search

DigitalPhonetics / IMS-Toucan

Controllable and fast Text-to-Speech for over 7000 languages!
Apache License 2.0
1.4k stars 158 forks source link

Finetuning example: Error(s) in loading state_dict for ToucanTTS #139

Closed stlohrey closed 1 year ago

stlohrey commented 1 year ago

Hi, I am trying to follow the fine-tuning example, but just after the dataset preparation and caching i run into the following error:

RuntimeError: Error(s) in loading state_dict for ToucanTTS:
Unexpected key(s) in state_dict: "conv_postnet.postnet.0.0.weight", "conv_postnet.postnet.0.1.weight", "conv_postnet.postnet.0.1.bias", "conv_postnet.postnet.1.0.weight", "conv_postnet.postnet.1.1.weight", "conv_postnet.postnet.1.1.bias", "conv_postnet.postnet.2.0.weight", "conv_postnet.postnet.2.1.weight", "conv_postnet.postnet.2.1.bias", "conv_postnet.postnet.3.0.weight", "conv_postnet.postnet.3.1.weight", "conv_postnet.postnet.3.1.bias", "conv_postnet.postnet.4.0.weight", "conv_postnet.postnet.4.1.weight", "conv_postnet.postnet.4.1.bias", "pitch_predictor.conv.5.0.weight", "pitch_predictor.conv.5.0.bias", "pitch_predictor.conv.6.0.weight", "pitch_predictor.conv.6.0.bias", "pitch_predictor.norms.5.W_scale.0.weight", "pitch_predictor.norms.5.W_scale.0.bias", "pitch_predictor.norms.5.W_scale.2.weight", "pitch_predictor.norms.5.W_scale.2.bias", "pitch_predictor.norms.5.W_scale.4.weight", "pitch_predictor.norms.5.W_scale.4.bias", "pitch_predictor.norms.5.W_bias.0.weight", "pitch_predictor.norms.5.W_bias.0.bias", "pitch_predictor.norms.5.W_bias.2.weight", "pitch_predictor.norms.5.W_bias.2.bias", "pitch_predictor.norms.5.W_bias.4.weight", "pitch_predictor.norms.5.W_bias.4.bias", "pitch_predictor.norms.6.W_scale.0.weight", "pitch_predictor.norms.6.W_scale.0.bias", "pitch_predictor.norms.6.W_scale.2.weight", "pitch_predictor.norms.6.W_scale.2.bias", "pitch_predictor.norms.6.W_scale.4.weight", "pitch_predictor.norms.6.W_scale.4.bias", "pitch_predictor.norms.6.W_bias.0.weight", "pitch_predictor.norms.6.W_bias.0.bias", "pitch_predictor.norms.6.W_bias.2.weight", "pitch_predictor.norms.6.W_bias.2.bias", "pitch_predictor.norms.6.W_bias.4.weight", "pitch_predictor.norms.6.W_bias.4.bias", "post_flow.flows.36.logs", "post_flow.flows.36.bias", "post_flow.flows.37.l", "post_flow.flows.37.log_s", "post_flow.flows.37.u", "post_flow.flows.37.p", "post_flow.flows.37.sign_s", "post_flow.flows.37.l_mask", "post_flow.flows.37.eye", "post_flow.flows.38.start.bias", "post_flow.flows.38.start.weight_g", "post_flow.flows.38.start.weight_v", "post_flow.flows.38.end.weight", "post_flow.flows.38.end.bias", "post_flow.flows.38.wn.in_layers.0.bias", "post_flow.flows.38.wn.in_layers.0.weight_g", "post_flow.flows.38.wn.in_layers.0.weight_v", "post_flow.flows.38.wn.in_layers.1.bias", "post_flow.flows.38.wn.in_layers.1.weight_g", "post_flow.flows.38.wn.in_layers.1.weight_v", "post_flow.flows.38.wn.in_layers.2.bias", "post_flow.flows.38.wn.in_layers.2.weight_g", "post_flow.flows.38.wn.in_layers.2.weight_v", "post_flow.flows.38.wn.in_layers.3.bias", "post_flow.flows.38.wn.in_layers.3.weight_g", "post_flow.flows.38.wn.in_layers.3.weight_v", "post_flow.flows.38.wn.res_skip_layers.0.bias", "post_flow.flows.38.wn.res_skip_layers.0.weight_g", "post_flow.flows.38.wn.res_skip_layers.0.weight_v", "post_flow.flows.38.wn.res_skip_layers.1.bias", "post_flow.flows.38.wn.res_skip_layers.1.weight_g", "post_flow.flows.38.wn.res_skip_layers.1.weight_v", "post_flow.flows.38.wn.res_skip_layers.2.bias", "post_flow.flows.38.wn.res_skip_layers.2.weight_g", "post_flow.flows.38.wn.res_skip_layers.2.weight_v", "post_flow.flows.38.wn.res_skip_layers.3.bias", "post_flow.flows.38.wn.res_skip_layers.3.weight_g", "post_flow.flows.38.wn.res_skip_layers.3.weight_v", "post_flow.flows.38.wn.cond_layer.bias", "post_flow.flows.38.wn.cond_layer.weight_g", "post_flow.flows.38.wn.cond_layer.weight_v", "post_flow.flows.39.logs", "post_flow.flows.39.bias", "post_flow.flows.40.l", "post_flow.flows.40.log_s", "post_flow.flows.40.u", "post_flow.flows.40.p", "post_flow.flows.40.sign_s", "post_flow.flows.40.l_mask", "post_flow.flows.40.eye", "post_flow.flows.41.start.bias", "post_flow.flows.41.start.weight_g", "post_flow.flows.41.start.weight_v", "post_flow.flows.41.end.weight", "post_flow.flows.41.end.bias", "post_flow.flows.41.wn.in_layers.0.bias", "post_flow.flows.41.wn.in_layers.0.weight_g", "post_flow.flows.41.wn.in_layers