Add MLX conversion with inference fixes

README.md

@@ -1,34 +1,38 @@
 ---
-license: apache-2.0
+tags:
+- mlx
+- text-to-speech
+- kitten-tts
 ---
-# Kitten TTS Mini 0.8 (MLX, 8-bit selective)
 
-This is a selectively quantized 8-bit MLX version of the KittenTTS Mini 0.8 model.
+# mlx-community/kitten-tts-mini-0.8-8bit
 
-## Source
-
-- Original model: `KittenML/kitten-tts-mini-0.8`
-- License: Apache-2.0
+This is the **INT8 (MLX 8-bit)** MLX conversion of `KittenML/kitten-tts-mini-0.8`.
 
 ## Usage
 
-````python
-from mlx_audio.tts.utils import load_model
-import soundfile as sf
+```bash
+pip install -U mlx-audio
+```
+
+```bash
+python -m mlx_audio.tts.generate --model mlx-community/kitten-tts-mini-0.8-8bit --text "This is a local MLX test voice." --voice "expr-voice-5-m"
+```
+
+## Inference Notes
 
-model = load_model("mlx-community/kitten-tts-mini-0.8-8bit")
+The MLX implementation includes small end-of-utterance smoothing to prevent abrupt cutoffs.
+You can override it with `fade_out_ms=0` and `tail_silence_ms=0` in `Model.generate()`.
 
-for out in model.generate("This model runs with MLX.", voice="Jasper"):
-    sf.write("out.wav", out.audio, model.sample_rate)
-````
+## Conversion Notes / Fixes
 
-Notes:
-- `phonemizer` (espeak backend) is required for `generate()`.
-- Available voices: Bella, Jasper, Luna, Bruno, Rosie, Hugo, Kiki, Leo.
+- AdaIN `fc.weight` orientation was corrected (ONNX stores as `(in, out)` even when square).
+- AdaIN Snake `alpha` parameters are loaded and used for generator resblocks.
+- ConvTranspose output padding matches the original (right-side pad for `output_padding=1`).
+- Phase slice is passed through `sin` before ISTFT, matching the ONNX graph.
+- ISTFT uses normalized windowing without phase unwrap (to match original behavior).
+- Tail trim + dynamic fade-out + tail silence are applied at inference time to avoid a trailing spurt.
 
-## Quantization
+## Original Model
 
-- Group size: 64
-- Bits: 8
-- Mode: affine
-- Decoder/ISTFT generator kept in fp32 to avoid audio artifacts.
+Refer to the original model card for details: https://huggingface.co/KittenML/kitten-tts-mini-0.8

