp.dense_4h_to_h.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.13.mlp.dense_4h_to_h.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.input_layernorm.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.input_layernorm.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.post_attention_layernorm.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.post_attention_layernorm.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.attention.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.attention.masked_bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.attention.rotary_emb.inv_freq": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.attention.query_key_value.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.attention.query_key_value.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.attention.dense.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.attention.dense.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.mlp.dense_h_to_4h.weight": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.mlp.dense_h_to_4h.bias": "hf_ckpt_7.pt",
"model.gpt_neox.layers.14.mlp.dense_4h_to_h.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.14.mlp.dense_4h_to_h.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.input_layernorm.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.input_layernorm.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.post_attention_layernorm.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.post_attention_layernorm.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.attention.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.attention.masked_bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.attention.rotary_emb.inv_freq": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.attention.query_key_value.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.attention.query_key_value.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.attention.dense.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.attention.dense.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.mlp.dense_h_to_4h.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.mlp.dense_h_to_4h.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.mlp.dense_4h_to_h.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.15.mlp.dense_4h_to_h.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.input_layernorm.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.input_layernorm.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.post_attention_layernorm.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.post_attention_layernorm.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.attention.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.attention.masked_bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.attention.rotary_emb.inv_freq": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.attention.query_key_value.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.attention.query_key_value.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.attention.dense.weight": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.attention.dense.bias": "hf_ckpt_8.pt",
"model.gpt_neox.layers.16.mlp.dense_h_to_4h.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.16.mlp.dense_h_to_4h.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.16.mlp.dense_4h_to_h.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.16.mlp.dense_4h_to_h.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.input_layernorm.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.input_layernorm.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.post_attention_layernorm.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.post_attention_layernorm.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.attention.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.attention.masked_bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.attention.rotary_emb.inv_freq": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.attention.query_key_value.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.attention.query_key_value.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.attention.dense.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.attention.dense.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.mlp.dense_h_to_4h.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.mlp.dense_h_to_4h.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.mlp.dense_4h_to_h.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.17.mlp.dense_4h_to_h.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.18.input_layernorm.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.18.input_layernorm.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.18.post_attention_layernorm.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.18.post_attention_layernorm.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.18.attention.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.18.attention.masked_bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.18.attention.rotary_emb.inv_freq": "hf_ckpt_9.pt",
"model.gpt_neox.layers.18.attention.query_key_value.weight": "hf_ckpt_9.pt",
"model.gpt_neox.layers.18.attention.query_key_value.bias": "hf_ckpt_9.pt",
"model.gpt_neox.layers.18.attention.dense.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.18.attention.dense.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.18.mlp.dense_h_to_4h.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.18.mlp.dense_h_to_4h.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.18.mlp.dense_4h_to_h.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.18.mlp.dense_4h_to_h.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.input_layernorm.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.input_layernorm.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.post_attention_layernorm.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.post_attention_layernorm.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.attention.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.attention.masked_bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.attention.rotary_emb.inv_freq": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.attention.query_key_value.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.attention.query_key_value.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.attention.dense.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.attention.dense.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.mlp.dense_h_to_4h.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.mlp.dense_h_to_4h.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.mlp.dense_4h_to_h.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.19.mlp.dense_4h_to_h.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.20.input_layernorm.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.20.input_layernorm.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.20.post_attention_layernorm.weight": "hf_ckpt_10.pt",
"model.gpt_neox.layers.20.post_attention_layernorm.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.20.attention.bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.20.attention.masked_bias": "hf_ckpt_10.pt",
"model.gpt_neox.layers.20.attention.rotary_emb.inv_freq": "hf_ckpt_10.pt",
"model.gpt_neox.layers.20.attention.query_key_value.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.20.attention.query_key_value.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.20.attention.dense.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.20.attention.dense.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.20.mlp.dense_h_to_4h.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.20.mlp.dense_h_to_4h.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.20.mlp.dense_4h_to_h.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.20.mlp.dense_4h_to_h.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.input_layernorm.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.input_layernorm.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.post_attention_layernorm.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.post_attention_layernorm.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.attention.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.attention.masked_bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.attention.rotary_emb.inv_freq": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.attention.query_key_value.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.attention.query_key_value.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.attention.dense.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.attention.dense.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.mlp.dense_h_to_4h.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.mlp.dense_h_to_4h.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.mlp.dense_4h_to_h.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.21.mlp.dense_4h_to_h.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.22.input_layernorm.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.22.input_layernorm.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.22.post_attention_layernorm.weight": "hf_ckpt_11.pt",
"model.gpt_neox.layers.22.post_attention_layernorm.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.22.attention.bias": "hf_ckpt_11.pt",
"model.gpt_neox.layers.22.attention.masked_bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.22.attention.rotary_emb.inv_freq": "hf_ckpt_12.pt",
