################################################################################ QWEN3-TTS MODEL INSPECTION REPORT ################################################################################ This report documents all tensors, shapes, and architecture details for GGUF conversion planning. ################################################################################ TTS BASE MODEL (Qwen3-TTS-12Hz-0.6B-Base) ################################################################################ ================================================================================ CONFIG: TTS Base ================================================================================ { "architectures": [ "Qwen3TTSForConditionalGeneration" ], "assistant_token_id": 77091, "im_end_token_id": 151645, "im_start_token_id": 151644, "tts_bos_token_id": 151672, "tts_eos_token_id": 151673, "tts_pad_token_id": 151671, "model_type": "qwen3_tts", "tokenizer_type": "qwen3_tts_tokenizer_12hz", "tts_model_size": "0b6", "tts_model_type": "base", "speaker_encoder_config": { "enc_dim": 1024, "sample_rate": 24000 }, "talker_config": { "attention_bias": false, "attention_dropout": 0, "code_predictor_config": { "_name_or_path": "", "add_cross_attention": false, "architectures": null, "attention_bias": false, "attention_dropout": 0, "bad_words_ids": null, "begin_suppress_tokens": null, "bos_token_id": null, "chunk_size_feed_forward": 0, "cross_attention_hidden_size": null, "decoder_start_token_id": null, "diversity_penalty": 0.0, "do_sample": false, "early_stopping": false, "encoder_no_repeat_ngram_size": 0, "eos_token_id": null, "exponential_decay_length_penalty": null, "finetuning_task": null, "forced_bos_token_id": null, "forced_eos_token_id": null, "head_dim": 128, "hidden_act": "silu", "hidden_size": 1024, "id2label": { "0": "LABEL_0", "1": "LABEL_1" }, "initializer_range": 0.02, "intermediate_size": 3072, "is_decoder": false, "is_encoder_decoder": false, "label2id": { "LABEL_0": 0, "LABEL_1": 1 }, "layer_types": [ "full_attention", "full_attention", "full_attention", "full_attention", "full_attention" ], "length_penalty": 1.0, "max_length": 20, "max_position_embeddings": 65536, "max_window_layers": 28, "min_length": 0, "model_type": "qwen3_tts_talker_code_predictor", "no_repeat_ngram_size": 0, "num_attention_heads": 16, "num_beam_groups": 1, "num_beams": 1, "num_code_groups": 16, "num_hidden_layers": 5, "num_key_value_heads": 8, "num_return_sequences": 1, "output_attentions": false, "output_hidden_states": false, "output_scores": false, "pad_token_id": null, "prefix": null, "problem_type": null, "pruned_heads": {}, "remove_invalid_values": false, "repetition_penalty": 1.0, "return_dict": true, "return_dict_in_generate": false, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 1000000, "sep_token_id": null, "sliding_window": null, "suppress_tokens": null, "task_specific_params": null, "temperature": 1.0, "tf_legacy_loss": false, "tie_encoder_decoder": false, "tie_word_embeddings": false, "tokenizer_class": null, "top_k": 50, "top_p": 1.0, "dtype": null, "torchscript": false, "typical_p": 1.0, "use_bfloat16": false, "use_cache": true, "use_sliding_window": false, "vocab_size": 2048 }, "codec_bos_id": 2149, "codec_eos_token_id": 2150, "codec_think_id": 2154, "codec_language_id": { "chinese": 2055, "english": 2050, "german": 2053, "italian": 2070, "portuguese": 2071, "spanish": 2054, "japanese": 2058, "korean": 2064, "french": 2061, "russian": 2069 }, "codec_nothink_id": 2155, "codec_pad_id": 2148, "codec_think_bos_id": 2156, "codec_think_eos_id": 2157, "spk_id": {}, "spk_is_dialect": {}, "head_dim": 128, "hidden_act": "silu", "hidden_size": 1024, "initializer_range": 0.02, "intermediate_size": 3072, "max_position_embeddings": 32768, "model_type": "qwen3_tts_talker", "num_attention_heads": 16, "num_code_groups": 16, "num_hidden_layers": 28, "num_key_value_heads": 8, "position_id_per_seconds": 13, "rms_norm_eps": 1e-06, "rope_scaling": { "interleaved": true, "mrope_section": [ 24, 20, 20 ], "rope_type": "default", "type": "default" }, "rope_theta": 1000000, "sliding_window": null, "text_hidden_size": 2048, "text_vocab_size": 151936, "use_cache": true, "use_sliding_window": false, "vocab_size": 3072 }, "transformers_version": "4.57.3" } -------------------------------------------------------------------------------- ARCHITECTURE ANALYSIS: TTS Base -------------------------------------------------------------------------------- Architecture: ['Qwen3TTSForConditionalGeneration'] Model Type: qwen3_tts --- Talker (Main TTS Transformer) --- Hidden Size: 1024 Intermediate Size: 3072 Num Hidden Layers: 28 Num Attention Heads: 16 Num KV Heads: 8 Head Dim: 128 Vocab Size: 3072 Text Vocab Size: 151936 Text Hidden Size: 2048 Num Code Groups (Codebooks): 16 RMS Norm Eps: 1e-06 RoPE Theta: 1000000 Max Position Embeddings: 32768 Hidden Act: silu RoPE Scaling: {'interleaved': True, 'mrope_section': [24, 20, 20], 'rope_type': 'default', 'type': 'default'} --- Code Predictor (Delay Pattern Transformer) --- Hidden Size: 1024 Intermediate Size: 3072 Num Hidden Layers: 5 Num Attention Heads: 16 Num KV Heads: 8 Head Dim: 128 Vocab Size: 2048 Num Code Groups: 16 Layer Types: ['full_attention', 'full_attention', 'full_attention', 'full_attention', 'full_attention'] --- Speaker Encoder --- Enc Dim: 1024 Sample Rate: 24000 ================================================================================ TENSORS: TTS Base Main ================================================================================ Total tensors: 478 Tensor Name Shape Dtype -------------------------------------------------------------------------------------------------------------- speaker_encoder.asp.conv.bias (1536,) torch.bfloat16 speaker_encoder.asp.conv.weight (1536, 128, 1) torch.bfloat16 speaker_encoder.asp.tdnn.conv.bias (128,) torch.bfloat16 speaker_encoder.asp.tdnn.conv.weight (128, 4608, 1) torch.bfloat16 speaker_encoder.blocks.0.conv.bias (512,) torch.bfloat16 speaker_encoder.blocks.0.conv.weight (512, 128, 5) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.0.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.0.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.1.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.1.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.2.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.2.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.3.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.3.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.4.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.4.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.5.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.5.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.6.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.1.res2net_block.blocks.6.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.1.se_block.conv1.bias (128,) torch.bfloat16 speaker_encoder.blocks.1.se_block.conv1.weight (128, 512, 1) torch.bfloat16 speaker_encoder.blocks.1.se_block.conv2.bias (512,) torch.bfloat16 speaker_encoder.blocks.1.se_block.conv2.weight (512, 128, 1) torch.bfloat16 speaker_encoder.blocks.1.tdnn1.conv.bias (512,) torch.bfloat16 speaker_encoder.blocks.1.tdnn1.conv.weight (512, 512, 1) torch.bfloat16 speaker_encoder.blocks.1.tdnn2.conv.bias (512,) torch.bfloat16 speaker_encoder.blocks.1.tdnn2.conv.weight (512, 512, 1) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.0.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.0.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.1.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.1.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.2.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.2.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.3.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.3.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.4.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.4.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.5.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.5.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.6.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.2.res2net_block.blocks.6.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.2.se_block.conv1.bias (128,) torch.bfloat16 speaker_encoder.blocks.2.se_block.conv1.weight (128, 512, 1) torch.bfloat16 speaker_encoder.blocks.2.se_block.conv2.bias (512,) torch.bfloat16 speaker_encoder.blocks.2.se_block.conv2.weight (512, 128, 1) torch.bfloat16 speaker_encoder.blocks.2.tdnn1.conv.bias (512,) torch.bfloat16 speaker_encoder.blocks.2.tdnn1.conv.weight (512, 512, 1) torch.bfloat16 speaker_encoder.blocks.2.tdnn2.conv.bias (512,) torch.bfloat16 speaker_encoder.blocks.2.tdnn2.conv.weight (512, 512, 1) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.0.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.0.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.1.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.1.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.2.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.2.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.3.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.3.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.4.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.4.