config.json

@@ -1,4 +1,155 @@
 {
+    "activation_quant_modules": [
+        "bert.encoder.albert_layer_groups.0.albert_layers.0.attention.dense",
+        "bert.encoder.albert_layer_groups.0.albert_layers.0.attention.key",
+        "bert.encoder.albert_layer_groups.0.albert_layers.0.attention.query",
+        "bert.encoder.albert_layer_groups.0.albert_layers.0.attention.value",
+        "bert.encoder.albert_layer_groups.0.albert_layers.0.ffn",
+        "bert.encoder.albert_layer_groups.0.albert_layers.0.ffn_output",
+        "bert.encoder.embedding_hidden_mapping_in",
+        "bert_encoder",
+        "decoder.asr_res.0",
+        "decoder.decode.0.conv1",
+        "decoder.decode.0.conv1x1",
+        "decoder.decode.0.conv2",
+        "decoder.decode.0.norm1.fc",
+        "decoder.decode.0.norm2.fc",
+        "decoder.decode.1.conv1",
+        "decoder.decode.1.conv1x1",
+        "decoder.decode.1.conv2",
+        "decoder.decode.1.norm1.fc",
+        "decoder.decode.1.norm2.fc",
+        "decoder.decode.2.conv1",
+        "decoder.decode.2.conv1x1",
+        "decoder.decode.2.conv2",
+        "decoder.decode.2.norm1.fc",
+        "decoder.decode.2.norm2.fc",
+        "decoder.decode.3.conv1",
+        "decoder.decode.3.conv1x1",
+        "decoder.decode.3.conv2",
+        "decoder.decode.3.norm1.fc",
+        "decoder.decode.3.norm2.fc",
+        "decoder.encode.conv1",
+        "decoder.encode.conv1x1",
+        "decoder.encode.conv2",
+        "decoder.encode.norm1.fc",
+        "decoder.encode.norm2.fc",
+        "decoder.generator.conv_post",
+        "decoder.generator.noise_convs.0",
+        "decoder.generator.noise_convs.1",
+        "decoder.generator.noise_res.0.adain1.0.fc",
+        "decoder.generator.noise_res.0.adain1.1.fc",
+        "decoder.generator.noise_res.0.adain1.2.fc",
+        "decoder.generator.noise_res.0.adain2.0.fc",
+        "decoder.generator.noise_res.0.adain2.1.fc",
+        "decoder.generator.noise_res.0.adain2.2.fc",
+        "decoder.generator.noise_res.0.convs1.0",
+        "decoder.generator.noise_res.0.convs1.1",
+        "decoder.generator.noise_res.0.convs1.2",
+        "decoder.generator.noise_res.0.convs2.0",
+        "decoder.generator.noise_res.0.convs2.1",
+        "decoder.generator.noise_res.0.convs2.2",
+        "decoder.generator.noise_res.1.adain1.0.fc",
+        "decoder.generator.noise_res.1.adain1.1.fc",
+        "decoder.generator.noise_res.1.adain1.2.fc",
+        "decoder.generator.noise_res.1.adain2.0.fc",
+        "decoder.generator.noise_res.1.adain2.1.fc",
+        "decoder.generator.noise_res.1.adain2.2.fc",
+        "decoder.generator.noise_res.1.convs1.0",
+        "decoder.generator.noise_res.1.convs1.1",
+        "decoder.generator.noise_res.1.convs1.2",
+        "decoder.generator.noise_res.1.convs2.0",
+        "decoder.generator.noise_res.1.convs2.1",
+        "decoder.generator.noise_res.1.convs2.2",
+        "decoder.generator.resblocks.0.adain1.0.fc",
+        "decoder.generator.resblocks.0.adain1.1.fc",
+        "decoder.generator.resblocks.0.adain1.2.fc",
+        "decoder.generator.resblocks.0.adain2.0.fc",
+        "decoder.generator.resblocks.0.adain2.1.fc",
+        "decoder.generator.resblocks.0.adain2.2.fc",
+        "decoder.generator.resblocks.0.convs1.0",
+        "decoder.generator.resblocks.0.convs1.1",
+        "decoder.generator.resblocks.0.convs1.2",
+        "decoder.generator.resblocks.0.convs2.0",