"model.gpt_neox.layers.22.attention.query_key_value.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.22.attention.query_key_value.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.22.attention.dense.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.22.attention.dense.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.22.mlp.dense_h_to_4h.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.22.mlp.dense_h_to_4h.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.22.mlp.dense_4h_to_h.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.22.mlp.dense_4h_to_h.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.input_layernorm.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.input_layernorm.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.post_attention_layernorm.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.post_attention_layernorm.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.attention.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.attention.masked_bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.attention.rotary_emb.inv_freq": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.attention.query_key_value.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.attention.query_key_value.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.attention.dense.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.attention.dense.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.mlp.dense_h_to_4h.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.mlp.dense_h_to_4h.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.mlp.dense_4h_to_h.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.23.mlp.dense_4h_to_h.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.24.input_layernorm.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.24.input_layernorm.bias": "hf_ckpt_12.pt",
"model.gpt_neox.layers.24.post_attention_layernorm.weight": "hf_ckpt_12.pt",
"model.gpt_neox.layers.24.post_attention_layernorm.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.attention.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.attention.masked_bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.attention.rotary_emb.inv_freq": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.attention.query_key_value.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.attention.query_key_value.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.attention.dense.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.attention.dense.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.mlp.dense_h_to_4h.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.mlp.dense_h_to_4h.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.mlp.dense_4h_to_h.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.24.mlp.dense_4h_to_h.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.input_layernorm.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.input_layernorm.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.post_attention_layernorm.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.post_attention_layernorm.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.attention.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.attention.masked_bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.attention.rotary_emb.inv_freq": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.attention.query_key_value.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.attention.query_key_value.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.attention.dense.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.attention.dense.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.mlp.dense_h_to_4h.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.mlp.dense_h_to_4h.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.mlp.dense_4h_to_h.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.25.mlp.dense_4h_to_h.bias": "hf_ckpt_13.pt",
"model.gpt_neox.layers.26.input_layernorm.weight": "hf_ckpt_13.pt",
"model.gpt_neox.layers.26.input_layernorm.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.post_attention_layernorm.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.post_attention_layernorm.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.attention.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.attention.masked_bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.attention.rotary_emb.inv_freq": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.attention.query_key_value.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.attention.query_key_value.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.attention.dense.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.attention.dense.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.mlp.dense_h_to_4h.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.mlp.dense_h_to_4h.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.mlp.dense_4h_to_h.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.26.mlp.dense_4h_to_h.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.input_layernorm.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.input_layernorm.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.post_attention_layernorm.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.post_attention_layernorm.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.attention.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.attention.masked_bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.attention.rotary_emb.inv_freq": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.attention.query_key_value.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.attention.query_key_value.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.attention.dense.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.attention.dense.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.mlp.dense_h_to_4h.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.mlp.dense_h_to_4h.bias": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.mlp.dense_4h_to_h.weight": "hf_ckpt_14.pt",
"model.gpt_neox.layers.27.mlp.dense_4h_to_h.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.input_layernorm.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.input_layernorm.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.post_attention_layernorm.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.post_attention_layernorm.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.attention.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.attention.masked_bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.attention.rotary_emb.inv_freq": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.attention.query_key_value.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.attention.query_key_value.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.attention.dense.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.attention.dense.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.mlp.dense_h_to_4h.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.mlp.dense_h_to_4h.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.mlp.dense_4h_to_h.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.28.mlp.dense_4h_to_h.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.input_layernorm.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.input_layernorm.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.post_attention_layernorm.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.post_attention_layernorm.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.attention.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.attention.masked_bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.attention.rotary_emb.inv_freq": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.attention.query_key_value.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.attention.query_key_value.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.attention.dense.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.attention.dense.bias": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.mlp.dense_h_to_4h.weight": "hf_ckpt_15.pt",
"model.gpt_neox.layers.29.mlp.dense_h_to_4h.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.29.mlp.dense_4h_to_h.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.29.mlp.dense_4h_to_h.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.input_layernorm.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.input_layernorm.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.post_attention_layernorm.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.post_attention_layernorm.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.attention.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.attention.masked_bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.attention.rotary_emb.inv_freq": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.attention.query_key_value.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.attention.query_key_value.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.attention.dense.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.attention.dense.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.mlp.dense_h_to_4h.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.mlp.dense_h_to_4h.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.mlp.dense_4h_to_h.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.30.mlp.dense_4h_to_h.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.input_layernorm.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.input_layernorm.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.post_attention_layernorm.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.post_attention_layernorm.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.attention.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.attention.masked_bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.attention.rotary_emb.inv_freq": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.attention.query_key_value.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.attention.query_key_value.bias": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.attention.dense.weight": "hf_ckpt_16.pt",