.0.weight_v", "post_flow.flows.41.wn.in_layers.1.bias", "post_flow.flows.41.wn.in_layers.1.weight_g", "post_flow.flows.41.wn.in_layers.1.weight_v", "post_flow.flows.41.wn.in_layers.2.bias", "post_flow.flows.41.wn.in_layers.2.weight_g", "post_flow.flows.41.wn.in_layers.2.weight_v", "post_flow.flows.41.wn.in_layers.3.bias", "post_flow.flows.41.wn.in_layers.3.weight_g", "post_flow.flows.41.wn.in_layers.3.weight_v", "post_flow.flows.41.wn.res_skip_layers.0.bias", "post_flow.flows.41.wn.res_skip_layers.0.weight_g", "post_flow.flows.41.wn.res_skip_layers.0.weight_v", "post_flow.flows.41.wn.res_skip_layers.1.bias", "post_flow.flows.41.wn.res_skip_layers.1.weight_g", "post_flow.flows.41.wn.res_skip_layers.1.weight_v", "post_flow.flows.41.wn.res_skip_layers.2.bias", "post_flow.flows.41.wn.res_skip_layers.2.weight_g", "post_flow.flows.41.wn.res_skip_layers.2.weight_v", "post_flow.flows.41.wn.res_skip_layers.3.bias", "post_flow.flows.41.wn.res_skip_layers.3.weight_g", "post_flow.flows.41.wn.res_skip_layers.3.weight_v", "post_flow.flows.41.wn.cond_layer.bias", "post_flow.flows.41.wn.cond_layer.weight_g", "post_flow.flows.41.wn.cond_layer.weight_v", "post_flow.flows.42.logs", "post_flow.flows.42.bias", "post_flow.flows.43.l", "post_flow.flows.43.log_s", "post_flow.flows.43.u", "post_flow.flows.43.p", "post_flow.flows.43.sign_s", "post_flow.flows.43.l_mask", "post_flow.flows.43.eye", "post_flow.flows.44.start.bias", "post_flow.flows.44.start.weight_g", "post_flow.flows.44.start.weight_v", "post_flow.flows.44.end.weight", "post_flow.flows.44.end.bias", "post_flow.flows.44.wn.in_layers.0.bias", "post_flow.flows.44.wn.in_layers.0.weight_g", "post_flow.flows.44.wn.in_layers.0.weight_v", "post_flow.flows.44.wn.in_layers.1.bias", "post_flow.flows.44.wn.in_layers.1.weight_g", "post_flow.flows.44.wn.in_layers.1.weight_v", "post_flow.flows.44.wn.in_layers.2.bias", "post_flow.flows.44.wn.in_layers.2.weight_g", "post_flow.flows.44.wn.in_layers.2.weight_v", "post_flow.flows.44.wn.in_layers.3.bias", "post_flow.flows.44.wn.in_layers.3.weight_g", "post_flow.flows.44.wn.in_layers.3.weight_v", "post_flow.flows.44.wn.res_skip_layers.0.bias", "post_flow.flows.44.wn.res_skip_layers.0.weight_g", "post_flow.flows.44.wn.res_skip_layers.0.weight_v", "post_flow.flows.44.wn.res_skip_layers.1.bias", "post_flow.flows.44.wn.res_skip_layers.1.weight_g", "post_flow.flows.44.wn.res_skip_layers.1.weight_v", "post_flow.flows.44.wn.res_skip_layers.2.bias", "post_flow.flows.44.wn.res_skip_layers.2.weight_g", "post_flow.flows.44.wn.res_skip_layers.2.weight_v", "post_flow.flows.44.wn.res_skip_layers.3.bias", "post_flow.flows.44.wn.res_skip_layers.3.weight_g", "post_flow.flows.44.wn.res_skip_layers.3.weight_v", "post_flow.flows.44.wn.cond_layer.bias", "post_flow.flows.44.wn.cond_layer.weight_g", "post_flow.flows.44.wn.cond_layer.weight_v", "post_flow.flows.45.logs", "post_flow.flows.45.bias", "post_flow.flows.46.l", "post_flow.flows.46.log_s", "post_flow.flows.46.u", "post_flow.flows.46.p", "post_flow.flows.46.sign_s", "post_flow.flows.46.l_mask", "post_flow.flows.46.eye", "post_flow.flows.47.start.bias", "post_flow.flows.47.start.weight_g", "post_flow.flows.47.start.weight_v", "post_flow.flows.47.end.weight", "post_flow.flows.47.end.bias", "post_flow.flows.47.wn.in_layers.0.bias", "post_flow.flows.47.wn.in_layers.0.weight_g", "post_flow.flows.47.wn.in_layers.0.weight_v", "post_flow.flows.47.wn.in_layers.1.bias", "post_flow.flows.47.wn.in_layers.1.weight_g", "post_flow.flows.47.wn.in_layers.1.weight_v", "post_flow.flows.47.wn.in_layers.2.bias", "post_flow.flows.47.wn.in_layers.2.weight_g", "post_flow.flows.47.wn.in_layers.2.weight_v", "post_flow.flows.47.wn.in_layers.3.bias", "post_flow.flows.47.wn.in_layers.3.weight_g", "post_flow.flows.47.wn.in_layers.3.weight_v", "post_flow.flows.47.wn.res_skip_layers.0.bias", "post_flow.flows.47.wn.res_skip_layers.0.weight_g", "post_flow.flows.47.wn.res_skip_layers.0.weight_v", "post_flow.flows.47.wn.res_skip_layers.1.bias", "post_flow.flows.47.wn.res_skip_layers.1.weight_g", "post_flow.flows.47.wn.res_skip_layers.1.weight_v", "post_flow.flows.47.wn.res_skip_layers.2.bias", "post_flow.flows.47.wn.res_skip_layers.2.weight_g", "post_flow.flows.47.wn.res_skip_layers.2.weight_v", "post_flow.flows.47.wn.res_skip_layers.3.bias", "post_flow.flows.47.wn.res_skip_layers.3.weight_g", "post_flow.flows.47.wn.res_skip_layers.3.weight_v", "post_flow