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.5.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.5.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.6.conv.bias (64,) torch.bfloat16 speaker_encoder.blocks.3.res2net_block.blocks.6.conv.weight (64, 64, 3) torch.bfloat16 speaker_encoder.blocks.3.se_block.conv1.bias (128,) torch.bfloat16 speaker_encoder.blocks.3.se_block.conv1.weight (128, 512, 1) torch.bfloat16 speaker_encoder.blocks.3.se_block.conv2.bias (512,) torch.bfloat16 speaker_encoder.blocks.3.se_block.conv2.weight (512, 128, 1) torch.bfloat16 speaker_encoder.blocks.3.tdnn1.conv.bias (512,) torch.bfloat16 speaker_encoder.blocks.3.tdnn1.conv.weight (512, 512, 1) torch.bfloat16 speaker_encoder.blocks.3.tdnn2.conv.bias (512,) torch.bfloat16 speaker_encoder.blocks.3.tdnn2.conv.weight (512, 512, 1) torch.bfloat16 speaker_encoder.fc.bias (1024,) torch.bfloat16 speaker_encoder.fc.weight (1024, 3072, 1) torch.bfloat16 speaker_encoder.mfa.conv.bias (1536,) torch.bfloat16 speaker_encoder.mfa.conv.weight (1536, 1536, 1) torch.bfloat16 talker.code_predictor.lm_head.0.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.1.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.10.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.11.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.12.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.13.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.14.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.2.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.3.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.4.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.5.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.6.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.7.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.8.weight (2048, 1024) torch.bfloat16 talker.code_predictor.lm_head.9.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.0.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.1.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.10.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.11.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.12.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.13.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.14.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.2.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.3.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.4.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.5.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.6.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.7.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.8.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.codec_embedding.9.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.layers.0.input_layernorm.weight (1024,) torch.bfloat16 talker.code_predictor.model.layers.0.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.code_predictor.model.layers.0.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.code_predictor.model.layers.0.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.code_predictor.model.layers.0.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.code_predictor.model.layers.0.self_attn.k_norm.weight (128,) torch.bfloat16 talker.code_predictor.model.layers.0.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.code_predictor.model.layers.0.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.code_predictor.model.layers.0.self_attn.q_norm.weight (128,) torch.bfloat16 talker.code_predictor.model.layers.0.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.layers.0.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.code_predictor.model.layers.1.input_layernorm.weight (1024,) torch.bfloat16 talker.code_predictor.model.layers.1.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.code_predictor.model.layers.1.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.code_predictor.model.layers.1.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.code_predictor.model.layers.1.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.code_predictor.model.layers.1.self_attn.k_norm.weight (128,) torch.bfloat16 talker.code_predictor.model.layers.1.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.code_predictor.model.layers.1.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.code_predictor.model.layers.1.self_attn.q_norm.weight (128,) torch.bfloat16 talker.code_predictor.model.layers.1.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.layers.1.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.code_predictor.model.layers.2.input_layernorm.weight (1024,) torch.bfloat16 talker.code_predictor.model.layers.2.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.code_predictor.model.layers.2.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.code_predictor.model.layers.2.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.code_predictor.model.layers.2.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.code_predictor.model.layers.2.self_attn.k_norm.weight (128,) torch.bfloat16 talker.code_predictor.model.layers.2.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.code_predictor.model.layers.2.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.code_predictor.model.layers.2.self_attn.q_norm.weight (128,) torch.bfloat16 talker.code_predictor.model.layers.2.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.layers.2.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.code_predictor.model.layers.3.input_layernorm.weight (1024,) torch.bfloat16 talker.code_predictor.model.layers.3.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.code_predictor.model.layers.3.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.code_predictor.model.layers.3.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.code_predictor.model.layers.3.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.code_predictor.model.layers.3.self_attn.k_norm.weight (128,) torch.bfloat16 talker.code_predictor.model.layers.3.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.code_predictor.model.layers.3.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.code_predictor.model.layers.3.self_attn.q_norm.weight (128,) torch.bfloat16 talker.code_predictor.model.layers.3.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.layers.3.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.code_predictor.model.layers.4.input_layernorm.weight (1024,) torch.bfloat16 talker.code_predictor.model.layers.4.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.code_predictor.model.layers.4.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.code_predictor.model.layers.4.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.code_predictor.model.layers.4.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.code_predictor.model.layers.4.self_attn.k_norm.weight (128,) torch.bfloat16 talker.code_predictor.model.layers.4.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.code_predictor.model.layers.4.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.code_predictor.model.layers.4.self_attn.q_norm.weight (128,) torch.bfloat16 talker.code_predictor.model.layers.4.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.code_predictor.model.layers.4.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.code_predictor.model.norm.weight (1024,) torch.bfloat16 talker.codec_head.weight (3072, 1024) torch.bfloat16 talker.model.codec_embedding.weight (3072, 1024) torch.bfloat16 talker.model.layers.0.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.0.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.0.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.0.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.0.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.0.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.0.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.0.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.0.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.0.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.0.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.1.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.1.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.1.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.1.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.1.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.1.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.1.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.1.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.1.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.1.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.1.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.10.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.10.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.10.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.10.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.10.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.10.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.10.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.10.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.10.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.10.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.10.