"model.gpt_neox.layers.31.attention.dense.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.31.mlp.dense_h_to_4h.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.31.mlp.dense_h_to_4h.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.31.mlp.dense_4h_to_h.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.31.mlp.dense_4h_to_h.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.input_layernorm.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.input_layernorm.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.post_attention_layernorm.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.post_attention_layernorm.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.attention.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.attention.masked_bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.attention.rotary_emb.inv_freq": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.attention.query_key_value.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.attention.query_key_value.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.attention.dense.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.attention.dense.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.mlp.dense_h_to_4h.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.mlp.dense_h_to_4h.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.mlp.dense_4h_to_h.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.32.mlp.dense_4h_to_h.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.33.input_layernorm.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.33.input_layernorm.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.33.post_attention_layernorm.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.33.post_attention_layernorm.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.33.attention.bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.33.attention.masked_bias": "hf_ckpt_17.pt",
"model.gpt_neox.layers.33.attention.rotary_emb.inv_freq": "hf_ckpt_17.pt",
"model.gpt_neox.layers.33.attention.query_key_value.weight": "hf_ckpt_17.pt",
"model.gpt_neox.layers.33.attention.query_key_value.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.33.attention.dense.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.33.attention.dense.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.33.mlp.dense_h_to_4h.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.33.mlp.dense_h_to_4h.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.33.mlp.dense_4h_to_h.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.33.mlp.dense_4h_to_h.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.input_layernorm.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.input_layernorm.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.post_attention_layernorm.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.post_attention_layernorm.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.attention.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.attention.masked_bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.attention.rotary_emb.inv_freq": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.attention.query_key_value.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.attention.query_key_value.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.attention.dense.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.attention.dense.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.mlp.dense_h_to_4h.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.mlp.dense_h_to_4h.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.mlp.dense_4h_to_h.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.34.mlp.dense_4h_to_h.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.35.input_layernorm.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.35.input_layernorm.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.35.post_attention_layernorm.weight": "hf_ckpt_18.pt",
"model.gpt_neox.layers.35.post_attention_layernorm.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.35.attention.bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.35.attention.masked_bias": "hf_ckpt_18.pt",
"model.gpt_neox.layers.35.attention.rotary_emb.inv_freq": "hf_ckpt_19.pt",
"model.gpt_neox.layers.35.attention.query_key_value.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.35.attention.query_key_value.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.35.attention.dense.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.35.attention.dense.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.35.mlp.dense_h_to_4h.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.35.mlp.dense_h_to_4h.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.35.mlp.dense_4h_to_h.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.35.mlp.dense_4h_to_h.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.input_layernorm.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.input_layernorm.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.post_attention_layernorm.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.post_attention_layernorm.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.attention.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.attention.masked_bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.attention.rotary_emb.inv_freq": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.attention.query_key_value.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.attention.query_key_value.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.attention.dense.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.attention.dense.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.mlp.dense_h_to_4h.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.mlp.dense_h_to_4h.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.mlp.dense_4h_to_h.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.36.mlp.dense_4h_to_h.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.37.input_layernorm.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.37.input_layernorm.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.37.post_attention_layernorm.weight": "hf_ckpt_19.pt",
"model.gpt_neox.layers.37.post_attention_layernorm.bias": "hf_ckpt_19.pt",
"model.gpt_neox.layers.37.attention.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.37.attention.masked_bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.37.attention.rotary_emb.inv_freq": "hf_ckpt_20.pt",
"model.gpt_neox.layers.37.attention.query_key_value.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.37.attention.query_key_value.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.37.attention.dense.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.37.attention.dense.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.37.mlp.dense_h_to_4h.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.37.mlp.dense_h_to_4h.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.37.mlp.dense_4h_to_h.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.37.mlp.dense_4h_to_h.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.input_layernorm.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.input_layernorm.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.post_attention_layernorm.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.post_attention_layernorm.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.attention.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.attention.masked_bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.attention.rotary_emb.inv_freq": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.attention.query_key_value.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.attention.query_key_value.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.attention.dense.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.attention.dense.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.mlp.dense_h_to_4h.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.mlp.dense_h_to_4h.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.mlp.dense_4h_to_h.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.38.mlp.dense_4h_to_h.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.39.input_layernorm.weight": "hf_ckpt_20.pt",
"model.gpt_neox.layers.39.input_layernorm.bias": "hf_ckpt_20.pt",
"model.gpt_neox.layers.39.post_attention_layernorm.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.post_attention_layernorm.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.attention.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.attention.masked_bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.attention.rotary_emb.inv_freq": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.attention.query_key_value.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.attention.query_key_value.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.attention.dense.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.attention.dense.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.mlp.dense_h_to_4h.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.mlp.dense_h_to_4h.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.mlp.dense_4h_to_h.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.39.mlp.dense_4h_to_h.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.input_layernorm.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.input_layernorm.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.post_attention_layernorm.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.post_attention_layernorm.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.attention.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.attention.masked_bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.attention.rotary_emb.inv_freq": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.attention.query_key_value.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.attention.query_key_value.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.attention.dense.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.attention.dense.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.mlp.dense_h_to_4h.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.mlp.dense_h_to_4h.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.mlp.dense_4h_to_h.weight": "hf_ckpt_21.pt",
