- state.param_states.decoder.decoder_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_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_10.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_10.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_12.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_13.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_13.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_14.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_15.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_16.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_17.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_17.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_19.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_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_20.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.layers_22.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_22.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_3.pre_mlp_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_mlp_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.layers_8.pre_cross_attention_layer_norm.scale.v
- state.param_states.decoder.layers_8.pre_mlp_layer_norm.scale.v
- state.param_states.decoder.layers_8.pre_self_attention_layer_norm.scale.v
- state.param_states.decoder.relpos_bias.rel_embedding.v
- state.param_states.encoder.layers_1.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_10.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_13.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_14.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_14.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_15.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_16.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_18.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_20.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_20.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_23.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_4.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_6.pre_attention_layer_norm.scale.v
- state.param_states.encoder.layers_7.pre_mlp_layer_norm.scale.v
- state.param_states.encoder.layers_9.pre_attention_layer_norm.scale.v
- target.decoder.layers_0.encoder_decoder_attention.value.kernel
- target.decoder.layers_0.mlp.wi_0.kernel
- target.decoder.layers_0.mlp.wi_1.kernel
- target.decoder.layers_0.pre_cross_attention_layer_norm.scale