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.11.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.11.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.11.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.11.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.11.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.11.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.11.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.11.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.11.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.11.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.11.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.12.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.12.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.12.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.12.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.12.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.12.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.12.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.12.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.12.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.12.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.12.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.13.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.13.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.13.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.13.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.13.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.13.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.13.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.13.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.13.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.13.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.13.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.14.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.14.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.14.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.14.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.14.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.14.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.14.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.14.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.14.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.14.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.14.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.15.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.15.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.15.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.15.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.15.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.15.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.15.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.15.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.15.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.15.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.15.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.16.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.16.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.16.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.16.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.16.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.16.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.16.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.16.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.16.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.16.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.16.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.17.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.17.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.17.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.17.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.17.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.17.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.17.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.17.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.17.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.17.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.17.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.18.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.18.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.18.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.18.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.18.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.18.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.18.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.18.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.18.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.18.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.18.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.19.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.19.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.19.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.19.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.19.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.19.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.19.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.19.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.19.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.19.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.19.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.2.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.2.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.2.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.2.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.2.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.2.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.2.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.2.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.2.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.2.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.2.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.20.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.20.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.20.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.20.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.20.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.20.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.20.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.20.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.20.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.20.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.20.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.21.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.21.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.21.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.21.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.21.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.21.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.21.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.21.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.21.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.21.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.21.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.22.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.22.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.22.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.22.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.22.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.22.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.22.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.22.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.22.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.22.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.22.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.23.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.23.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.23.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.23.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.23.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.23.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.23.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.23.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.23.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.23.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.23.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.24.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.24.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.24.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.24.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.24.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.24.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.24.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.24.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.24.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.24.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.24.