"model.gpt_neox.layers.40.mlp.dense_4h_to_h.bias": "hf_ckpt_21.pt",
"model.gpt_neox.layers.41.input_layernorm.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.input_layernorm.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.post_attention_layernorm.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.post_attention_layernorm.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.attention.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.attention.masked_bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.attention.rotary_emb.inv_freq": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.attention.query_key_value.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.attention.query_key_value.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.attention.dense.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.attention.dense.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.mlp.dense_h_to_4h.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.mlp.dense_h_to_4h.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.mlp.dense_4h_to_h.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.41.mlp.dense_4h_to_h.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.input_layernorm.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.input_layernorm.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.post_attention_layernorm.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.post_attention_layernorm.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.attention.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.attention.masked_bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.attention.rotary_emb.inv_freq": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.attention.query_key_value.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.attention.query_key_value.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.attention.dense.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.attention.dense.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.mlp.dense_h_to_4h.weight": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.mlp.dense_h_to_4h.bias": "hf_ckpt_22.pt",
"model.gpt_neox.layers.42.mlp.dense_4h_to_h.weight": "hf_ckpt_23.pt",
"model.gpt_neox.layers.42.mlp.dense_4h_to_h.bias": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.input_layernorm.weight": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.input_layernorm.bias": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.post_attention_layernorm.weight": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.post_attention_layernorm.bias": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.attention.bias": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.attention.masked_bias": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.attention.rotary_emb.inv_freq": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.attention.query_key_value.weight": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.attention.query_key_value.bias": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.attention.dense.weight": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.attention.dense.bias": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.mlp.dense_h_to_4h.weight": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.mlp.dense_h_to_4h.bias": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.mlp.dense_4h_to_h.weight": "hf_ckpt_23.pt",
"model.gpt_neox.layers.43.mlp.dense_4h_to_h.bias": "hf_ckpt_23.pt",
"model.gpt_neox.final_layer_norm.weight": "hf_ckpt_23.pt",
"model.gpt_neox.final_layer_norm.bias": "hf_ckpt_23.pt",
"model.embed_out.weight": "hf_ckpt_23.pt",
"transformer.embed_in.weight": "hf_ckpt_23.pt",
"transformer.layers.0.input_layernorm.weight": "hf_ckpt_23.pt",
"transformer.layers.0.input_layernorm.bias": "hf_ckpt_23.pt",
"transformer.layers.0.post_attention_layernorm.weight": "hf_ckpt_23.pt",
"transformer.layers.0.post_attention_layernorm.bias": "hf_ckpt_23.pt",
"transformer.layers.0.attention.bias": "hf_ckpt_23.pt",
"transformer.layers.0.attention.masked_bias": "hf_ckpt_23.pt",
"transformer.layers.0.attention.rotary_emb.inv_freq": "hf_ckpt_23.pt",
"transformer.layers.0.attention.query_key_value.weight": "hf_ckpt_24.pt",
"transformer.layers.0.attention.query_key_value.bias": "hf_ckpt_24.pt",
"transformer.layers.0.attention.dense.weight": "hf_ckpt_24.pt",
"transformer.layers.0.attention.dense.bias": "hf_ckpt_24.pt",
"transformer.layers.0.mlp.dense_h_to_4h.weight": "hf_ckpt_24.pt",
"transformer.layers.0.mlp.dense_h_to_4h.bias": "hf_ckpt_24.pt",
"transformer.layers.0.mlp.dense_4h_to_h.weight": "hf_ckpt_24.pt",
"transformer.layers.0.mlp.dense_4h_to_h.bias": "hf_ckpt_24.pt",
"transformer.layers.1.input_layernorm.weight": "hf_ckpt_24.pt",
"transformer.layers.1.input_layernorm.bias": "hf_ckpt_24.pt",
"transformer.layers.1.post_attention_layernorm.weight": "hf_ckpt_24.pt",
"transformer.layers.1.post_attention_layernorm.bias": "hf_ckpt_24.pt",
"transformer.layers.1.attention.bias": "hf_ckpt_24.pt",
"transformer.layers.1.attention.masked_bias": "hf_ckpt_24.pt",
"transformer.layers.1.attention.rotary_emb.inv_freq": "hf_ckpt_24.pt",
"transformer.layers.1.attention.query_key_value.weight": "hf_ckpt_24.pt",
"transformer.layers.1.attention.query_key_value.bias": "hf_ckpt_24.pt",
"transformer.layers.1.attention.dense.weight": "hf_ckpt_24.pt",
"transformer.layers.1.attention.dense.bias": "hf_ckpt_24.pt",
"transformer.layers.1.mlp.dense_h_to_4h.weight": "hf_ckpt_24.pt",
"transformer.layers.1.mlp.dense_h_to_4h.bias": "hf_ckpt_24.pt",
"transformer.layers.1.mlp.dense_4h_to_h.weight": "hf_ckpt_24.pt",
"transformer.layers.1.mlp.dense_4h_to_h.bias": "hf_ckpt_24.pt",
"transformer.layers.2.input_layernorm.weight": "hf_ckpt_24.pt",
"transformer.layers.2.input_layernorm.bias": "hf_ckpt_24.pt",
"transformer.layers.2.post_attention_layernorm.weight": "hf_ckpt_24.pt",
"transformer.layers.2.post_attention_layernorm.bias": "hf_ckpt_24.pt",
"transformer.layers.2.attention.bias": "hf_ckpt_24.pt",
"transformer.layers.2.attention.masked_bias": "hf_ckpt_25.pt",
"transformer.layers.2.attention.rotary_emb.inv_freq": "hf_ckpt_25.pt",
"transformer.layers.2.attention.query_key_value.weight": "hf_ckpt_25.pt",
"transformer.layers.2.attention.query_key_value.bias": "hf_ckpt_25.pt",
"transformer.layers.2.attention.dense.weight": "hf_ckpt_25.pt",
"transformer.layers.2.attention.dense.bias": "hf_ckpt_25.pt",
"transformer.layers.2.mlp.dense_h_to_4h.weight": "hf_ckpt_25.pt",
"transformer.layers.2.mlp.dense_h_to_4h.bias": "hf_ckpt_25.pt",
"transformer.layers.2.mlp.dense_4h_to_h.weight": "hf_ckpt_25.pt",
"transformer.layers.2.mlp.dense_4h_to_h.bias": "hf_ckpt_25.pt",
"transformer.layers.3.input_layernorm.weight": "hf_ckpt_25.pt",
"transformer.layers.3.input_layernorm.bias": "hf_ckpt_25.pt",
"transformer.layers.3.post_attention_layernorm.weight": "hf_ckpt_25.pt",
"transformer.layers.3.post_attention_layernorm.bias": "hf_ckpt_25.pt",
"transformer.layers.3.attention.bias": "hf_ckpt_25.pt",
"transformer.layers.3.attention.masked_bias": "hf_ckpt_25.pt",
"transformer.layers.3.attention.rotary_emb.inv_freq": "hf_ckpt_25.pt",
"transformer.layers.3.attention.query_key_value.weight": "hf_ckpt_25.pt",
"transformer.layers.3.attention.query_key_value.bias": "hf_ckpt_25.pt",
"transformer.layers.3.attention.dense.weight": "hf_ckpt_25.pt",
"transformer.layers.3.attention.dense.bias": "hf_ckpt_25.pt",
"transformer.layers.3.mlp.dense_h_to_4h.weight": "hf_ckpt_25.pt",
"transformer.layers.3.mlp.dense_h_to_4h.bias": "hf_ckpt_25.pt",
"transformer.layers.3.mlp.dense_4h_to_h.weight": "hf_ckpt_25.pt",
"transformer.layers.3.mlp.dense_4h_to_h.bias": "hf_ckpt_25.pt",
"transformer.layers.4.input_layernorm.weight": "hf_ckpt_25.pt",
"transformer.layers.4.input_layernorm.bias": "hf_ckpt_25.pt",
"transformer.layers.4.post_attention_layernorm.weight": "hf_ckpt_25.pt",
"transformer.layers.4.post_attention_layernorm.bias": "hf_ckpt_26.pt",
"transformer.layers.4.attention.bias": "hf_ckpt_26.pt",
"transformer.layers.4.attention.masked_bias": "hf_ckpt_26.pt",
"transformer.layers.4.attention.rotary_emb.inv_freq": "hf_ckpt_26.pt",
"transformer.layers.4.attention.query_key_value.weight": "hf_ckpt_26.pt",
"transformer.layers.4.attention.query_key_value.bias": "hf_ckpt_26.pt",
"transformer.layers.4.attention.dense.weight": "hf_ckpt_26.pt",
"transformer.layers.4.attention.dense.bias": "hf_ckpt_26.pt",
"transformer.layers.4.mlp.dense_h_to_4h.weight": "hf_ckpt_26.pt",
"transformer.layers.4.mlp.dense_h_to_4h.bias": "hf_ckpt_26.pt",
"transformer.layers.4.mlp.dense_4h_to_h.weight": "hf_ckpt_26.pt",
"transformer.layers.4.mlp.dense_4h_to_h.bias": "hf_ckpt_26.pt",
"transformer.layers.5.input_layernorm.weight": "hf_ckpt_26.pt",
"transformer.layers.5.input_layernorm.bias": "hf_ckpt_26.pt",
"transformer.layers.5.post_attention_layernorm.weight": "hf_ckpt_26.pt",
"transformer.layers.5.post_attention_layernorm.bias": "hf_ckpt_26.pt",