.flows.47.wn.cond_layer.bias", "post_flow.flows.47.wn.cond_layer.weight_g", "post_flow.flows.47.wn.cond_layer.weight_v", "post_flow.flows.48.logs", "post_flow.flows.48.bias", "post_flow.flows.49.l", "post_flow.flows.49.log_s", "post_flow.flows.49.u", "post_flow.flows.49.p", "post_flow.flows.49.sign_s", "post_flow.flows.49.l_mask", "post_flow.flows.49.eye", "post_flow.flows.50.start.bias", "post_flow.flows.50.start.weight_g", "post_flow.flows.50.start.weight_v", "post_flow.flows.50.end.weight", "post_flow.flows.50.end.bias", "post_flow.flows.50.wn.in_layers.0.bias", "post_flow.flows.50.wn.in_layers.0.weight_g", "post_flow.flows.50.wn.in_layers.0.weight_v", "post_flow.flows.50.wn.in_layers.1.bias", "post_flow.flows.50.wn.in_layers.1.weight_g", "post_flow.flows.50.wn.in_layers.1.weight_v", "post_flow.flows.50.wn.in_layers.2.bias", "post_flow.flows.50.wn.in_layers.2.weight_g", "post_flow.flows.50.wn.in_layers.2.weight_v", "post_flow.flows.50.wn.in_layers.3.bias", "post_flow.flows.50.wn.in_layers.3.weight_g", "post_flow.flows.50.wn.in_layers.3.weight_v", "post_flow.flows.50.wn.res_skip_layers.0.bias", "post_flow.flows.50.wn.res_skip_layers.0.weight_g", "post_flow.flows.50.wn.res_skip_layers.0.weight_v", "post_flow.flows.50.wn.res_skip_layers.1.bias", "post_flow.flows.50.wn.res_skip_layers.1.weight_g", "post_flow.flows.50.wn.res_skip_layers.1.weight_v", "post_flow.flows.50.wn.res_skip_layers.2.bias", "post_flow.flows.50.wn.res_skip_layers.2.weight_g", "post_flow.flows.50.wn.res_skip_layers.2.weight_v", "post_flow.flows.50.wn.res_skip_layers.3.bias", "post_flow.flows.50.wn.res_skip_layers.3.weight_g", "post_flow.flows.50.wn.res_skip_layers.3.weight_v", "post_flow.flows.50.wn.cond_layer.bias", "post_flow.flows.50.wn.cond_layer.weight_g", "post_flow.flows.50.wn.cond_layer.weight_v", "post_flow.flows.51.logs", "post_flow.flows.51.bias", "post_flow.flows.52.l", "post_flow.flows.52.log_s", "post_flow.flows.52.u", "post_flow.flows.52.p", "post_flow.flows.52.sign_s", "post_flow.flows.52.l_mask", "post_flow.flows.52.eye", "post_flow.flows.53.start.bias", "post_flow.flows.53.start.weight_g", "post_flow.flows.53.start.weight_v", "post_flow.flows.53.end.weight", "post_flow.flows.53.end.bias", "post_flow.flows.53.wn.in_layers.0.bias", "post_flow.flows.53.wn.in_layers.0.weight_g", "post_flow.flows.53.wn.in_layers.0.weight_v", "post_flow.flows.53.wn.in_layers.1.bias", "post_flow.flows.53.wn.in_layers.1.weight_g", "post_flow.flows.53.wn.in_layers.1.weight_v", "post_flow.flows.53.wn.in_layers.2.bias", "post_flow.flows.53.wn.in_layers.2.weight_g", "post_flow.flows.53.wn.in_layers.2.weight_v", "post_flow.flows.53.wn.in_layers.3.bias", "post_flow.flows.53.wn.in_layers.3.weight_g", "post_flow.flows.53.wn.in_layers.3.weight_v", "post_flow.flows.53.wn.res_skip_layers.0.bias", "post_flow.flows.53.wn.res_skip_layers.0.weight_g", "post_flow.flows.53.wn.res_skip_layers.0.weight_v", "post_flow.flows.53.wn.res_skip_layers.1.bias", "post_flow.flows.53.wn.res_skip_layers.1.weight_g", "post_flow.flows.53.wn.res_skip_layers.1.weight_v", "post_flow.flows.53.wn.res_skip_layers.2.bias", "post_flow.flows.53.wn.res_skip_layers.2.weight_g", "post_flow.flows.53.wn.res_skip_layers.2.weight_v", "post_flow.flows.53.wn.res_skip_layers.3.bias", "post_flow.flows.53.wn.res_skip_layers.3.weight_g", "post_flow.flows.53.wn.res_skip_layers.3.weight_v", "post_flow.flows.53.wn.cond_layer.bias", "post_flow.flows.53.wn.cond_layer.weight_g", "post_flow.flows.53.wn.cond_layer.weight_v", "post_flow.flows.2.wn.in_layers.3.bias", "post_flow.flows.2.wn.in_layers.3.weight_g", "post_flow.flows.2.wn.in_layers.3.weight_v", "post_flow.flows.2.wn.res_skip_layers.3.bias", "post_flow.flows.2.wn.res_skip_layers.3.weight_g", "post_flow.flows.2.wn.res_skip_layers.3.weight_v", "post_flow.flows.5.wn.in_layers.3.bias", "post_flow.flows.5.wn.in_layers.3.weight_g", "post_flow.flows.5.wn.in_layers.3.weight_v", "post_flow.flows.5.wn.res_skip_layers.3.bias", "post_flow.flows.5.wn.res_skip_layers.3.weight_g", "post_flow.flows.5.wn.res_skip_layers.3.weight_v", "post_flow.flows.8.wn.in_layers.3.bias", "post_flow.flows.8.wn.in_layers.3.weight_g", "post_flow.flows.8.wn.in_layers.3.weight_v", "post_flow.flows.8.wn.res_skip_layers.3.bias", "post_flow.flows.8.wn.res_skip_layers.3.weight_g", "post_flow.flows.8.wn.res_skip_layers.3.weight_v", "post_flow.flows.11.wn.in_layers.3.bias", "post_flow.flows.11.wn.in_layers