+        "decoder.generator.resblocks.0.convs2.1",
+        "decoder.generator.resblocks.0.convs2.2",
+        "decoder.generator.resblocks.1.adain1.0.fc",
+        "decoder.generator.resblocks.1.adain1.1.fc",
+        "decoder.generator.resblocks.1.adain1.2.fc",
+        "decoder.generator.resblocks.1.adain2.0.fc",
+        "decoder.generator.resblocks.1.adain2.1.fc",
+        "decoder.generator.resblocks.1.adain2.2.fc",
+        "decoder.generator.resblocks.1.convs1.0",
+        "decoder.generator.resblocks.1.convs1.1",
+        "decoder.generator.resblocks.1.convs1.2",
+        "decoder.generator.resblocks.1.convs2.0",
+        "decoder.generator.resblocks.1.convs2.1",
+        "decoder.generator.resblocks.1.convs2.2",
+        "decoder.generator.resblocks.2.adain1.0.fc",
+        "decoder.generator.resblocks.2.adain1.1.fc",
+        "decoder.generator.resblocks.2.adain1.2.fc",
+        "decoder.generator.resblocks.2.adain2.0.fc",
+        "decoder.generator.resblocks.2.adain2.1.fc",
+        "decoder.generator.resblocks.2.adain2.2.fc",
+        "decoder.generator.resblocks.2.convs1.0",
+        "decoder.generator.resblocks.2.convs1.1",
+        "decoder.generator.resblocks.2.convs1.2",
+        "decoder.generator.resblocks.2.convs2.0",
+        "decoder.generator.resblocks.2.convs2.1",
+        "decoder.generator.resblocks.2.convs2.2",
+        "decoder.generator.resblocks.3.adain1.0.fc",
+        "decoder.generator.resblocks.3.adain1.1.fc",
+        "decoder.generator.resblocks.3.adain1.2.fc",
+        "decoder.generator.resblocks.3.adain2.0.fc",
+        "decoder.generator.resblocks.3.adain2.1.fc",
+        "decoder.generator.resblocks.3.adain2.2.fc",
+        "decoder.generator.resblocks.3.convs1.0",
+        "decoder.generator.resblocks.3.convs1.1",
+        "decoder.generator.resblocks.3.convs1.2",
+        "decoder.generator.resblocks.3.convs2.0",
+        "decoder.generator.resblocks.3.convs2.1",
+        "decoder.generator.resblocks.3.convs2.2",
+        "predictor.F0.0.conv1",
+        "predictor.F0.0.conv2",
+        "predictor.F0.0.norm1.fc",
+        "predictor.F0.0.norm2.fc",
+        "predictor.F0.1.conv1",
+        "predictor.F0.1.conv1x1",
+        "predictor.F0.1.conv2",
+        "predictor.F0.1.norm1.fc",
+        "predictor.F0.1.norm2.fc",
+        "predictor.F0.2.conv1",
+        "predictor.F0.2.conv2",
+        "predictor.F0.2.norm1.fc",
+        "predictor.F0.2.norm2.fc",
+        "predictor.F0_proj",
+        "predictor.N.0.conv1",
+        "predictor.N.0.conv2",
+        "predictor.N.0.norm1.fc",
+        "predictor.N.0.norm2.fc",
+        "predictor.N.1.conv1",
+        "predictor.N.1.conv1x1",
+        "predictor.N.1.conv2",
+        "predictor.N.1.norm1.fc",
+        "predictor.N.1.norm2.fc",
+        "predictor.N.2.conv1",
+        "predictor.N.2.conv2",
+        "predictor.N.2.norm1.fc",
+        "predictor.N.2.norm2.fc",
+        "predictor.N_proj",
+        "predictor.lstm",
+        "predictor.shared",
+        "predictor.text_encoder.lstms.0",
+        "predictor.text_encoder.lstms.1.fc",
+        "predictor.text_encoder.lstms.2",
+        "predictor.text_encoder.lstms.3.fc",
+        "predictor.text_encoder.lstms.4",
+        "predictor.text_encoder.lstms.5.fc",
+        "text_encoder.cnn.0.0",
+        "text_encoder.cnn.1.0",
+        "text_encoder.cnn.2.0",
+        "text_encoder.lstm"
+    ],
     "asr_res_dim": 64,
     "decoder_out_dim": 512,
     "hidden_dim": 512,