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.25.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.25.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.25.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.25.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.25.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.25.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.25.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.25.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.25.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.25.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.25.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.26.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.26.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.26.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.26.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.26.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.26.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.26.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.26.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.26.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.26.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.26.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.27.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.27.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.27.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.27.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.27.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.27.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.27.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.27.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.27.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.27.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.27.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.3.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.3.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.3.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.3.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.3.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.3.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.3.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.3.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.3.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.3.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.3.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.4.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.4.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.4.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.4.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.4.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.4.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.4.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.4.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.4.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.4.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.4.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.5.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.5.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.5.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.5.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.5.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.5.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.5.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.5.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.5.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.5.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.5.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.6.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.6.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.6.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.6.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.6.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.6.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.6.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.6.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.6.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.6.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.6.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.7.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.7.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.7.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.7.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.7.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.7.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.7.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.7.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.7.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.7.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.7.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.8.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.8.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.8.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.8.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.8.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.8.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.8.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.8.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.8.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.8.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.8.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.9.input_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.9.mlp.down_proj.weight (1024, 3072) torch.bfloat16 talker.model.layers.9.mlp.gate_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.9.mlp.up_proj.weight (3072, 1024) torch.bfloat16 talker.model.layers.9.post_attention_layernorm.weight (1024,) torch.bfloat16 talker.model.layers.9.self_attn.k_norm.weight (128,) torch.bfloat16 talker.model.layers.9.self_attn.k_proj.weight (1024, 1024) torch.bfloat16 talker.model.layers.9.self_attn.o_proj.weight (1024, 2048) torch.bfloat16 talker.model.layers.9.self_attn.q_norm.weight (128,) torch.bfloat16 talker.model.layers.9.self_attn.q_proj.weight (2048, 1024) torch.bfloat16 talker.model.layers.9.self_attn.v_proj.weight (1024, 1024) torch.bfloat16 talker.model.norm.weight (1024,) torch.bfloat16 talker.model.text_embedding.weight (151936, 2048) torch.bfloat16 talker.text_projection.linear_fc1.bias (2048,) torch.bfloat16 talker.text_projection.linear_fc1.weight (2048, 2048) torch.bfloat16 talker.text_projection.linear_fc2.bias (1024,) torch.bfloat16 talker.text_projection.linear_fc2.weight (1024, 2048) torch.bfloat16 -------------------------------------------------------------------------------- TENSOR CATEGORIES: TTS Base Main -------------------------------------------------------------------------------- speaker_encoder.asp (4 tensors): - speaker_encoder.asp.conv.bias - speaker_encoder.asp.conv.weight - speaker_encoder.asp.tdnn.conv.bias - speaker_encoder.asp.tdnn.conv.weight speaker_encoder.blocks (68 tensors): - speaker_encoder.blocks.0.conv.bias - speaker_encoder.blocks.0.conv.weight - speaker_encoder.blocks.1.res2net_block.blocks.0.conv.bias - speaker_encoder.blocks.1.res2net_block.blocks.0.conv.weight - speaker_encoder.blocks.1.res2net_block.blocks.1.conv.bias ... and 63 more speaker_encoder.fc (2 tensors): - speaker_encoder.fc.bias - speaker_encoder.fc.weight speaker_encoder.mfa (2 tensors): - speaker_encoder.mfa.conv.bias - speaker_encoder.mfa.conv.weight talker.code_predictor (86 tensors): - talker.code_predictor.lm_head.0.weight - talker.code_predictor.lm_head.1.weight - talker.code_predictor.lm_head.10.weight - talker.code_predictor.lm_head.11.weight - talker.code_predictor.lm_head.12.weight ... and 81 more talker.codec_head (1 tensors): - talker.codec_head.weight talker.model (311 tensors): - talker.model.codec_embedding.weight - talker.model.layers.0.input_layernorm.weight - talker.model.layers.0.mlp.down_proj.weight - talker.model.layers.0.mlp.gate_proj.weight - talker.model.layers.0.mlp.up_proj.weight ... and 306 more talker.text_projection (4 tensors): - talker.text_projection.linear_fc1.bias - talker.text_projection.linear_fc1.weight - talker.text_projection.linear_fc2.bias - talker.text_projection.linear_fc2.weight ================================================================================ CONFIG: TTS Base Speech Tokenizer ================================================================================ { "architectures": [ "Qwen3TTSTokenizerV2Model" ], "model_type": "qwen3_tts_tokenizer_12hz", "encoder_valid_num_quantizers": 16, "input_sample_rate": 24000, "output_sample_rate": 24000, "decode_upsample_rate": 1920, "encode_downsample_rate": 1920, "decoder_config": { "attention_bias": false, "attention_dropout": 0.0, "latent_dim": 1024, "codebook_dim": 512, "codebook_size": 2048, "decoder_dim": 1536, "hidden_act": "silu", "hidden_size": 512, "intermediate_size": 1024, "layer_scale_initial_scale": 0.01, "max_position_embeddings": 8000, "head_dim": 64, "num_attention_heads": 16, "num_hidden_layers": 8, "num_key_value_heads": 16, "num_quantizers": 16, "num_semantic_quantizers": 1, "rms_norm_eps": 1e-05, "rope_theta": 