"transformer.layers.5.attention.bias": "hf_ckpt_26.pt",
"transformer.layers.5.attention.masked_bias": "hf_ckpt_26.pt",
"transformer.layers.5.attention.rotary_emb.inv_freq": "hf_ckpt_26.pt",
"transformer.layers.5.attention.query_key_value.weight": "hf_ckpt_26.pt",
"transformer.layers.5.attention.query_key_value.bias": "hf_ckpt_26.pt",
"transformer.layers.5.attention.dense.weight": "hf_ckpt_26.pt",
"transformer.layers.5.attention.dense.bias": "hf_ckpt_26.pt",
"transformer.layers.5.mlp.dense_h_to_4h.weight": "hf_ckpt_26.pt",
"transformer.layers.5.mlp.dense_h_to_4h.bias": "hf_ckpt_26.pt",
"transformer.layers.5.mlp.dense_4h_to_h.weight": "hf_ckpt_26.pt",
"transformer.layers.5.mlp.dense_4h_to_h.bias": "hf_ckpt_26.pt",
"transformer.layers.6.input_layernorm.weight": "hf_ckpt_26.pt",
"transformer.layers.6.input_layernorm.bias": "hf_ckpt_27.pt",
"transformer.layers.6.post_attention_layernorm.weight": "hf_ckpt_27.pt",
"transformer.layers.6.post_attention_layernorm.bias": "hf_ckpt_27.pt",
"transformer.layers.6.attention.bias": "hf_ckpt_27.pt",
"transformer.layers.6.attention.masked_bias": "hf_ckpt_27.pt",
"transformer.layers.6.attention.rotary_emb.inv_freq": "hf_ckpt_27.pt",
"transformer.layers.6.attention.query_key_value.weight": "hf_ckpt_27.pt",
"transformer.layers.6.attention.query_key_value.bias": "hf_ckpt_27.pt",
"transformer.layers.6.attention.dense.weight": "hf_ckpt_27.pt",
"transformer.layers.6.attention.dense.bias": "hf_ckpt_27.pt",
"transformer.layers.6.mlp.dense_h_to_4h.weight": "hf_ckpt_27.pt",
"transformer.layers.6.mlp.dense_h_to_4h.bias": "hf_ckpt_27.pt",
"transformer.layers.6.mlp.dense_4h_to_h.weight": "hf_ckpt_27.pt",
"transformer.layers.6.mlp.dense_4h_to_h.bias": "hf_ckpt_27.pt",
"transformer.layers.7.input_layernorm.weight": "hf_ckpt_27.pt",
"transformer.layers.7.input_layernorm.bias": "hf_ckpt_27.pt",
"transformer.layers.7.post_attention_layernorm.weight": "hf_ckpt_27.pt",
"transformer.layers.7.post_attention_layernorm.bias": "hf_ckpt_27.pt",
"transformer.layers.7.attention.bias": "hf_ckpt_27.pt",
"transformer.layers.7.attention.masked_bias": "hf_ckpt_27.pt",
"transformer.layers.7.attention.rotary_emb.inv_freq": "hf_ckpt_27.pt",
"transformer.layers.7.attention.query_key_value.weight": "hf_ckpt_27.pt",
"transformer.layers.7.attention.query_key_value.bias": "hf_ckpt_27.pt",
"transformer.layers.7.attention.dense.weight": "hf_ckpt_27.pt",
"transformer.layers.7.attention.dense.bias": "hf_ckpt_27.pt",
"transformer.layers.7.mlp.dense_h_to_4h.weight": "hf_ckpt_27.pt",
"transformer.layers.7.mlp.dense_h_to_4h.bias": "hf_ckpt_27.pt",
"transformer.layers.7.mlp.dense_4h_to_h.weight": "hf_ckpt_27.pt",
"transformer.layers.7.mlp.dense_4h_to_h.bias": "hf_ckpt_28.pt",
"transformer.layers.8.input_layernorm.weight": "hf_ckpt_28.pt",
"transformer.layers.8.input_layernorm.bias": "hf_ckpt_28.pt",
"transformer.layers.8.post_attention_layernorm.weight": "hf_ckpt_28.pt",
"transformer.layers.8.post_attention_layernorm.bias": "hf_ckpt_28.pt",
"transformer.layers.8.attention.bias": "hf_ckpt_28.pt",
"transformer.layers.8.attention.masked_bias": "hf_ckpt_28.pt",
"transformer.layers.8.attention.rotary_emb.inv_freq": "hf_ckpt_28.pt",
"transformer.layers.8.attention.query_key_value.weight": "hf_ckpt_28.pt",
"transformer.layers.8.attention.query_key_value.bias": "hf_ckpt_28.pt",
"transformer.layers.8.attention.dense.weight": "hf_ckpt_28.pt",
"transformer.layers.8.attention.dense.bias": "hf_ckpt_28.pt",
"transformer.layers.8.mlp.dense_h_to_4h.weight": "hf_ckpt_28.pt",
"transformer.layers.8.mlp.dense_h_to_4h.bias": "hf_ckpt_28.pt",
"transformer.layers.8.mlp.dense_4h_to_h.weight": "hf_ckpt_28.pt",
"transformer.layers.8.mlp.dense_4h_to_h.bias": "hf_ckpt_28.pt",
"transformer.layers.9.input_layernorm.weight": "hf_ckpt_28.pt",
"transformer.layers.9.input_layernorm.bias": "hf_ckpt_28.pt",
"transformer.layers.9.post_attention_layernorm.weight": "hf_ckpt_28.pt",
"transformer.layers.9.post_attention_layernorm.bias": "hf_ckpt_28.pt",
"transformer.layers.9.attention.bias": "hf_ckpt_28.pt",
"transformer.layers.9.attention.masked_bias": "hf_ckpt_28.pt",
"transformer.layers.9.attention.rotary_emb.inv_freq": "hf_ckpt_28.pt",
"transformer.layers.9.attention.query_key_value.weight": "hf_ckpt_28.pt",
"transformer.layers.9.attention.query_key_value.bias": "hf_ckpt_28.pt",
"transformer.layers.9.attention.dense.weight": "hf_ckpt_28.pt",
"transformer.layers.9.attention.dense.bias": "hf_ckpt_28.pt",
"transformer.layers.9.mlp.dense_h_to_4h.weight": "hf_ckpt_28.pt",
"transformer.layers.9.mlp.dense_h_to_4h.bias": "hf_ckpt_29.pt",
"transformer.layers.9.mlp.dense_4h_to_h.weight": "hf_ckpt_29.pt",
"transformer.layers.9.mlp.dense_4h_to_h.bias": "hf_ckpt_29.pt",
"transformer.layers.10.input_layernorm.weight": "hf_ckpt_29.pt",
"transformer.layers.10.input_layernorm.bias": "hf_ckpt_29.pt",
"transformer.layers.10.post_attention_layernorm.weight": "hf_ckpt_29.pt",
"transformer.layers.10.post_attention_layernorm.bias": "hf_ckpt_29.pt",
"transformer.layers.10.attention.bias": "hf_ckpt_29.pt",
"transformer.layers.10.attention.masked_bias": "hf_ckpt_29.pt",
"transformer.layers.10.attention.rotary_emb.inv_freq": "hf_ckpt_29.pt",
"transformer.layers.10.attention.query_key_value.weight": "hf_ckpt_29.pt",
"transformer.layers.10.attention.query_key_value.bias": "hf_ckpt_29.pt",
"transformer.layers.10.attention.dense.weight": "hf_ckpt_29.pt",
"transformer.layers.10.attention.dense.bias": "hf_ckpt_29.pt",
"transformer.layers.10.mlp.dense_h_to_4h.weight": "hf_ckpt_29.pt",
"transformer.layers.10.mlp.dense_h_to_4h.bias": "hf_ckpt_29.pt",
"transformer.layers.10.mlp.dense_4h_to_h.weight": "hf_ckpt_29.pt",
"transformer.layers.10.mlp.dense_4h_to_h.bias": "hf_ckpt_29.pt",
"transformer.layers.11.input_layernorm.weight": "hf_ckpt_29.pt",
"transformer.layers.11.input_layernorm.bias": "hf_ckpt_29.pt",
"transformer.layers.11.post_attention_layernorm.weight": "hf_ckpt_29.pt",
"transformer.layers.11.post_attention_layernorm.bias": "hf_ckpt_29.pt",
"transformer.layers.11.attention.bias": "hf_ckpt_29.pt",
"transformer.layers.11.attention.masked_bias": "hf_ckpt_29.pt",
"transformer.layers.11.attention.rotary_emb.inv_freq": "hf_ckpt_29.pt",
"transformer.layers.11.attention.query_key_value.weight": "hf_ckpt_29.pt",
"transformer.layers.11.attention.query_key_value.bias": "hf_ckpt_29.pt",
"transformer.layers.11.attention.dense.weight": "hf_ckpt_29.pt",
"transformer.layers.11.attention.dense.bias": "hf_ckpt_30.pt",
"transformer.layers.11.mlp.dense_h_to_4h.weight": "hf_ckpt_30.pt",
"transformer.layers.11.mlp.dense_h_to_4h.bias": "hf_ckpt_30.pt",
"transformer.layers.11.mlp.dense_4h_to_h.weight": "hf_ckpt_30.pt",
"transformer.layers.11.mlp.dense_4h_to_h.bias": "hf_ckpt_30.pt",
"transformer.layers.12.input_layernorm.weight": "hf_ckpt_30.pt",
"transformer.layers.12.input_layernorm.bias": "hf_ckpt_30.pt",
"transformer.layers.12.post_attention_layernorm.weight": "hf_ckpt_30.pt",
"transformer.layers.12.post_attention_layernorm.bias": "hf_ckpt_30.pt",
"transformer.layers.12.attention.bias": "hf_ckpt_30.pt",
"transformer.layers.12.attention.masked_bias": "hf_ckpt_30.pt",
"transformer.layers.12.attention.rotary_emb.inv_freq": "hf_ckpt_30.pt",
"transformer.layers.12.attention.query_key_value.weight": "hf_ckpt_30.pt",
"transformer.layers.12.attention.query_key_value.bias": "hf_ckpt_30.pt",
"transformer.layers.12.attention.dense.weight": "hf_ckpt_30.pt",
"transformer.layers.12.attention.dense.bias": "hf_ckpt_30.pt",
"transformer.layers.12.mlp.dense_h_to_4h.weight": "hf_ckpt_30.pt",
"transformer.layers.12.mlp.dense_h_to_4h.bias": "hf_ckpt_30.pt",
"transformer.layers.12.mlp.dense_4h_to_h.weight": "hf_ckpt_30.pt",
"transformer.layers.12.mlp.dense_4h_to_h.bias": "hf_ckpt_30.pt",
"transformer.layers.13.input_layernorm.weight": "hf_ckpt_30.pt",
"transformer.layers.13.input_layernorm.bias": "hf_ckpt_30.pt",
"transformer.layers.13.post_attention_layernorm.weight": "hf_ckpt_30.pt",
"transformer.layers.13.post_attention_layernorm.bias": "hf_ckpt_30.pt",
"transformer.layers.13.attention.bias": "hf_ckpt_30.pt",
"transformer.layers.13.attention.masked_bias": "hf_ckpt_30.pt",
"transformer.layers.13.attention.rotary_emb.inv_freq": "hf_ckpt_30.pt",
"transformer.layers.13.attention.query_key_value.weight": "hf_ckpt_30.pt",
"transformer.layers.13.attention.query_key_value.bias": "hf_ckpt_31.pt",
"transformer.layers.13.attention.dense.weight": "hf_ckpt_31.pt",
"transformer.layers.13.attention.dense.bias": "hf_ckpt_31.pt",
"transformer.layers.13.mlp.dense_h_to_4h.weight": "hf_ckpt_31.pt",
"transformer.layers.13.mlp.dense_h_to_4h.bias": "hf_ckpt_31.pt",
"transformer.layers.13.mlp.dense_4h_to_h.weight": "hf_ckpt_31.pt",
"transformer.layers.13.mlp.dense_4h_to_h.bias": "hf_ckpt_31.pt",
"transformer.layers.14.input_layernorm.weight": "hf_ckpt_31.pt",
"transformer.layers.14.input_layernorm.bias": "hf_ckpt_31.pt",
"transformer.layers.14.post_attention_layernorm.weight": "hf_ckpt_31.pt",
"transformer.layers.14.post_attention_layernorm.bias": "hf_ckpt_31.pt",
"transformer.layers.14.attention.bias": "hf_ckpt_31.pt",
"transformer.layers.14.attention.masked_bias": "hf_ckpt_31.pt",
"transformer.layers.14.attention.rotary_emb.inv_freq": "hf_ckpt_31.pt",
"transformer.layers.14.attention.query_key_value.weight": "hf_ckpt_31.pt",
"transformer.layers.14.attention.query_key_value.bias": "hf_ckpt_31.pt",
"transformer.layers.14.attention.dense.weight": "hf_ckpt_31.pt",
"transformer.layers.14.attention.dense.bias": "hf_ckpt_31.pt",
"transformer.layers.14.mlp.dense_h_to_4h.weight": "hf_ckpt_31.pt",
"transformer.layers.14.mlp.dense_h_to_4h.bias": "hf_ckpt_31.pt",
"transformer.layers.14.mlp.dense_4h_to_h.weight": "hf_ckpt_31.pt",
"transformer.layers.14.mlp.dense_4h_to_h.bias": "hf_ckpt_31.pt",
"transformer.layers.15.input_layernorm.weight": "hf_ckpt_31.pt",
