- state.param_states.decoder.decoder_norm.scale.v
- state.param_states.decoder.layers_0.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_0.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_0.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_1.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_1.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_1.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_2.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_2.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_2.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_3.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_3.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_3.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_4.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_4.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_4.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_5.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_5.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_5.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_6.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_6.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_6.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_7.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_7.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_7.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.relpos_bias.rel_embedding.v
- state.param_states.encoder.encoder_norm.scale.v
- state.param_states.encoder.layers_0.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_0.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_1.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_1.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_2.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_2.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_3.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_3.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_4.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_4.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_5.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_5.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_6.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_6.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_7.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_7.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.relpos_bias.rel_embedding.v
- target.decoder.decoder_norm.scale
- target.decoder.layers_0.encoder_decoder_attention.key.kernel
- target.decoder.layers_0.encoder_decoder_attention.out.kernel
- target.decoder.layers_0.encoder_decoder_attention.query.kernel
- target.decoder.layers_0.encoder_decoder_attention.value.kernel
- target.decoder.layers_0.mlp.wi_0.kernel