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0fa1fa125e44a01ef2d87658305f648edfcd58b12173f7915e9452c5e28e823
-size 265112122
+oid sha256:e04ae3840b450fedf3743bc24c81818395aa074f010cf2726b5c0e4949d8ab8b
+size 253056500

model.safetensors.index.json

@@ -1,6 +1,6 @@
 {
     "metadata": {
-        "total_size": 265053108,
+        "total_size": 252987124,
         "total_parameters": 73782354
     },
     "weight_map": {
@@ -73,8 +73,12 @@
         "decoder.decode.0.conv2.weight_g": "model.safetensors",
         "decoder.decode.0.conv2.weight_v": "model.safetensors",
         "decoder.decode.0.norm1.fc.bias": "model.safetensors",
+        "decoder.decode.0.norm1.fc.biases": "model.safetensors",
+        "decoder.decode.0.norm1.fc.scales": "model.safetensors",
         "decoder.decode.0.norm1.fc.weight": "model.safetensors",
         "decoder.decode.0.norm2.fc.bias": "model.safetensors",
+        "decoder.decode.0.norm2.fc.biases": "model.safetensors",
+        "decoder.decode.0.norm2.fc.scales": "model.safetensors",
         "decoder.decode.0.norm2.fc.weight": "model.safetensors",
         "decoder.decode.1.conv1.bias": "model.safetensors",
         "decoder.decode.1.conv1.weight_g": "model.safetensors",
@@ -85,8 +89,12 @@
         "decoder.decode.1.conv2.weight_g": "model.safetensors",
         "decoder.decode.1.conv2.weight_v": "model.safetensors",
         "decoder.decode.1.norm1.fc.bias": "model.safetensors",
+        "decoder.decode.1.norm1.fc.biases": "model.safetensors",
+        "decoder.decode.1.norm1.fc.scales": "model.safetensors",
         "decoder.decode.1.norm1.fc.weight": "model.safetensors",
         "decoder.decode.1.norm2.fc.bias": "model.safetensors",
+        "decoder.decode.1.norm2.fc.biases": "model.safetensors",
+        "decoder.decode.1.norm2.fc.scales": "model.safetensors",
         "decoder.decode.1.norm2.fc.weight": "model.safetensors",
         "decoder.decode.2.conv1.bias": "model.safetensors",
         "decoder.decode.2.conv1.weight_g": "model.safetensors",
@@ -97,8 +105,12 @@
         "decoder.decode.2.conv2.weight_g": "model.safetensors",
         "decoder.decode.2.conv2.weight_v": "model.safetensors",
         "decoder.decode.2.norm1.fc.bias": "model.safetensors",
+        "decoder.decode.2.norm1.fc.biases": "model.safetensors",
+        "decoder.decode.2.norm1.fc.scales": "model.safetensors",
         "decoder.decode.2.norm1.fc.weight": "model.safetensors",
         "decoder.decode.2.norm2.fc.bias": "model.safetensors",
+        "decoder.decode.2.norm2.fc.biases": "model.safetensors",
+        "decoder.decode.2.norm2.fc.scales": "model.safetensors",
         "decoder.decode.2.norm2.fc.weight": "model.safetensors",
         "decoder.decode.3.conv1.bias": "model.safetensors",
         "decoder.decode.3.conv1.weight_g": "model.safetensors",
@@ -109,8 +121,12 @@
         "decoder.decode.3.conv2.weight_g": "model.safetensors",
         "decoder.decode.3.conv2.weight_v": "model.safetensors",
         "decoder.decode.3.norm1.fc.bias": "model.safetensors",
+        "decoder.decode.3.norm1.fc.biases": "model.safetensors",
+        "decoder.decode.3.norm1.fc.scales": "model.safetensors",
         "decoder.decode.3.norm1.fc.weight": "model.safetensors",
         "decoder.decode.3.norm2.fc.bias": "model.safetensors",
+        "decoder.decode.3.norm2.fc.biases": "model.safetensors",
+        "decoder.decode.3.norm2.fc.scales": "model.safetensors",
         "decoder.decode.3.norm2.fc.weight": "model.safetensors",
         "decoder.decode.3.pool.bias": "model.safetensors",
         "decoder.decode.3.pool.weight_g": "model.safetensors",
@@ -124,8 +140,12 @@
         "decoder.encode.conv2.weight_g": "model.safetensors",
         "decoder.encode.conv2.weight_v": "model.safetensors",
         "decoder.encode.norm1.fc.bias": "model.safetensors",
+        "decoder.encode.norm1.fc.biases": "model.safetensors",
+        "decoder.encode.norm1.fc.scales": "model.safetensors",
         "decoder.encode.norm1.fc.weight": "model.safetensors",
         "decoder.encode.norm2.fc.bias": "model.safetensors",
+        "decoder.encode.norm2.fc.biases": "model.safetensors",
+        "decoder.encode.norm2.fc.scales": "model.safetensors",
         "decoder.encode.norm2.fc.weight": "model.safetensors",
         "decoder.generator.conv_post.bias": "model.safetensors",
         "decoder.generator.conv_post.weight_g": "model.safetensors",
@@ -137,23 +157,35 @@
         "decoder.generator.noise_convs.1.bias": "model.safetensors",
         "decoder.generator.noise_convs.1.weight": "model.safetensors",
         "decoder.generator.noise_res.0.adain1.0.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.0.adain1.0.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.0.adain1.0.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.0.adain1.0.fc.weight": "model.safetensors",
         "decoder.generator.noise_res.0.adain1.1.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.0.adain1.1.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.0.adain1.1.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.0.adain1.1.fc.weight": "model.safetensors",
         "decoder.generator.noise_res.0.adain1.2.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.0.adain1.2.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.0.adain1.2.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.0.adain1.2.fc.weight": "model.safetensors",