.3.weight_g", "post_flow.flows.11.wn.in_layers.3.weight_v", "post_flow.flows.11.wn.res_skip_layers.3.bias", "post_flow.flows.11.wn.res_skip_layers.3.weight_g", "post_flow.flows.11.wn.res_skip_layers.3.weight_v", "post_flow.flows.14.wn.in_layers.3.bias", "post_flow.flows.14.wn.in_layers.3.weight_g", "post_flow.flows.14.wn.in_layers.3.weight_v", "post_flow.flows.14.wn.res_skip_layers.3.bias", "post_flow.flows.14.wn.res_skip_layers.3.weight_g", "post_flow.flows.14.wn.res_skip_layers.3.weight_v", "post_flow.flows.17.wn.in_layers.3.bias", "post_flow.flows.17.wn.in_layers.3.weight_g", "post_flow.flows.17.wn.in_layers.3.weight_v", "post_flow.flows.17.wn.res_skip_layers.3.bias", "post_flow.flows.17.wn.res_skip_layers.3.weight_g", "post_flow.flows.17.wn.res_skip_layers.3.weight_v", "post_flow.flows.20.wn.in_layers.3.bias", "post_flow.flows.20.wn.in_layers.3.weight_g", "post_flow.flows.20.wn.in_layers.3.weight_v", "post_flow.flows.20.wn.res_skip_layers.3.bias", "post_flow.flows.20.wn.res_skip_layers.3.weight_g", "post_flow.flows.20.wn.res_skip_layers.3.weight_v", "post_flow.flows.23.wn.in_layers.3.bias", "post_flow.flows.23.wn.in_layers.3.weight_g", "post_flow.flows.23.wn.in_layers.3.weight_v", "post_flow.flows.23.wn.res_skip_layers.3.bias", "post_flow.flows.23.wn.res_skip_layers.3.weight_g", "post_flow.flows.23.wn.res_skip_layers.3.weight_v", "post_flow.flows.26.wn.in_layers.3.bias", "post_flow.flows.26.wn.in_layers.3.weight_g", "post_flow.flows.26.wn.in_layers.3.weight_v", "post_flow.flows.26.wn.res_skip_layers.3.bias", "post_flow.flows.26.wn.res_skip_layers.3.weight_g", "post_flow.flows.26.wn.res_skip_layers.3.weight_v", "post_flow.flows.29.wn.in_layers.3.bias", "post_flow.flows.29.wn.in_layers.3.weight_g", "post_flow.flows.29.wn.in_layers.3.weight_v", "post_flow.flows.29.wn.res_skip_layers.3.bias", "post_flow.flows.29.wn.res_skip_layers.3.weight_g", "post_flow.flows.29.wn.res_skip_layers.3.weight_v", "post_flow.flows.32.wn.in_layers.3.bias", "post_flow.flows.32.wn.in_layers.3.weight_g", "post_flow.flows.32.wn.in_layers.3.weight_v", "post_flow.flows.32.wn.res_skip_layers.3.bias", "post_flow.flows.32.wn.res_skip_layers.3.weight_g", "post_flow.flows.32.wn.res_skip_layers.3.weight_v", "post_flow.flows.35.wn.in_layers.3.bias", "post_flow.flows.35.wn.in_layers.3.weight_g", "post_flow.flows.35.wn.in_layers.3.weight_v", "post_flow.flows.35.wn.res_skip_layers.3.bias", "post_flow.flows.35.wn.res_skip_layers.3.weight_g", "post_flow.flows.35.wn.res_skip_layers.3.weight_v". 
    size mismatch for duration_predictor.conv.0.0.weight: copying a param with shape torch.Size([256, 192, 3]) from checkpoint, the shape in current model is torch.Size([192, 192, 3]).
    size mismatch for duration_predictor.conv.0.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.conv.1.0.weight: copying a param with shape torch.Size([256, 256, 3]) from checkpoint, the shape in current model is torch.Size([192, 192, 3]).
    size mismatch for duration_predictor.conv.1.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.conv.2.0.weight: copying a param with shape torch.Size([256, 256, 3]) from checkpoint, the shape in current model is torch.Size([192, 192, 3]).
    size mismatch for duration_predictor.conv.2.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.0.W_scale.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for duration_predictor.norms.0.W_scale.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.0.W_scale.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for duration_predictor.norms.0.W_scale.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.0.W_bias.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for duration_predictor.norms.0.W_bias.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.0.W_bias.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for duration_predictor.norms.0.W_bias.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.1.W_scale.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for duration_predictor.norms.1.W_scale.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.1.W_scale.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for duration_predictor.norms.1.W_scale.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.1.W_bias.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for duration_predictor.norms.1.W_bias.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.1.W_bias.