10000, "semantic_codebook_size": 4096, "sliding_window": 72, "upsample_rates": [ 8, 5, 4, 3 ], "upsampling_ratios": [ 2, 2 ], "vector_quantization_hidden_dimension": 512 }, "encoder_config": { "_frame_rate": 12.5, "attention_bias": false, "attention_dropout": 0.0, "audio_channels": 1, "codebook_dim": 256, "codebook_size": 2048, "compress": 2, "dilation_growth_rate": 2, "dtype": "float32", "head_dim": 64, "hidden_act": "gelu", "hidden_size": 512, "initializer_range": 0.02, "intermediate_size": 2048, "kernel_size": 7, "last_kernel_size": 3, "layer_scale_initial_scale": 0.01, "max_position_embeddings": 8000, "norm_eps": 1e-05, "normalize": false, "num_attention_heads": 8, "num_filters": 64, "num_hidden_layers": 8, "num_key_value_heads": 8, "num_quantizers": 32, "num_residual_layers": 1, "num_semantic_quantizers": 1, "pad_mode": "constant", "residual_kernel_size": 3, "rope_theta": 10000.0, "sampling_rate": 24000, "sliding_window": 250, "transformers_version": "4.57.0.dev0", "trim_right_ratio": 1.0, "upsample_groups": 512, "upsampling_ratios": [ 8, 6, 5, 4 ], "use_cache": false, "use_causal_conv": true, "use_conv_shortcut": false, "use_streaming": false, "vector_quantization_hidden_dimension": 256 }, "transformers_version": "4.57.3" } -------------------------------------------------------------------------------- ARCHITECTURE ANALYSIS: TTS Base Speech Tokenizer -------------------------------------------------------------------------------- Architecture: ['Qwen3TTSTokenizerV2Model'] Model Type: qwen3_tts_tokenizer_12hz --- Tokenizer Encoder (Audio -> Codes) --- Frame Rate: 12.5 Hidden Size: 512 Intermediate Size: 2048 Num Hidden Layers: 8 Num Attention Heads: 8 Num KV Heads: 8 Head Dim: 64 Codebook Size: 2048 Codebook Dim: 256 Num Quantizers: 32 Num Semantic Quantizers: 1 Sampling Rate: 24000 Upsampling Ratios: [8, 6, 5, 4] Num Filters: 64 Kernel Size: 7 Use Causal Conv: True --- Tokenizer Decoder (Codes -> Audio, Vocoder) --- Latent Dim: 1024 Decoder Dim: 1536 Hidden Size: 512 Intermediate Size: 1024 Num Hidden Layers: 8 Num Attention Heads: 16 Num KV Heads: 16 Head Dim: 64 Codebook Size: 2048 Codebook Dim: 512 Semantic Codebook Size: 4096 Num Quantizers: 16 Num Semantic Quantizers: 1 Upsample Rates: [8, 5, 4, 3] Upsampling Ratios: [2, 2] Sliding Window: 72 VQ Hidden Dim: 512 --- Tokenizer Global --- Valid Num Quantizers: 16 Input Sample Rate: 24000 Output Sample Rate: 24000 Encode Downsample Rate: 1920 Decode Upsample Rate: 1920 ================================================================================ TENSORS: TTS Base Speech Tokenizer ================================================================================ Total tensors: 496 Tensor Name Shape Dtype -------------------------------------------------------------------------------------------------------------- decoder.decoder.0.conv.bias (1536,) torch.float32 decoder.decoder.0.conv.weight (1536, 1024, 7) torch.float32 decoder.decoder.1.block.0.alpha (1536,) torch.float32 decoder.decoder.1.block.0.beta (1536,) torch.float32 decoder.decoder.1.block.1.conv.bias (768,) torch.float32 decoder.decoder.1.block.1.conv.weight (1536, 768, 16) torch.float32 decoder.decoder.1.block.2.act1.alpha (768,) torch.float32 decoder.decoder.1.block.2.act1.beta (768,) torch.float32 decoder.decoder.1.block.2.act2.alpha (768,) torch.float32 decoder.decoder.1.block.2.act2.beta (768,) torch.float32 decoder.decoder.1.block.2.conv1.conv.bias (768,) torch.float32 decoder.decoder.1.block.2.conv1.conv.weight (768, 768, 7) torch.float32 decoder.decoder.1.block.2.conv2.conv.bias (768,) torch.float32 decoder.decoder.1.block.2.conv2.conv.weight (768, 768, 1) torch.float32 decoder.decoder.1.block.3.act1.alpha (768,) torch.float32 decoder.decoder.1.block.3.act1.beta (768,) torch.float32 decoder.decoder.1.block.3.act2.alpha (768,) torch.float32 decoder.decoder.1.block.3.act2.beta (768,) torch.float32 decoder.decoder.1.block.3.conv1.conv.bias (768,) torch.float32 decoder.decoder.1.block.3.conv1.conv.weight (768, 768, 7) torch.float32 decoder.decoder.1.block.3.conv2.conv.bias (768,) torch.float32 decoder.decoder.1.block.3.conv2.conv.weight (768, 768, 1) torch.float32 decoder.decoder.1.block.4.act1.alpha (768,) torch.float32 decoder.decoder.1.block.4.act1.beta (768,) torch.float32 decoder.decoder.1.block.4.act2.alpha (768,) torch.float32 decoder.decoder.1.block.4.act2.beta (768,) torch.float32 decoder.decoder.1.block.4.conv1.conv.bias (768,) torch.float32 decoder.decoder.1.block.4.conv1.conv.weight (768, 768, 7) torch.float32 decoder.decoder.1.block.4.conv2.conv.bias (768,) torch.float32 decoder.decoder.1.block.4.conv2.conv.weight (768, 768, 1) torch.float32 decoder.decoder.2.block.0.alpha (768,) torch.float32 decoder.decoder.2.block.0.beta (768,) torch.float32 decoder.decoder.2.block.1.conv.bias (384,) torch.float32 decoder.decoder.2.block.1.conv.weight (768, 384, 10) torch.float32 decoder.decoder.2.block.2.act1.alpha (384,) torch.float32 decoder.decoder.2.block.2.act1.beta (384,) torch.float32 decoder.decoder.2.block.2.act2.alpha (384,) torch.float32 decoder.decoder.2.block.2.act2.beta (384,) torch.float32 decoder.decoder.2.block.2.conv1.conv.bias (384,) torch.float32 decoder.decoder.2.block.2.conv1.conv.weight (384, 384, 7) torch.float32 decoder.decoder.2.block.2.conv2.conv.bias (384,) torch.float32 decoder.decoder.2.block.2.conv2.conv.weight (384, 384, 1) torch.float32 decoder.decoder.2.block.3.act1.alpha (384,) torch.float32 decoder.decoder.2.block.3.act1.beta (384,) torch.float32 decoder.decoder.2.block.3.act2.alpha (384,) torch.float32 decoder.decoder.2.block.3.act2.beta (384,) torch.float32 decoder.decoder.2.block.3.conv1.conv.bias (384,) torch.float32 decoder.decoder.2.block.3.conv1.conv.weight (384, 384, 7) torch.float32 decoder.decoder.2.block.3.conv2.conv.bias (384,) torch.float32 decoder.decoder.2.block.3.conv2.conv.weight (384, 384, 1) torch.float32 decoder.decoder.2.block.4.act1.alpha (384,) torch.float32 decoder.decoder.2.block.4.act1.beta (384,) torch.float32 decoder.decoder.2.block.4.act2.alpha (384,) torch.float32 decoder.decoder.2.block.4.act2.beta (384,) torch.float32 decoder.decoder.2.block.4.conv1.conv.bias (384,) torch.float32 decoder.decoder.2.block.4.conv1.conv.weight (384, 384, 7) torch.float32 decoder.decoder.2.block.4.conv2.conv.bias (384,) torch.float32 decoder.decoder.2.block.4.conv2.conv.weight (384, 384, 1) torch.float32 decoder.decoder.3.block.0.alpha (384,) torch.float32 decoder.decoder.3.block.0.beta (384,) torch.float32 decoder.decoder.3.block.1.conv.bias (192,) torch.float32 decoder.decoder.3.block.1.conv.weight (384, 192, 8) torch.float32 decoder.decoder.3.block.2.act1.alpha (192,) torch.float32 decoder.decoder.3.block.2.act1.beta (192,) torch.float32 decoder.decoder.3.block.2.act2.alpha (192,) torch.float32 decoder.decoder.3.block.2.act2.beta (192,) torch.float32 decoder.decoder.3.block.2.conv1.conv.bias (192,) torch.float32 decoder.decoder.3.block.2.conv1.conv.weight (192, 192, 7) torch.float32 decoder.decoder.3.block.2.conv2.conv.bias (192,) torch.float32 decoder.decoder.3.block.2.conv2.conv.weight (192, 192, 1) torch.float32 decoder.decoder.3.block.3.act1.alpha (192,) torch.float32 decoder.decoder.3.block.3.act1.beta (192,) torch.float32 decoder.decoder.3.block.3.act2.alpha (192,) torch.float32 decoder.decoder.3.block.3.act2.beta (192,) torch.float32 decoder.decoder.3.block.3.conv1.conv.bias (192,) torch.float32 decoder.decoder.3.block.3.conv1.conv.weight (192, 192, 7) torch.float32 decoder.decoder.3.block.3.conv2.conv.bias (192,) torch.float32 decoder.decoder.3.block.3.conv2.conv.weight (192, 192, 1) torch.float32 decoder.decoder.3.block.4.act1.alpha (192,) torch.float32 decoder.decoder.3.block.4.act1.beta (192,) torch.float32 decoder.decoder.3.block.4.act2.alpha (192,) torch.float32 decoder.decoder.3.block.4.act2.beta (192,) torch.float32 decoder.decoder.3.block.4.conv1.conv.bias (192,) torch.float32 decoder.decoder.3.block.4.conv1.conv.weight (192, 192, 7) torch.float32 decoder.decoder.3.block.4.conv2.conv.bias (192,) torch.float32 decoder.decoder.3.block.4.conv2.conv.weight (192, 192, 1) torch.float32 decoder.decoder.4.block.0.alpha (192,) torch.float32 decoder.decoder.4.block.0.beta (192,) torch.float32 decoder.decoder.4.block.1.conv.bias (96,) torch.float32 decoder.decoder.4.block.1.conv.weight (192, 96, 6) torch.float32 decoder.decoder.4.block.2.act1.alpha (96,) torch.float32 decoder.decoder.4.block.2.act1.beta (96,) torch.float32 decoder.decoder.4.block.2.act2.alpha (96,) torch.float32 decoder.decoder.4.block.2.act2.beta (96,) torch.float32 decoder.decoder.4.block.2.conv1.conv.bias (96,) torch.float32 decoder.decoder.4.block.2.conv1.conv.weight (96, 96, 7) torch.float32 decoder.decoder.4.block.2.conv2.conv.bias (96,) torch.float32 decoder.decoder.4.block.2.conv2.conv.weight (96, 96, 1) torch.float32 decoder.decoder.4.block.3.act1.alpha (96,) torch.float32 decoder.decoder.4.block.3.act1.beta (96,) torch.float32 decoder.decoder.4.block.3.act2.alpha (96,) torch.float32 decoder.decoder.4.block.3.act2.beta (96,) torch.float32 decoder.decoder.4.block.3.conv1.conv.bias (96,) torch.float32 decoder.decoder.4.block.3.conv1.conv.weight (96, 96, 7) torch.float32 decoder.decoder.4.block.3.conv2.conv.bias (96,) torch.float32 decoder.decoder.4.block.3.conv2.conv.weight (96, 96, 1) torch.float32 decoder.decoder.4.block.4.act1.alpha (96,) torch.float32 decoder.decoder.4.block.4.act1.beta (96,) torch.float32 decoder.decoder.4.block.4.act2.alpha (96,) torch.float32 decoder.decoder.4.block.4.act2.beta (96,) torch.float32 decoder.decoder.4.block.4.conv1.conv.bias (96,) torch.float32 decoder.decoder.4.block.4.conv1.conv.weight (96, 96, 7) torch.float32 decoder.decoder.4.block.4.conv2.conv.bias (96,) torch.float32 decoder.decoder.4.block.4.conv2.conv.weight (96, 96, 1) torch.float32 decoder.decoder.5.alpha (96,) torch.float32 decoder.decoder.5.beta (96,) torch.float32 decoder.decoder.6.conv.bias (1,) torch.float32 decoder.decoder.6.conv.weight (1, 96, 7) torch.float32 decoder.pre_conv.conv.bias (1024,) torch.float32 decoder.pre_conv.conv.weight (1024, 512, 3) torch.float32 decoder.pre_transformer.input_proj.bias (512,) torch.float32 decoder.pre_transformer.input_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.0.