         "decoder.generator.noise_res.0.adain2.0.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.0.adain2.0.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.0.adain2.0.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.0.adain2.0.fc.weight": "model.safetensors",
         "decoder.generator.noise_res.0.adain2.1.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.0.adain2.1.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.0.adain2.1.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.0.adain2.1.fc.weight": "model.safetensors",
         "decoder.generator.noise_res.0.adain2.2.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.0.adain2.2.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.0.adain2.2.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.0.adain2.2.fc.weight": "model.safetensors",
-        "decoder.generator.noise_res.0.alpha1.0": "model.safetensors",
-        "decoder.generator.noise_res.0.alpha1.1": "model.safetensors",
-        "decoder.generator.noise_res.0.alpha1.2": "model.safetensors",
-        "decoder.generator.noise_res.0.alpha2.0": "model.safetensors",
-        "decoder.generator.noise_res.0.alpha2.1": "model.safetensors",
-        "decoder.generator.noise_res.0.alpha2.2": "model.safetensors",
+        "decoder.generator.noise_res.0.alpha1_0": "model.safetensors",
+        "decoder.generator.noise_res.0.alpha1_1": "model.safetensors",
+        "decoder.generator.noise_res.0.alpha1_2": "model.safetensors",
+        "decoder.generator.noise_res.0.alpha2_0": "model.safetensors",
+        "decoder.generator.noise_res.0.alpha2_1": "model.safetensors",
+        "decoder.generator.noise_res.0.alpha2_2": "model.safetensors",
         "decoder.generator.noise_res.0.convs1.0.bias": "model.safetensors",
         "decoder.generator.noise_res.0.convs1.0.weight_g": "model.safetensors",
         "decoder.generator.noise_res.0.convs1.0.weight_v": "model.safetensors",
@@ -173,23 +205,35 @@
         "decoder.generator.noise_res.0.convs2.2.weight_g": "model.safetensors",
         "decoder.generator.noise_res.0.convs2.2.weight_v": "model.safetensors",
         "decoder.generator.noise_res.1.adain1.0.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.1.adain1.0.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.1.adain1.0.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.1.adain1.0.fc.weight": "model.safetensors",
         "decoder.generator.noise_res.1.adain1.1.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.1.adain1.1.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.1.adain1.1.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.1.adain1.1.fc.weight": "model.safetensors",
         "decoder.generator.noise_res.1.adain1.2.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.1.adain1.2.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.1.adain1.2.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.1.adain1.2.fc.weight": "model.safetensors",
         "decoder.generator.noise_res.1.adain2.0.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.1.adain2.0.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.1.adain2.0.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.1.adain2.0.fc.weight": "model.safetensors",
         "decoder.generator.noise_res.1.adain2.1.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.1.adain2.1.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.1.adain2.1.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.1.adain2.1.fc.weight": "model.safetensors",
         "decoder.generator.noise_res.1.adain2.2.fc.bias": "model.safetensors",
+        "decoder.generator.noise_res.1.adain2.2.fc.biases": "model.safetensors",
+        "decoder.generator.noise_res.1.adain2.2.fc.scales": "model.safetensors",
         "decoder.generator.noise_res.1.adain2.2.fc.weight": "model.safetensors",
-        "decoder.generator.noise_res.1.alpha1.0": "model.safetensors",
-        "decoder.generator.noise_res.1.alpha1.1": "model.safetensors",
-        "decoder.generator.noise_res.1.alpha1.2": "model.safetensors",
-        "decoder.generator.noise_res.1.alpha2.0": "model.safetensors",
-        "decoder.generator.noise_res.1.alpha2.1": "model.safetensors",
-        "decoder.generator.noise_res.1.alpha2.2": "model.safetensors",
+        "decoder.generator.noise_res.1.alpha1_0": "model.safetensors",
+        "decoder.generator.noise_res.1.alpha1_1": "model.safetensors",
+        "decoder.generator.noise_res.1.alpha1_2": "model.safetensors",
+        "decoder.generator.noise_res.1.alpha2_0": "model.safetensors",
+        "decoder.generator.noise_res.1.alpha2_1": "model.safetensors",
+        "decoder.generator.noise_res.1.alpha2_2": "model.safetensors",
         "decoder.generator.noise_res.1.convs1.0.bias": "model.safetensors",
         "decoder.generator.noise_res.1.convs1.0.weight_g": "model.safetensors",
         "decoder.generator.noise_res.1.convs1.0.weight_v": "model.safetensors",
@@ -209,23 +253,35 @@
         "decoder.generator.noise_res.1.convs2.2.weight_g": "model.safetensors",
         "decoder.generator.noise_res.1.convs2.2.weight_v": "model.safetensors",