"transformer.layers.15.input_layernorm.bias": "hf_ckpt_31.pt",
"transformer.layers.15.post_attention_layernorm.weight": "hf_ckpt_31.pt",
"transformer.layers.15.post_attention_layernorm.bias": "hf_ckpt_31.pt",
"transformer.layers.15.attention.bias": "hf_ckpt_31.pt",
"transformer.layers.15.attention.masked_bias": "hf_ckpt_31.pt",
"transformer.layers.15.attention.rotary_emb.inv_freq": "hf_ckpt_32.pt",
"transformer.layers.15.attention.query_key_value.weight": "hf_ckpt_32.pt",
"transformer.layers.15.attention.query_key_value.bias": "hf_ckpt_32.pt",
"transformer.layers.15.attention.dense.weight": "hf_ckpt_32.pt",
"transformer.layers.15.attention.dense.bias": "hf_ckpt_32.pt",
"transformer.layers.15.mlp.dense_h_to_4h.weight": "hf_ckpt_32.pt",
"transformer.layers.15.mlp.dense_h_to_4h.bias": "hf_ckpt_32.pt",
"transformer.layers.15.mlp.dense_4h_to_h.weight": "hf_ckpt_32.pt",
"transformer.layers.15.mlp.dense_4h_to_h.bias": "hf_ckpt_32.pt",
"transformer.layers.16.input_layernorm.weight": "hf_ckpt_32.pt",
"transformer.layers.16.input_layernorm.bias": "hf_ckpt_32.pt",
"transformer.layers.16.post_attention_layernorm.weight": "hf_ckpt_32.pt",
"transformer.layers.16.post_attention_layernorm.bias": "hf_ckpt_32.pt",
"transformer.layers.16.attention.bias": "hf_ckpt_32.pt",
"transformer.layers.16.attention.masked_bias": "hf_ckpt_32.pt",
"transformer.layers.16.attention.rotary_emb.inv_freq": "hf_ckpt_32.pt",
"transformer.layers.16.attention.query_key_value.weight": "hf_ckpt_32.pt",
"transformer.layers.16.attention.query_key_value.bias": "hf_ckpt_32.pt",
"transformer.layers.16.attention.dense.weight": "hf_ckpt_32.pt",
"transformer.layers.16.attention.dense.bias": "hf_ckpt_32.pt",
"transformer.layers.16.mlp.dense_h_to_4h.weight": "hf_ckpt_32.pt",
"transformer.layers.16.mlp.dense_h_to_4h.bias": "hf_ckpt_32.pt",
"transformer.layers.16.mlp.dense_4h_to_h.weight": "hf_ckpt_32.pt",
"transformer.layers.16.mlp.dense_4h_to_h.bias": "hf_ckpt_32.pt",
"transformer.layers.17.input_layernorm.weight": "hf_ckpt_32.pt",
"transformer.layers.17.input_layernorm.bias": "hf_ckpt_32.pt",
"transformer.layers.17.post_attention_layernorm.weight": "hf_ckpt_32.pt",
"transformer.layers.17.post_attention_layernorm.bias": "hf_ckpt_32.pt",
"transformer.layers.17.attention.bias": "hf_ckpt_33.pt",
"transformer.layers.17.attention.masked_bias": "hf_ckpt_33.pt",
"transformer.layers.17.attention.rotary_emb.inv_freq": "hf_ckpt_33.pt",
"transformer.layers.17.attention.query_key_value.weight": "hf_ckpt_33.pt",
"transformer.layers.17.attention.query_key_value.bias": "hf_ckpt_33.pt",
"transformer.layers.17.attention.dense.weight": "hf_ckpt_33.pt",
"transformer.layers.17.attention.dense.bias": "hf_ckpt_33.pt",
"transformer.layers.17.mlp.dense_h_to_4h.weight": "hf_ckpt_33.pt",
"transformer.layers.17.mlp.dense_h_to_4h.bias": "hf_ckpt_33.pt",
"transformer.layers.17.mlp.dense_4h_to_h.weight": "hf_ckpt_33.pt",
"transformer.layers.17.mlp.dense_4h_to_h.bias": "hf_ckpt_33.pt",
"transformer.layers.18.input_layernorm.weight": "hf_ckpt_33.pt",
"transformer.layers.18.input_layernorm.bias": "hf_ckpt_33.pt",
"transformer.layers.18.post_attention_layernorm.weight": "hf_ckpt_33.pt",
"transformer.layers.18.post_attention_layernorm.bias": "hf_ckpt_33.pt",
"transformer.layers.18.attention.bias": "hf_ckpt_33.pt",
"transformer.layers.18.attention.masked_bias": "hf_ckpt_33.pt",
"transformer.layers.18.attention.rotary_emb.inv_freq": "hf_ckpt_33.pt",
"transformer.layers.18.attention.query_key_value.weight": "hf_ckpt_33.pt",
"transformer.layers.18.attention.query_key_value.bias": "hf_ckpt_33.pt",
"transformer.layers.18.attention.dense.weight": "hf_ckpt_33.pt",
"transformer.layers.18.attention.dense.bias": "hf_ckpt_33.pt",
"transformer.layers.18.mlp.dense_h_to_4h.weight": "hf_ckpt_33.pt",
"transformer.layers.18.mlp.dense_h_to_4h.bias": "hf_ckpt_33.pt",
"transformer.layers.18.mlp.dense_4h_to_h.weight": "hf_ckpt_33.pt",
"transformer.layers.18.mlp.dense_4h_to_h.bias": "hf_ckpt_33.pt",
"transformer.layers.19.input_layernorm.weight": "hf_ckpt_33.pt",
"transformer.layers.19.input_layernorm.bias": "hf_ckpt_33.pt",
"transformer.layers.19.post_attention_layernorm.weight": "hf_ckpt_34.pt",
"transformer.layers.19.post_attention_layernorm.bias": "hf_ckpt_34.pt",
"transformer.layers.19.attention.bias": "hf_ckpt_34.pt",
"transformer.layers.19.attention.masked_bias": "hf_ckpt_34.pt",
"transformer.layers.19.attention.rotary_emb.inv_freq": "hf_ckpt_34.pt",
"transformer.layers.19.attention.query_key_value.weight": "hf_ckpt_34.pt",
"transformer.layers.19.attention.query_key_value.bias": "hf_ckpt_34.pt",
"transformer.layers.19.attention.dense.weight": "hf_ckpt_34.pt",
"transformer.layers.19.attention.dense.bias": "hf_ckpt_34.pt",
"transformer.layers.19.mlp.dense_h_to_4h.weight": "hf_ckpt_34.pt",
"transformer.layers.19.mlp.dense_h_to_4h.bias": "hf_ckpt_34.pt",
"transformer.layers.19.mlp.dense_4h_to_h.weight": "hf_ckpt_34.pt",
"transformer.layers.19.mlp.dense_4h_to_h.bias": "hf_ckpt_34.pt",
"transformer.layers.20.input_layernorm.weight": "hf_ckpt_34.pt",
"transformer.layers.20.input_layernorm.bias": "hf_ckpt_34.pt",
"transformer.layers.20.post_attention_layernorm.weight": "hf_ckpt_34.pt",
"transformer.layers.20.post_attention_layernorm.bias": "hf_ckpt_34.pt",
"transformer.layers.20.attention.bias": "hf_ckpt_34.pt",
"transformer.layers.20.attention.masked_bias": "hf_ckpt_34.pt",
"transformer.layers.20.attention.rotary_emb.inv_freq": "hf_ckpt_34.pt",
"transformer.layers.20.attention.query_key_value.weight": "hf_ckpt_34.pt",
"transformer.layers.20.attention.query_key_value.bias": "hf_ckpt_34.pt",
"transformer.layers.20.attention.dense.weight": "hf_ckpt_34.pt",
"transformer.layers.20.attention.dense.bias": "hf_ckpt_34.pt",
"transformer.layers.20.mlp.dense_h_to_4h.weight": "hf_ckpt_34.pt",
"transformer.layers.20.mlp.dense_h_to_4h.bias": "hf_ckpt_34.pt",
"transformer.layers.20.mlp.dense_4h_to_h.weight": "hf_ckpt_34.pt",
"transformer.layers.20.mlp.dense_4h_to_h.bias": "hf_ckpt_34.pt",
"transformer.layers.21.input_layernorm.weight": "hf_ckpt_35.pt",
"transformer.layers.21.input_layernorm.bias": "hf_ckpt_35.pt",
"transformer.layers.21.post_attention_layernorm.weight": "hf_ckpt_35.pt",
"transformer.layers.21.post_attention_layernorm.bias": "hf_ckpt_35.pt",
"transformer.layers.21.attention.bias": "hf_ckpt_35.pt",
"transformer.layers.21.attention.masked_bias": "hf_ckpt_35.pt",
"transformer.layers.21.attention.rotary_emb.inv_freq": "hf_ckpt_35.pt",
"transformer.layers.21.attention.query_key_value.weight": "hf_ckpt_35.pt",
"transformer.layers.21.attention.query_key_value.bias": "hf_ckpt_35.pt",
"transformer.layers.21.attention.dense.weight": "hf_ckpt_35.pt",
"transformer.layers.21.attention.dense.bias": "hf_ckpt_35.pt",
"transformer.layers.21.mlp.dense_h_to_4h.weight": "hf_ckpt_35.pt",
"transformer.layers.21.mlp.dense_h_to_4h.bias": "hf_ckpt_35.pt",
"transformer.layers.21.mlp.dense_4h_to_h.weight": "hf_ckpt_35.pt",
"transformer.layers.21.mlp.dense_4h_to_h.bias": "hf_ckpt_35.pt",
"transformer.layers.22.input_layernorm.weight": "hf_ckpt_35.pt",
"transformer.layers.22.input_layernorm.bias": "hf_ckpt_35.pt",
"transformer.layers.22.post_attention_layernorm.weight": "hf_ckpt_35.pt",
"transformer.layers.22.post_attention_layernorm.bias": "hf_ckpt_35.pt",
"transformer.layers.22.attention.bias": "hf_ckpt_35.pt",
"transformer.layers.22.attention.masked_bias": "hf_ckpt_35.pt",
"transformer.layers.22.attention.rotary_emb.inv_freq": "hf_ckpt_35.pt",
"transformer.layers.22.attention.query_key_value.weight": "hf_ckpt_35.pt",
"transformer.layers.22.attention.query_key_value.bias": "hf_ckpt_35.pt",
"transformer.layers.22.attention.dense.weight": "hf_ckpt_35.pt",
"transformer.layers.22.attention.dense.bias": "hf_ckpt_35.pt",
"transformer.layers.22.mlp.dense_h_to_4h.weight": "hf_ckpt_35.pt",
"transformer.layers.22.mlp.dense_h_to_4h.bias": "hf_ckpt_35.pt",
"transformer.layers.22.mlp.dense_4h_to_h.weight": "hf_ckpt_36.pt",
"transformer.layers.22.mlp.dense_4h_to_h.bias": "hf_ckpt_36.pt",
"transformer.layers.23.input_layernorm.weight": "hf_ckpt_36.pt",
"transformer.layers.23.input_layernorm.bias": "hf_ckpt_36.pt",
"transformer.layers.23.post_attention_layernorm.weight": "hf_ckpt_36.pt",
"transformer.layers.23.post_attention_layernorm.bias": "hf_ckpt_36.pt",
"transformer.layers.23.attention.bias": "hf_ckpt_36.pt",
"transformer.layers.23.attention.masked_bias": "hf_ckpt_36.pt",
"transformer.layers.23.attention.rotary_emb.inv_freq": "hf_ckpt_36.pt",
"transformer.layers.23.attention.query_key_value.weight": "hf_ckpt_36.pt",
"transformer.layers.23.attention.query_key_value.bias": "hf_ckpt_36.pt",
"transformer.layers.23.attention.dense.weight": "hf_ckpt_36.pt",
"transformer.layers.23.attention.dense.bias": "hf_ckpt_36.pt",
"transformer.layers.23.mlp.dense_h_to_4h.weight": "hf_ckpt_36.pt",
"transformer.layers.23.mlp.dense_h_to_4h.bias": "hf_ckpt_36.pt",
"transformer.layers.23.mlp.dense_4h_to_h.weight": "hf_ckpt_36.pt",
"transformer.layers.23.mlp.dense_4h_to_h.bias": "hf_ckpt_36.pt",
"transformer.layers.24.input_layernorm.weight": "hf_ckpt_36.pt",
"transformer.layers.24.input_layernorm.bias": "hf_ckpt_36.pt",
"transformer.layers.24.post_attention_layernorm.weight": "hf_ckpt_36.pt",
"transformer.layers.24.post_attention_layernorm.bias": "hf_ckpt_36.pt",
"transformer.layers.24.attention.bias": "hf_ckpt_36.pt",
"transformer.layers.24.attention.masked_bias": "hf_ckpt_36.pt",
"transformer.layers.24.attention.rotary_emb.inv_freq": "hf_ckpt_36.pt",
"transformer.layers.24.attention.query_key_value.weight": "hf_ckpt_36.pt",
"transformer.layers.24.attention.query_key_value.bias": "hf_ckpt_36.pt",
"transformer.layers.24.attention.dense.weight": "hf_ckpt_36.pt",
"transformer.layers.24.attention.dense.bias": "hf_ckpt_36.pt",