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.0.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.0.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.0.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.0.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.0.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.0.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.0.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.0.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.0.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.0.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.1.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.1.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.1.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.1.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.1.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.1.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.1.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.1.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.1.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.1.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.1.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.2.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.2.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.2.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.2.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.2.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.2.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.2.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.2.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.2.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.2.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.2.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.3.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.3.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.3.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.3.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.3.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.3.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.3.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.3.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.3.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.3.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.3.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.4.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.4.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.4.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.4.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.4.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.4.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.4.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.4.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.4.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.4.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.4.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.5.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.5.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.5.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.5.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.5.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.5.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.5.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.5.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.5.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.5.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.5.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.6.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.6.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.6.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.6.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.6.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.6.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.6.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.6.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.6.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.6.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.6.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.7.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.7.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.7.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.7.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.7.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.7.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.7.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.7.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.7.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.7.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.7.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.norm.weight (512,) torch.float32 decoder.pre_transformer.output_proj.bias (1024,) torch.float32 decoder.pre_transformer.output_proj.weight (1024, 512) torch.float32 decoder.quantizer.rvq_first.input_proj.weight (256, 512, 1) torch.float32 decoder.quantizer.rvq_first.output_proj.weight (512, 256, 1) torch.float32 decoder.quantizer.rvq_first.vq.layers.0._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_first.vq.layers.0._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.input_proj.weight (256, 512, 1) torch.float32 decoder.quantizer.rvq_rest.output_proj.weight (512, 256, 1) torch.float32 decoder.quantizer.rvq_rest.vq.layers.0._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.0._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.1._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.1._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.10._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.10._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.11._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.11._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.12._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.12._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.13._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.13._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.14._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.14._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.2._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.2._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.3._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.3._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.4._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.4._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.5._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.5._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.6._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.6._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.7._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.7._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.8._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.8._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.9._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.9._codebook.embedding_sum (2048, 256) torch.float32 decoder.upsample.0.0.conv.bias (1024,) torch.float32 decoder.upsample.0.0.conv.weight (1024, 1024, 2) torch.float32 decoder.upsample.0.1.dwconv.conv.bias (1024,) torch.float32 decoder.upsample.0.1.dwconv.conv.weight (1024, 1, 7) torch.float32 decoder.upsample.0.1.gamma (1024,) torch.float32 decoder.upsample.0.1.norm.bias (1024,) torch.float32 decoder.upsample.0.1.norm.weight (1024,) torch.float32 decoder.upsample.0.1.pwconv1.bias (4096,) torch.float32 decoder.upsample.0.1.pwconv1.weight (4096, 1024) torch.float32 decoder.upsample.0.1.pwconv2.bias (1024,) torch.float32 decoder.upsample.0.1.pwconv2.weight (1024, 4096) torch.float32 decoder.upsample.1.0.conv.bias (1024,) torch.float32 decoder.upsample.1.0.conv.weight (1024, 1024, 2) torch.float32 decoder.upsample.1.1.dwconv.conv.bias (1024,) torch.float32 decoder.upsample.1.1.dwconv.conv.weight (1024, 1, 7) torch.float32 decoder.upsample.1.1.gamma (1024,) torch.float32 decoder.upsample.1.1.norm.bias (1024,) torch.float32 decoder.upsample.1.1.norm.weight (1024,) torch.float32 decoder.upsample.1.1.pwconv1.bias (4096,) torch.float32 decoder.upsample.1.1.pwconv1.weight (4096, 