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for duration_predictor.norms.1.W_bias.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.2.W_scale.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for duration_predictor.norms.2.W_scale.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.2.W_scale.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for duration_predictor.norms.2.W_scale.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.2.W_bias.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for duration_predictor.norms.2.W_bias.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.norms.2.W_bias.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for duration_predictor.norms.2.W_bias.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for duration_predictor.linear.weight: copying a param with shape torch.Size([1, 256]) from checkpoint, the shape in current model is torch.Size([1, 192]).
    size mismatch for pitch_predictor.conv.0.0.weight: copying a param with shape torch.Size([256, 192, 5]) from checkpoint, the shape in current model is torch.Size([192, 192, 3]).
    size mismatch for pitch_predictor.conv.0.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.conv.1.0.weight: copying a param with shape torch.Size([256, 256, 5]) from checkpoint, the shape in current model is torch.Size([192, 192, 3]).
    size mismatch for pitch_predictor.conv.1.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.conv.2.0.weight: copying a param with shape torch.Size([256, 256, 5]) from checkpoint, the shape in current model is torch.Size([192, 192, 3]).
    size mismatch for pitch_predictor.conv.2.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.conv.3.0.weight: copying a param with shape torch.Size([256, 256, 5]) from checkpoint, the shape in current model is torch.Size([192, 192, 3]).
    size mismatch for pitch_predictor.conv.3.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.conv.4.0.weight: copying a param with shape torch.Size([256, 256, 5]) from checkpoint, the shape in current model is torch.Size([192, 192, 3]).
    size mismatch for pitch_predictor.conv.4.0.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.0.W_scale.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for pitch_predictor.norms.0.W_scale.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.0.W_scale.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for pitch_predictor.norms.0.W_scale.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.0.W_bias.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for pitch_predictor.norms.0.W_bias.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.0.W_bias.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for pitch_predictor.norms.0.W_bias.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.1.W_scale.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for pitch_predictor.norms.1.W_scale.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.1.W_scale.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for pitch_predictor.norms.1.W_scale.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.1.W_bias.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for pitch_predictor.norms.1.W_bias.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.1.W_bias.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for pitch_predictor.norms.1.W_bias.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.2.W_scale.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for pitch_predictor.norms.2.W_scale.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.2.W_scale.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for pitch_predictor.norms.2.W_scale.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.2.W_bias.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for pitch_predictor.norms.2.W_bias.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.2.W_bias.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for pitch_predictor.norms.2.W_bias.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.3.W_scale.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for pitch_predictor.norms.3.W_scale.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.3.W_scale.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for pitch_predictor.norms.3.W_scale.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.3.W_bias.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for pitch_predictor.norms.3.W_bias.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.3.W_bias.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for pitch_predictor.norms.3.W_bias.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.4.W_scale.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for pitch_predictor.norms.4.W_scale.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.4.W_scale.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for pitch_predictor.norms.4.W_scale.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.4.W_bias.2.weight: copying a param with shape torch.Size([256, 64]) from checkpoint, the shape in current model is torch.Size([192, 64]).