         "decoder.generator.resblocks.0.adain1.0.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.0.adain1.0.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.0.adain1.0.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.0.adain1.0.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.0.adain1.1.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.0.adain1.1.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.0.adain1.1.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.0.adain1.1.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.0.adain1.2.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.0.adain1.2.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.0.adain1.2.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.0.adain1.2.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.0.adain2.0.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.0.adain2.0.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.0.adain2.0.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.0.adain2.0.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.0.adain2.1.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.0.adain2.1.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.0.adain2.1.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.0.adain2.1.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.0.adain2.2.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.0.adain2.2.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.0.adain2.2.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.0.adain2.2.fc.weight": "model.safetensors",
-        "decoder.generator.resblocks.0.alpha1.0": "model.safetensors",
-        "decoder.generator.resblocks.0.alpha1.1": "model.safetensors",
-        "decoder.generator.resblocks.0.alpha1.2": "model.safetensors",
-        "decoder.generator.resblocks.0.alpha2.0": "model.safetensors",
-        "decoder.generator.resblocks.0.alpha2.1": "model.safetensors",
-        "decoder.generator.resblocks.0.alpha2.2": "model.safetensors",
+        "decoder.generator.resblocks.0.alpha1_0": "model.safetensors",
+        "decoder.generator.resblocks.0.alpha1_1": "model.safetensors",
+        "decoder.generator.resblocks.0.alpha1_2": "model.safetensors",
+        "decoder.generator.resblocks.0.alpha2_0": "model.safetensors",
+        "decoder.generator.resblocks.0.alpha2_1": "model.safetensors",
+        "decoder.generator.resblocks.0.alpha2_2": "model.safetensors",
         "decoder.generator.resblocks.0.convs1.0.bias": "model.safetensors",
         "decoder.generator.resblocks.0.convs1.0.weight_g": "model.safetensors",
         "decoder.generator.resblocks.0.convs1.0.weight_v": "model.safetensors",
@@ -245,23 +301,35 @@
         "decoder.generator.resblocks.0.convs2.2.weight_g": "model.safetensors",
         "decoder.generator.resblocks.0.convs2.2.weight_v": "model.safetensors",
         "decoder.generator.resblocks.1.adain1.0.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.1.adain1.0.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.1.adain1.0.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.1.adain1.0.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.1.adain1.1.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.1.adain1.1.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.1.adain1.1.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.1.adain1.1.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.1.adain1.2.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.1.adain1.2.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.1.adain1.2.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.1.adain1.2.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.1.adain2.0.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.1.adain2.0.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.1.adain2.0.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.1.adain2.0.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.1.adain2.1.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.1.adain2.1.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.1.adain2.1.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.1.adain2.1.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.1.adain2.2.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.1.adain2.2.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.1.adain2.2.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.1.adain2.2.fc.weight": "model.safetensors",
-        "decoder.generator.resblocks.1.alpha1.0": "model.safetensors",
-        "decoder.generator.resblocks.1.alpha1.1": "model.safetensors",
-        "decoder.generator.resblocks.1.alpha1.2": "model.safetensors",
-        "decoder.generator.resblocks.1.alpha2.0": "model.safetensors",
-        "decoder.generator.resblocks.1.alpha2.1": "model.safetensors",
-        "decoder.generator.resblocks.1.alpha2.2": "model.safetensors",
+        "decoder.generator.resblocks.1.alpha1_0": "model.safetensors",
+        "decoder.generator.resblocks.1.alpha1_1": "model.safetensors",
+        "decoder.generator.resblocks.1.alpha1_2": "model.safetensors",
+        "decoder.generator.resblocks.1.alpha2_0": "model.safetensors",
+        "decoder.generator.resblocks.1.alpha2_1": "model.safetensors",
+        "decoder.generator.resblocks.1.alpha2_2": "model.safetensors",
         "decoder.generator.resblocks.1.convs1.0.bias": "model.safetensors",
         "decoder.generator.resblocks.1.convs1.0.weight_g": "model.safetensors",
         "decoder.generator.resblocks.1.convs1.0.weight_v": "model.safetensors",
@@ -281,23 +349,35 @@
         "decoder.generator.resblocks.1.convs2.2.weight_g": "model.safetensors",
         "decoder.generator.resblocks.1.convs2.2.weight_v": "model.safetensors",