1024) torch.float32 decoder.upsample.1.1.pwconv2.bias (1024,) torch.float32 decoder.upsample.1.1.pwconv2.weight (1024, 4096) torch.float32 encoder.downsample.conv.weight (512, 512, 4) torch.float32 encoder.encoder.layers.0.conv.bias (64,) torch.float32 encoder.encoder.layers.0.conv.weight (64, 1, 7) torch.float32 encoder.encoder.layers.1.block.1.conv.bias (32,) torch.float32 encoder.encoder.layers.1.block.1.conv.weight (32, 64, 3) torch.float32 encoder.encoder.layers.1.block.3.conv.bias (64,) torch.float32 encoder.encoder.layers.1.block.3.conv.weight (64, 32, 1) torch.float32 encoder.encoder.layers.10.block.1.conv.bias (256,) torch.float32 encoder.encoder.layers.10.block.1.conv.weight (256, 512, 3) torch.float32 encoder.encoder.layers.10.block.3.conv.bias (512,) torch.float32 encoder.encoder.layers.10.block.3.conv.weight (512, 256, 1) torch.float32 encoder.encoder.layers.12.conv.bias (1024,) torch.float32 encoder.encoder.layers.12.conv.weight (1024, 512, 16) torch.float32 encoder.encoder.layers.14.conv.bias (512,) torch.float32 encoder.encoder.layers.14.conv.weight (512, 1024, 3) torch.float32 encoder.encoder.layers.3.conv.bias (128,) torch.float32 encoder.encoder.layers.3.conv.weight (128, 64, 8) torch.float32 encoder.encoder.layers.4.block.1.conv.bias (64,) torch.float32 encoder.encoder.layers.4.block.1.conv.weight (64, 128, 3) torch.float32 encoder.encoder.layers.4.block.3.conv.bias (128,) torch.float32 encoder.encoder.layers.4.block.3.conv.weight (128, 64, 1) torch.float32 encoder.encoder.layers.6.conv.bias (256,) torch.float32 encoder.encoder.layers.6.conv.weight (256, 128, 10) torch.float32 encoder.encoder.layers.7.block.1.conv.bias (128,) torch.float32 encoder.encoder.layers.7.block.1.conv.weight (128, 256, 3) torch.float32 encoder.encoder.layers.7.block.3.conv.bias (256,) torch.float32 encoder.encoder.layers.7.block.3.conv.weight (256, 128, 1) torch.float32 encoder.encoder.layers.9.conv.bias (512,) torch.float32 encoder.encoder.layers.9.conv.weight (512, 256, 12) torch.float32 encoder.encoder_transformer.layers.0.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.0.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.0.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.0.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.0.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.0.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.0.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.0.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.0.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.0.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.0.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.0.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.1.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.1.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.1.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.1.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.1.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.1.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.1.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.1.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.1.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.1.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.1.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.1.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.2.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.2.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.2.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.2.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.2.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.2.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.2.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.2.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.2.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.2.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.2.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.2.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.3.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.3.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.3.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.3.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.3.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.3.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.3.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.3.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.3.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.3.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.3.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.3.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.4.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.4.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.4.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.4.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.4.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.4.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.4.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.4.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.4.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.4.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.4.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.4.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.5.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.5.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.5.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.5.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.5.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.5.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.5.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.5.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.5.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.5.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.5.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.5.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.6.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.6.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.6.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.6.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.6.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.6.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.6.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.6.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.6.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.6.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.6.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.6.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.7.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.7.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.7.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.7.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.7.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.7.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.7.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.7.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.7.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.7.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.7.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.7.self_attn_layer_scale.scale (512,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.input_proj.weight (256, 512, 1) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.1.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.1.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.1.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.10.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.10.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.10.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.11.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.11.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.11.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.12.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.12.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.12.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.13.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.13.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.13.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.14.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.14.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.14.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.15.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.15.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.15.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.16.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.16.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.16.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.17.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.17.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.17.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.18.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.18.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.18.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.19.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.19.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.19.