    size mismatch for pitch_predictor.norms.4.W_bias.2.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.norms.4.W_bias.4.weight: copying a param with shape torch.Size([256, 256]) from checkpoint, the shape in current model is torch.Size([192, 192]).
    size mismatch for pitch_predictor.norms.4.W_bias.4.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for pitch_predictor.linear.weight: copying a param with shape torch.Size([1, 256]) from checkpoint, the shape in current model is torch.Size([1, 192]).
    size mismatch for post_flow.flows.2.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.2.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.2.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.2.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.2.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.2.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.2.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.2.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.2.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.5.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.5.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.5.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.5.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.5.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.5.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.5.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.5.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.5.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.8.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.8.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.8.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.8.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.8.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.8.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.8.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.8.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.8.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.11.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.11.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.11.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.11.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.11.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.11.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.11.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.11.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.11.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.14.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.14.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.14.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.14.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.14.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.14.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.14.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.14.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.14.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.17.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.17.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.17.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.17.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.17.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.17.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.17.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.17.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.17.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.20.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.20.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.20.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.20.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.20.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.20.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.20.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.20.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.20.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.23.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.23.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.23.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.23.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.23.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.23.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.23.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.23.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.23.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.26.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.26.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.26.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.26.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.26.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.26.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.26.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.26.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.26.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.29.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.29.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.29.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.29.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.29.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.29.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.29.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.29.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.29.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.32.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.32.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.32.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.32.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.32.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.32.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.32.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.32.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.32.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).
    size mismatch for post_flow.flows.35.wn.in_layers.0.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.35.wn.in_layers.1.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.35.wn.in_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 5]) from checkpoint, the shape in current model is torch.Size([384, 192, 3]).
    size mismatch for post_flow.flows.35.wn.res_skip_layers.2.bias: copying a param with shape torch.Size([384]) from checkpoint, the shape in current model is torch.Size([192]).
    size mismatch for post_flow.flows.35.wn.res_skip_layers.2.weight_g: copying a param with shape torch.Size([384, 1, 1]) from checkpoint, the shape in current model is torch.Size([192, 1, 1]).
    size mismatch for post_flow.flows.35.wn.res_skip_layers.2.weight_v: copying a param with shape torch.Size([384, 192, 1]) from checkpoint, the shape in current model is torch.Size([192, 192, 1]).
    size mismatch for post_flow.flows.35.wn.cond_layer.bias: copying a param with shape torch.Size([1536]) from checkpoint, the shape in current model is torch.Size([1152]).
    size mismatch for post_flow.flows.35.wn.cond_layer.weight_g: copying a param with shape torch.Size([1536, 1, 1]) from checkpoint, the shape in current model is torch.Size([1152, 1, 1]).
    size mismatch for post_flow.flows.35.wn.cond_layer.weight_v: copying a param with shape torch.Size([1536, 384, 1]) from checkpoint, the shape in current model is torch.Size([1152, 384, 1]).

It seems to work just fine if I do not resume from the Meta checkpoint. Is the latest available Meta checkpoint not working with this example or am I doing something wrong here?

Thanks!

Flux9665 commented 1 year ago

sorry, that was my mistake, I made an experimental change on the main branch instead of an experimental one. It should work again now.

I'm trying a smaller model configuration that is hopefully easier to finetune, but there is no matching pretrained model yet.

stlohrey commented 1 year ago

allright, i will try again, thank you!