"transformer.layers.24.mlp.dense_h_to_4h.weight": "hf_ckpt_37.pt",
"transformer.layers.24.mlp.dense_h_to_4h.bias": "hf_ckpt_37.pt",
"transformer.layers.24.mlp.dense_4h_to_h.weight": "hf_ckpt_37.pt",
"transformer.layers.24.mlp.dense_4h_to_h.bias": "hf_ckpt_37.pt",
"transformer.layers.25.input_layernorm.weight": "hf_ckpt_37.pt",
"transformer.layers.25.input_layernorm.bias": "hf_ckpt_37.pt",
"transformer.layers.25.post_attention_layernorm.weight": "hf_ckpt_37.pt",
"transformer.layers.25.post_attention_layernorm.bias": "hf_ckpt_37.pt",
"transformer.layers.25.attention.bias": "hf_ckpt_37.pt",
"transformer.layers.25.attention.masked_bias": "hf_ckpt_37.pt",
"transformer.layers.25.attention.rotary_emb.inv_freq": "hf_ckpt_37.pt",
"transformer.layers.25.attention.query_key_value.weight": "hf_ckpt_37.pt",
"transformer.layers.25.attention.query_key_value.bias": "hf_ckpt_37.pt",
"transformer.layers.25.attention.dense.weight": "hf_ckpt_37.pt",
"transformer.layers.25.attention.dense.bias": "hf_ckpt_37.pt",
"transformer.layers.25.mlp.dense_h_to_4h.weight": "hf_ckpt_37.pt",
"transformer.layers.25.mlp.dense_h_to_4h.bias": "hf_ckpt_37.pt",
"transformer.layers.25.mlp.dense_4h_to_h.weight": "hf_ckpt_37.pt",
"transformer.layers.25.mlp.dense_4h_to_h.bias": "hf_ckpt_37.pt",
"transformer.layers.26.input_layernorm.weight": "hf_ckpt_37.pt",
"transformer.layers.26.input_layernorm.bias": "hf_ckpt_37.pt",
"transformer.layers.26.post_attention_layernorm.weight": "hf_ckpt_37.pt",
"transformer.layers.26.post_attention_layernorm.bias": "hf_ckpt_37.pt",
"transformer.layers.26.attention.bias": "hf_ckpt_37.pt",
"transformer.layers.26.attention.masked_bias": "hf_ckpt_37.pt",
"transformer.layers.26.attention.rotary_emb.inv_freq": "hf_ckpt_37.pt",
"transformer.layers.26.attention.query_key_value.weight": "hf_ckpt_37.pt",
"transformer.layers.26.attention.query_key_value.bias": "hf_ckpt_37.pt",
"transformer.layers.26.attention.dense.weight": "hf_ckpt_38.pt",
"transformer.layers.26.attention.dense.bias": "hf_ckpt_38.pt",
"transformer.layers.26.mlp.dense_h_to_4h.weight": "hf_ckpt_38.pt",
"transformer.layers.26.mlp.dense_h_to_4h.bias": "hf_ckpt_38.pt",
"transformer.layers.26.mlp.dense_4h_to_h.weight": "hf_ckpt_38.pt",
"transformer.layers.26.mlp.dense_4h_to_h.bias": "hf_ckpt_38.pt",
"transformer.layers.27.input_layernorm.weight": "hf_ckpt_38.pt",
"transformer.layers.27.input_layernorm.bias": "hf_ckpt_38.pt",
"transformer.layers.27.post_attention_layernorm.weight": "hf_ckpt_38.pt",
"transformer.layers.27.post_attention_layernorm.bias": "hf_ckpt_38.pt",
"transformer.layers.27.attention.bias": "hf_ckpt_38.pt",
"transformer.layers.27.attention.masked_bias": "hf_ckpt_38.pt",
"transformer.layers.27.attention.rotary_emb.inv_freq": "hf_ckpt_38.pt",
"transformer.layers.27.attention.query_key_value.weight": "hf_ckpt_38.pt",
"transformer.layers.27.attention.query_key_value.bias": "hf_ckpt_38.pt",
"transformer.layers.27.attention.dense.weight": "hf_ckpt_38.pt",
"transformer.layers.27.attention.dense.bias": "hf_ckpt_38.pt",
"transformer.layers.27.mlp.dense_h_to_4h.weight": "hf_ckpt_38.pt",
"transformer.layers.27.mlp.dense_h_to_4h.bias": "hf_ckpt_38.pt",
"transformer.layers.27.mlp.dense_4h_to_h.weight": "hf_ckpt_38.pt",
"transformer.layers.27.mlp.dense_4h_to_h.bias": "hf_ckpt_38.pt",
"transformer.layers.28.input_layernorm.weight": "hf_ckpt_38.pt",
"transformer.layers.28.input_layernorm.bias": "hf_ckpt_38.pt",
"transformer.layers.28.post_attention_layernorm.weight": "hf_ckpt_38.pt",
"transformer.layers.28.post_attention_layernorm.bias": "hf_ckpt_38.pt",
"transformer.layers.28.attention.bias": "hf_ckpt_38.pt",
"transformer.layers.28.attention.masked_bias": "hf_ckpt_38.pt",
"transformer.layers.28.attention.rotary_emb.inv_freq": "hf_ckpt_38.pt",
"transformer.layers.28.attention.query_key_value.weight": "hf_ckpt_39.pt",
"transformer.layers.28.attention.query_key_value.bias": "hf_ckpt_39.pt",
"transformer.layers.28.attention.dense.weight": "hf_ckpt_39.pt",
"transformer.layers.28.attention.dense.bias": "hf_ckpt_39.pt",
"transformer.layers.28.mlp.dense_h_to_4h.weight": "hf_ckpt_39.pt",
"transformer.layers.28.mlp.dense_h_to_4h.bias": "hf_ckpt_39.pt",
"transformer.layers.28.mlp.dense_4h_to_h.weight": "hf_ckpt_39.pt",
"transformer.layers.28.mlp.dense_4h_to_h.bias": "hf_ckpt_39.pt",
"transformer.layers.29.input_layernorm.weight": "hf_ckpt_39.pt",
"transformer.layers.29.input_layernorm.bias": "hf_ckpt_39.pt",
"transformer.layers.29.post_attention_layernorm.weight": "hf_ckpt_39.pt",
"transformer.layers.29.post_attention_layernorm.bias": "hf_ckpt_39.pt",
"transformer.layers.29.attention.bias": "hf_ckpt_39.pt",
"transformer.layers.29.attention.masked_bias": "hf_ckpt_39.pt",
"transformer.layers.29.attention.rotary_emb.inv_freq": "hf_ckpt_39.pt",
"transformer.layers.29.attention.query_key_value.weight": "hf_ckpt_39.pt",
"transformer.layers.29.attention.query_key_value.bias": "hf_ckpt_39.pt",
"transformer.layers.29.attention.dense.weight": "hf_ckpt_39.pt",
"transformer.layers.29.attention.dense.bias": "hf_ckpt_39.pt",
"transformer.layers.29.mlp.dense_h_to_4h.weight": "hf_ckpt_39.pt",
"transformer.layers.29.mlp.dense_h_to_4h.bias": "hf_ckpt_39.pt",
"transformer.layers.29.mlp.dense_4h_to_h.weight": "hf_ckpt_39.pt",
"transformer.layers.29.mlp.dense_4h_to_h.bias": "hf_ckpt_39.pt",
"transformer.layers.30.input_layernorm.weight": "hf_ckpt_39.pt",
"transformer.layers.30.input_layernorm.bias": "hf_ckpt_39.pt",
"transformer.layers.30.post_attention_layernorm.weight": "hf_ckpt_39.pt",
"transformer.layers.30.post_attention_layernorm.bias": "hf_ckpt_39.pt",
"transformer.layers.30.attention.bias": "hf_ckpt_39.pt",
"transformer.layers.30.attention.masked_bias": "hf_ckpt_40.pt",
"transformer.layers.30.attention.rotary_emb.inv_freq": "hf_ckpt_40.pt",
"transformer.layers.30.attention.query_key_value.weight": "hf_ckpt_40.pt",
"transformer.layers.30.attention.query_key_value.bias": "hf_ckpt_40.pt",
"transformer.layers.30.attention.dense.weight": "hf_ckpt_40.pt",
"transformer.layers.30.attention.dense.bias": "hf_ckpt_40.pt",
"transformer.layers.30.mlp.dense_h_to_4h.weight": "hf_ckpt_40.pt",
"transformer.layers.30.mlp.dense_h_to_4h.bias": "hf_ckpt_40.pt",
"transformer.layers.30.mlp.dense_4h_to_h.weight": "hf_ckpt_40.pt",
"transformer.layers.30.mlp.dense_4h_to_h.bias": "hf_ckpt_40.pt",
"transformer.layers.31.input_layernorm.weight": "hf_ckpt_40.pt",
"transformer.layers.31.input_layernorm.bias": "hf_ckpt_40.pt",
"transformer.layers.31.post_attention_layernorm.weight": "hf_ckpt_40.pt",
"transformer.layers.31.post_attention_layernorm.bias": "hf_ckpt_40.pt",
"transformer.layers.31.attention.bias": "hf_ckpt_40.pt",
"transformer.layers.31.attention.masked_bias": "hf_ckpt_40.pt",
"transformer.layers.31.attention.rotary_emb.inv_freq": "hf_ckpt_40.pt",
"transformer.layers.31.attention.query_key_value.weight": "hf_ckpt_40.pt",
"transformer.layers.31.attention.query_key_value.bias": "hf_ckpt_40.pt",
"transformer.layers.31.attention.dense.weight": "hf_ckpt_40.pt",
"transformer.layers.31.attention.dense.bias": "hf_ckpt_40.pt",
"transformer.layers.31.mlp.dense_h_to_4h.weight": "hf_ckpt_40.pt",
"transformer.layers.31.mlp.dense_h_to_4h.bias": "hf_ckpt_40.pt",
"transformer.layers.31.mlp.dense_4h_to_h.weight": "hf_ckpt_40.pt",
"transformer.layers.31.mlp.dense_4h_to_h.bias": "hf_ckpt_40.pt",
"transformer.layers.32.input_layernorm.weight": "hf_ckpt_40.pt",
"transformer.layers.32.input_layernorm.bias": "hf_ckpt_40.pt",
"transformer.layers.32.post_attention_layernorm.weight": "hf_ckpt_40.pt",
"transformer.layers.32.post_attention_layernorm.bias": "hf_ckpt_41.pt",
"transformer.layers.32.attention.bias": "hf_ckpt_41.pt",
"transformer.layers.32.attention.masked_bias": "hf_ckpt_41.pt",
"transformer.layers.32.attention.rotary_emb.inv_freq": "hf_ckpt_41.pt",
"transformer.layers.32.attention.query_key_value.weight": "hf_ckpt_41.pt",
"transformer.layers.32.attention.query_key_value.bias": "hf_ckpt_41.pt",
"transformer.layers.32.attention.dense.weight": "hf_ckpt_41.pt",
"transformer.layers.32.attention.dense.bias": "hf_ckpt_41.pt",
"transformer.layers.32.mlp.dense_h_to_4h.weight": "hf_ckpt_41.pt",
"transformer.layers.32.mlp.dense_h_to_4h.bias": "hf_ckpt_41.pt",
"transformer.layers.32.mlp.dense_4h_to_h.weight": "hf_ckpt_41.pt",
"transformer.layers.32.mlp.dense_4h_to_h.bias": "hf_ckpt_41.pt",
"transformer.layers.33.input_layernorm.weight": "hf_ckpt_41.pt",
"transformer.layers.33.input_layernorm.bias": "hf_ckpt_41.pt",
"transformer.layers.33.post_attention_layernorm.weight": "hf_ckpt_41.pt",
"transformer.layers.33.post_attention_layernorm.bias": "hf_ckpt_41.pt",
"transformer.layers.33.attention.bias": "hf_ckpt_41.pt",
"transformer.layers.33.attention.masked_bias": "hf_ckpt_41.pt",
"transformer.layers.33.attention.rotary_emb.inv_freq": "hf_ckpt_41.pt",
"transformer.layers.33.attention.query_key_value.weight": "hf_ckpt_41.pt",
"transformer.layers.33.attention.query_key_value.bias": "hf_ckpt_41.pt",
"transformer.layers.33.attention.dense.weight": "hf_ckpt_41.pt",
"transformer.layers.33.attention.dense.bias": "hf_ckpt_41.pt",
"transformer.layers.33.mlp.dense_h_to_4h.weight": "hf_ckpt_41.pt",
"transformer.layers.33.mlp.dense_h_to_4h.bias": "hf_ckpt_41.pt",
"transformer.layers.33.mlp.dense_4h_to_h.weight": "hf_ckpt_41.pt",
"transformer.layers.33.mlp.dense_4h_to_h.bias": "hf_ckpt_41.pt",
"transformer.layers.34.input_layernorm.weight": "hf_ckpt_41.pt",
"transformer.layers.34.input_layernorm.bias": "hf_ckpt_42.pt",
"transformer.layers.34.post_attention_layernorm.weight": "hf_ckpt_42.pt",
"transformer.layers.34.post_attention_layernorm.bias": "hf_ckpt_42.pt",
"transformer.layers.34.attention.bias": "hf_ckpt_42.pt",
"transformer.layers.34.attention.masked_bias": "hf_ckpt_42.pt",
"transformer.layers.34.attention.rotary_emb.inv_freq": "hf_ckpt_42.pt",