         "decoder.generator.resblocks.2.adain1.0.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.2.adain1.0.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.2.adain1.0.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.2.adain1.0.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.2.adain1.1.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.2.adain1.1.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.2.adain1.1.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.2.adain1.1.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.2.adain1.2.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.2.adain1.2.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.2.adain1.2.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.2.adain1.2.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.2.adain2.0.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.2.adain2.0.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.2.adain2.0.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.2.adain2.0.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.2.adain2.1.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.2.adain2.1.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.2.adain2.1.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.2.adain2.1.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.2.adain2.2.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.2.adain2.2.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.2.adain2.2.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.2.adain2.2.fc.weight": "model.safetensors",
-        "decoder.generator.resblocks.2.alpha1.0": "model.safetensors",
-        "decoder.generator.resblocks.2.alpha1.1": "model.safetensors",
-        "decoder.generator.resblocks.2.alpha1.2": "model.safetensors",
-        "decoder.generator.resblocks.2.alpha2.0": "model.safetensors",
-        "decoder.generator.resblocks.2.alpha2.1": "model.safetensors",
-        "decoder.generator.resblocks.2.alpha2.2": "model.safetensors",
+        "decoder.generator.resblocks.2.alpha1_0": "model.safetensors",
+        "decoder.generator.resblocks.2.alpha1_1": "model.safetensors",
+        "decoder.generator.resblocks.2.alpha1_2": "model.safetensors",
+        "decoder.generator.resblocks.2.alpha2_0": "model.safetensors",
+        "decoder.generator.resblocks.2.alpha2_1": "model.safetensors",
+        "decoder.generator.resblocks.2.alpha2_2": "model.safetensors",
         "decoder.generator.resblocks.2.convs1.0.bias": "model.safetensors",
         "decoder.generator.resblocks.2.convs1.0.weight_g": "model.safetensors",
         "decoder.generator.resblocks.2.convs1.0.weight_v": "model.safetensors",
@@ -317,23 +397,35 @@
         "decoder.generator.resblocks.2.convs2.2.weight_g": "model.safetensors",
         "decoder.generator.resblocks.2.convs2.2.weight_v": "model.safetensors",
         "decoder.generator.resblocks.3.adain1.0.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.3.adain1.0.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.3.adain1.0.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.3.adain1.0.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.3.adain1.1.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.3.adain1.1.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.3.adain1.1.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.3.adain1.1.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.3.adain1.2.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.3.adain1.2.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.3.adain1.2.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.3.adain1.2.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.3.adain2.0.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.3.adain2.0.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.3.adain2.0.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.3.adain2.0.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.3.adain2.1.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.3.adain2.1.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.3.adain2.1.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.3.adain2.1.fc.weight": "model.safetensors",
         "decoder.generator.resblocks.3.adain2.2.fc.bias": "model.safetensors",
+        "decoder.generator.resblocks.3.adain2.2.fc.biases": "model.safetensors",
+        "decoder.generator.resblocks.3.adain2.2.fc.scales": "model.safetensors",
         "decoder.generator.resblocks.3.adain2.2.fc.weight": "model.safetensors",
-        "decoder.generator.resblocks.3.alpha1.0": "model.safetensors",
-        "decoder.generator.resblocks.3.alpha1.1": "model.safetensors",
-        "decoder.generator.resblocks.3.alpha1.2": "model.safetensors",
-        "decoder.generator.resblocks.3.alpha2.0": "model.safetensors",
-        "decoder.generator.resblocks.3.alpha2.1": "model.safetensors",
-        "decoder.generator.resblocks.3.alpha2.2": "model.safetensors",
+        "decoder.generator.resblocks.3.alpha1_0": "model.safetensors",
+        "decoder.generator.resblocks.3.alpha1_1": "model.safetensors",
+        "decoder.generator.resblocks.3.alpha1_2": "model.safetensors",
+        "decoder.generator.resblocks.3.alpha2_0": "model.safetensors",
+        "decoder.generator.resblocks.3.alpha2_1": "model.safetensors",
+        "decoder.generator.resblocks.3.alpha2_2": "model.safetensors",
         "decoder.generator.resblocks.3.convs1.0.bias": "model.safetensors",
         "decoder.generator.resblocks.3.convs1.0.weight_g": "model.safetensors",
         "decoder.generator.resblocks.3.convs1.0.weight_v": "model.safetensors",