"transformer.layers.34.attention.query_key_value.weight": "hf_ckpt_42.pt",
"transformer.layers.34.attention.query_key_value.bias": "hf_ckpt_42.pt",
"transformer.layers.34.attention.dense.weight": "hf_ckpt_42.pt",
"transformer.layers.34.attention.dense.bias": "hf_ckpt_42.pt",
"transformer.layers.34.mlp.dense_h_to_4h.weight": "hf_ckpt_42.pt",
"transformer.layers.34.mlp.dense_h_to_4h.bias": "hf_ckpt_42.pt",
"transformer.layers.34.mlp.dense_4h_to_h.weight": "hf_ckpt_42.pt",
"transformer.layers.34.mlp.dense_4h_to_h.bias": "hf_ckpt_42.pt",
"transformer.layers.35.input_layernorm.weight": "hf_ckpt_42.pt",
"transformer.layers.35.input_layernorm.bias": "hf_ckpt_42.pt",
"transformer.layers.35.post_attention_layernorm.weight": "hf_ckpt_42.pt",
"transformer.layers.35.post_attention_layernorm.bias": "hf_ckpt_42.pt",
"transformer.layers.35.attention.bias": "hf_ckpt_42.pt",
"transformer.layers.35.attention.masked_bias": "hf_ckpt_42.pt",
"transformer.layers.35.attention.rotary_emb.inv_freq": "hf_ckpt_42.pt",
"transformer.layers.35.attention.query_key_value.weight": "hf_ckpt_42.pt",
"transformer.layers.35.attention.query_key_value.bias": "hf_ckpt_42.pt",
"transformer.layers.35.attention.dense.weight": "hf_ckpt_42.pt",
"transformer.layers.35.attention.dense.bias": "hf_ckpt_42.pt",
"transformer.layers.35.mlp.dense_h_to_4h.weight": "hf_ckpt_42.pt",
"transformer.layers.35.mlp.dense_h_to_4h.bias": "hf_ckpt_42.pt",
"transformer.layers.35.mlp.dense_4h_to_h.weight": "hf_ckpt_42.pt",
"transformer.layers.35.mlp.dense_4h_to_h.bias": "hf_ckpt_43.pt",
"transformer.layers.36.input_layernorm.weight": "hf_ckpt_43.pt",
"transformer.layers.36.input_layernorm.bias": "hf_ckpt_43.pt",
"transformer.layers.36.post_attention_layernorm.weight": "hf_ckpt_43.pt",
"transformer.layers.36.post_attention_layernorm.bias": "hf_ckpt_43.pt",
"transformer.layers.36.attention.bias": "hf_ckpt_43.pt",
"transformer.layers.36.attention.masked_bias": "hf_ckpt_43.pt",
"transformer.layers.36.attention.rotary_emb.inv_freq": "hf_ckpt_43.pt",
"transformer.layers.36.attention.query_key_value.weight": "hf_ckpt_43.pt",
"transformer.layers.36.attention.query_key_value.bias": "hf_ckpt_43.pt",
"transformer.layers.36.attention.dense.weight": "hf_ckpt_43.pt",
"transformer.layers.36.attention.dense.bias": "hf_ckpt_43.pt",
"transformer.layers.36.mlp.dense_h_to_4h.weight": "hf_ckpt_43.pt",
"transformer.layers.36.mlp.dense_h_to_4h.bias": "hf_ckpt_43.pt",
"transformer.layers.36.mlp.dense_4h_to_h.weight": "hf_ckpt_43.pt",
"transformer.layers.36.mlp.dense_4h_to_h.bias": "hf_ckpt_43.pt",
"transformer.layers.37.input_layernorm.weight": "hf_ckpt_43.pt",
"transformer.layers.37.input_layernorm.bias": "hf_ckpt_43.pt",
"transformer.layers.37.post_attention_layernorm.weight": "hf_ckpt_43.pt",
"transformer.layers.37.post_attention_layernorm.bias": "hf_ckpt_43.pt",
"transformer.layers.37.attention.bias": "hf_ckpt_43.pt",
"transformer.layers.37.attention.masked_bias": "hf_ckpt_43.pt",
"transformer.layers.37.attention.rotary_emb.inv_freq": "hf_ckpt_43.pt",
"transformer.layers.37.attention.query_key_value.weight": "hf_ckpt_43.pt",
"transformer.layers.37.attention.query_key_value.bias": "hf_ckpt_43.pt",
"transformer.layers.37.attention.dense.weight": "hf_ckpt_43.pt",
"transformer.layers.37.attention.dense.bias": "hf_ckpt_43.pt",
"transformer.layers.37.mlp.dense_h_to_4h.weight": "hf_ckpt_43.pt",
"transformer.layers.37.mlp.dense_h_to_4h.bias": "hf_ckpt_44.pt",
"transformer.layers.37.mlp.dense_4h_to_h.weight": "hf_ckpt_44.pt",
"transformer.layers.37.mlp.dense_4h_to_h.bias": "hf_ckpt_44.pt",
"transformer.layers.38.input_layernorm.weight": "hf_ckpt_44.pt",
"transformer.layers.38.input_layernorm.bias": "hf_ckpt_44.pt",
"transformer.layers.38.post_attention_layernorm.weight": "hf_ckpt_44.pt",
"transformer.layers.38.post_attention_layernorm.bias": "hf_ckpt_44.pt",
"transformer.layers.38.attention.bias": "hf_ckpt_44.pt",
"transformer.layers.38.attention.masked_bias": "hf_ckpt_44.pt",
"transformer.layers.38.attention.rotary_emb.inv_freq": "hf_ckpt_44.pt",
"transformer.layers.38.attention.query_key_value.weight": "hf_ckpt_44.pt",
"transformer.layers.38.attention.query_key_value.bias": "hf_ckpt_44.pt",
"transformer.layers.38.attention.dense.weight": "hf_ckpt_44.pt",
"transformer.layers.38.attention.dense.bias": "hf_ckpt_44.pt",
"transformer.layers.38.mlp.dense_h_to_4h.weight": "hf_ckpt_44.pt",
"transformer.layers.38.mlp.dense_h_to_4h.bias": "hf_ckpt_44.pt",
"transformer.layers.38.mlp.dense_4h_to_h.weight": "hf_ckpt_44.pt",
"transformer.layers.38.mlp.dense_4h_to_h.bias": "hf_ckpt_44.pt",
"transformer.layers.39.input_layernorm.weight": "hf_ckpt_44.pt",
"transformer.layers.39.input_layernorm.bias": "hf_ckpt_44.pt",
"transformer.layers.39.post_attention_layernorm.weight": "hf_ckpt_44.pt",
"transformer.layers.39.post_attention_layernorm.bias": "hf_ckpt_44.pt",
"transformer.layers.39.attention.bias": "hf_ckpt_44.pt",
"transformer.layers.39.attention.masked_bias": "hf_ckpt_44.pt",
"transformer.layers.39.attention.rotary_emb.inv_freq": "hf_ckpt_44.pt",
"transformer.layers.39.attention.query_key_value.weight": "hf_ckpt_44.pt",
"transformer.layers.39.attention.query_key_value.bias": "hf_ckpt_44.pt",
"transformer.layers.39.attention.dense.weight": "hf_ckpt_44.pt",
"transformer.layers.39.attention.dense.bias": "hf_ckpt_45.pt",
"transformer.layers.39.mlp.dense_h_to_4h.weight": "hf_ckpt_45.pt",
"transformer.layers.39.mlp.dense_h_to_4h.bias": "hf_ckpt_45.pt",
"transformer.layers.39.mlp.dense_4h_to_h.weight": "hf_ckpt_45.pt",
"transformer.layers.39.mlp.dense_4h_to_h.bias": "hf_ckpt_45.pt",
"transformer.layers.40.input_layernorm.weight": "hf_ckpt_45.pt",
"transformer.layers.40.input_layernorm.bias": "hf_ckpt_45.pt",
"transformer.layers.40.post_attention_layernorm.weight": "hf_ckpt_45.pt",
"transformer.layers.40.post_attention_layernorm.bias": "hf_ckpt_45.pt",
"transformer.layers.40.attention.bias": "hf_ckpt_45.pt",
"transformer.layers.40.attention.masked_bias": "hf_ckpt_45.pt",
"transformer.layers.40.attention.rotary_emb.inv_freq": "hf_ckpt_45.pt",
"transformer.layers.40.attention.query_key_value.weight": "hf_ckpt_45.pt",
"transformer.layers.40.attention.query_key_value.bias": "hf_ckpt_45.pt",
"transformer.layers.40.attention.dense.weight": "hf_ckpt_45.pt",
"transformer.layers.40.attention.dense.bias": "hf_ckpt_45.pt",
"transformer.layers.40.mlp.dense_h_to_4h.weight": "hf_ckpt_45.pt",
"transformer.layers.40.mlp.dense_h_to_4h.bias": "hf_ckpt_45.pt",
"transformer.layers.40.mlp.dense_4h_to_h.weight": "hf_ckpt_45.pt",
"transformer.layers.40.mlp.dense_4h_to_h.bias": "hf_ckpt_45.pt",
"transformer.layers.41.input_layernorm.weight": "hf_ckpt_45.pt",
"transformer.layers.41.input_layernorm.bias": "hf_ckpt_45.pt",
"transformer.layers.41.post_attention_layernorm.weight": "hf_ckpt_45.pt",
"transformer.layers.41.post_attention_layernorm.bias": "hf_ckpt_45.pt",
"transformer.layers.41.attention.bias": "hf_ckpt_45.pt",
"transformer.layers.41.attention.masked_bias": "hf_ckpt_45.pt",
"transformer.layers.41.attention.rotary_emb.inv_freq": "hf_ckpt_45.pt",
"transformer.layers.41.attention.query_key_value.weight": "hf_ckpt_45.pt",
"transformer.layers.41.attention.query_key_value.bias": "hf_ckpt_46.pt",
"transformer.layers.41.attention.dense.weight": "hf_ckpt_46.pt",
"transformer.layers.41.attention.dense.bias": "hf_ckpt_46.pt",
"transformer.layers.41.mlp.dense_h_to_4h.weight": "hf_ckpt_46.pt",
"transformer.layers.41.mlp.dense_h_to_4h.bias": "hf_ckpt_46.pt",
"transformer.layers.41.mlp.dense_4h_to_h.weight": "hf_ckpt_46.pt",
"transformer.layers.41.mlp.dense_4h_to_h.bias": "hf_ckpt_46.pt",
"transformer.layers.42.input_layernorm.weight": "hf_ckpt_46.pt",
"transformer.layers.42.input_layernorm.bias": "hf_ckpt_46.pt",
"transformer.layers.42.post_attention_layernorm.weight": "hf_ckpt_46.pt",
"transformer.layers.42.post_attention_layernorm.bias": "hf_ckpt_46.pt",
"transformer.layers.42.attention.bias": "hf_ckpt_46.pt",
"transformer.layers.42.attention.masked_bias": "hf_ckpt_46.pt",
"transformer.layers.42.attention.rotary_emb.inv_freq": "hf_ckpt_46.pt",
"transformer.layers.42.attention.query_key_value.weight": "hf_ckpt_46.pt",
"transformer.layers.42.attention.query_key_value.bias": "hf_ckpt_46.pt",
"transformer.layers.42.attention.dense.weight": "hf_ckpt_46.pt",
"transformer.layers.42.attention.dense.bias": "hf_ckpt_46.pt",
"transformer.layers.42.mlp.dense_h_to_4h.weight": "hf_ckpt_46.pt",
"transformer.layers.42.mlp.dense_h_to_4h.bias": "hf_ckpt_46.pt",
"transformer.layers.42.mlp.dense_4h_to_h.weight": "hf_ckpt_46.pt",
"transformer.layers.42.mlp.dense_4h_to_h.bias": "hf_ckpt_46.pt",
"transformer.layers.43.input_layernorm.weight": "hf_ckpt_46.pt",
"transformer.layers.43.input_layernorm.bias": "hf_ckpt_46.pt",
"transformer.layers.43.post_attention_layernorm.weight": "hf_ckpt_46.pt",
"transformer.layers.43.post_attention_layernorm.bias": "hf_ckpt_46.pt",
"transformer.layers.43.attention.bias": "hf_ckpt_46.pt",
"transformer.layers.43.attention.masked_bias": "hf_ckpt_46.pt",
"transformer.layers.43.attention.rotary_emb.inv_freq": "hf_ckpt_47.pt",
"transformer.layers.43.attention.query_key_value.weight": "hf_ckpt_47.pt",
"transformer.layers.43.attention.query_key_value.bias": "hf_ckpt_47.pt",
"transformer.layers.43.attention.dense.weight": "hf_ckpt_47.pt",
"transformer.layers.43.attention.dense.bias": "hf_ckpt_47.pt",
"transformer.layers.43.mlp.dense_h_to_4h.weight": "hf_ckpt_47.pt",
"transformer.layers.43.mlp.dense_h_to_4h.bias": "hf_ckpt_47.pt",
"transformer.layers.43.mlp.dense_4h_to_h.weight": "hf_ckpt_47.pt",
"transformer.layers.43.mlp.dense_4h_to_h.bias": "hf_ckpt_47.pt",
"transformer.final_layer_norm.weight": "hf_ckpt_47.pt",
"transformer.final_layer_norm.bias": "hf_ckpt_47.pt",
"v_head.weight": "hf_ckpt_47.pt"
}