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.2.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.2.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.2.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.20.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.20.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.20.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.21.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.21.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.21.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.22.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.22.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.22.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.23.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.23.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.23.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.24.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.24.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.24.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.25.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.25.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.25.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.26.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.26.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.26.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.27.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.27.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.27.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.28.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.28.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.28.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.29.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.29.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.29.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.3.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.3.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.3.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.30.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.30.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.30.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.4.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.4.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.4.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.5.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.5.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.5.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.6.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.6.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.6.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.7.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.7.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.7.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.8.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.8.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.8.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.9.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.9.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.9.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.output_proj.weight (512, 256, 1) torch.float32 encoder.quantizer.semantic_residual_vector_quantizer.input_proj.weight (256, 512, 1) torch.float32 encoder.quantizer.semantic_residual_vector_quantizer.layers.0.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.semantic_residual_vector_quantizer.layers.0.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.semantic_residual_vector_quantizer.layers.0.codebook.initialized (1,) torch.float32 encoder.quantizer.semantic_residual_vector_quantizer.output_proj.weight (512, 256, 1) torch.float32 -------------------------------------------------------------------------------- TENSOR CATEGORIES: TTS Base Speech Tokenizer -------------------------------------------------------------------------------- decoder.decoder (118 tensors): - decoder.decoder.0.conv.bias - decoder.decoder.0.conv.weight - decoder.decoder.1.block.0.alpha - decoder.decoder.1.block.0.beta - decoder.decoder.1.block.1.conv.bias ... and 113 more decoder.pre_conv (2 tensors): - decoder.pre_conv.conv.bias - decoder.pre_conv.conv.weight decoder.pre_transformer (93 tensors): - decoder.pre_transformer.input_proj.bias - decoder.pre_transformer.input_proj.weight - decoder.pre_transformer.layers.0.input_layernorm.weight - decoder.pre_transformer.layers.0.mlp.down_proj.weight - decoder.pre_transformer.layers.0.mlp.gate_proj.weight ... and 88 more decoder.quantizer (36 tensors): - decoder.quantizer.rvq_first.input_proj.weight - decoder.quantizer.rvq_first.output_proj.weight - decoder.quantizer.rvq_first.vq.layers.0._codebook.cluster_usage - decoder.quantizer.rvq_first.vq.layers.0._codebook.embedding_sum - decoder.quantizer.rvq_rest.input_proj.weight ... and 31 more decoder.upsample (22 tensors): - decoder.upsample.0.0.conv.bias - decoder.upsample.0.0.conv.weight - decoder.upsample.0.1.dwconv.conv.bias - decoder.upsample.0.1.dwconv.conv.weight - decoder.upsample.0.1.gamma ... and 17 more encoder.downsample (1 tensors): - encoder.downsample.conv.weight encoder.encoder (28 tensors): - encoder.encoder.layers.0.conv.bias - encoder.encoder.layers.0.conv.weight - encoder.encoder.layers.1.block.1.conv.bias - encoder.encoder.layers.1.block.1.conv.weight - encoder.encoder.layers.1.block.3.conv.bias ... and 23 more encoder.encoder_transformer (96 tensors): - encoder.encoder_transformer.layers.0.input_layernorm.bias - encoder.encoder_transformer.layers.0.input_layernorm.weight - encoder.encoder_transformer.layers.0.mlp.fc1.weight - encoder.encoder_transformer.layers.0.mlp.fc2.weight - encoder.encoder_transformer.layers.0.mlp_layer_scale.scale ... and 91 more encoder.quantizer (100 tensors): - encoder.quantizer.acoustic_residual_vector_quantizer.input_proj.weight - encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.cluster_usage - encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.embed_sum - encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.initialized - encoder.quantizer.acoustic_residual_vector_quantizer.layers.1.codebook.cluster_usage ... and 95 more ################################################################################ STANDALONE TOKENIZER (Qwen3-TTS-Tokenizer-12Hz) ################################################################################ ================================================================================ CONFIG: Standalone Tokenizer ================================================================================ { "architectures": [ "Qwen3TTSTokenizerV2Model" ], "model_type": "qwen3_tts_tokenizer_12hz", "encoder_valid_num_quantizers": 16, "input_sample_rate": 24000, "output_sample_rate": 24000, "decode_upsample_rate": 1920, "encode_downsample_rate": 1920, "decoder_config": { "attention_bias": false, "attention_dropout": 0.0, "latent_dim": 1024, "codebook_dim": 512, "codebook_size": 2048, "decoder_dim": 1536, "hidden_act": "silu", "hidden_size": 512, "intermediate_size": 1024, "layer_scale_initial_scale": 0.01, "max_position_embeddings": 8000, "head_dim": 64, "num_attention_heads": 16, "num_hidden_layers": 8, "num_key_value_heads": 16, "num_quantizers": 16, "num_semantic_quantizers": 1, "rms_norm_eps": 1e-05, "rope_theta": 10000, "semantic_codebook_size": 4096, "sliding_window": 72, "upsample_rates": [ 8, 5, 4, 3 ], "upsampling_ratios": [ 2, 2 ], "vector_quantization_hidden_dimension": 512 }, "encoder_config": { "_frame_rate": 12.5, "attention_bias": false, "attention_dropout": 0.0, "audio_channels": 1, "codebook_dim": 256, "codebook_size": 2048, "compress": 2, "dilation_growth_rate": 2, "dtype": "float32", "head_dim": 64, "hidden_act": "gelu", "hidden_size": 512, "initializer_range": 0.02, "intermediate_size": 2048, "kernel_size": 7, "last_kernel_size": 3, "layer_scale_initial_scale": 0.01, "max_position_embeddings": 8000, "norm_eps": 1e-05, "normalize": false, "num_attention_heads": 8, "num_filters": 64, "num_hidden_layers": 8, "num_key_value_heads": 8, "num_quantizers": 32, "num_residual_layers": 1, "num_semantic_quantizers": 1, "pad_mode": "constant", "residual_kernel_size": 3, "rope_theta": 10000.0, "sampling_rate": 24000, "sliding_window": 250, "transformers_version": "4.57.0.dev0", "trim_right_ratio": 1.0, "upsample_groups": 512, "upsampling_ratios": [ 8, 6, 5, 4 ], "use_cache": false, "use_causal_conv": true, "use_conv_shortcut": false, "use_streaming": false, "vector_quantization_hidden_dimension": 256 }, "transformers_version": "4.57.3" } -------------------------------------------------------------------------------- ARCHITECTURE ANALYSIS: Standalone Tokenizer -------------------------------------------------------------------------------- Architecture: ['Qwen3TTSTokenizerV2Model'] Model Type: qwen3_tts_tokenizer_12hz --- Tokenizer Encoder (Audio -> Codes) --- Frame Rate: 12.5 Hidden Size: 512 Intermediate Size: 2048 Num Hidden Layers: 8 Num Attention Heads: 8 Num KV Heads: 8 Head Dim: 64 Codebook Size: 2048 Codebook Dim: 256 Num Quantizers: 32 Num Semantic Quantizers: 1 Sampling Rate: 24000 Upsampling Ratios: [8, 6, 5, 4] Num Filters: 64 Kernel Size: 7 Use Causal Conv: True --- Tokenizer Decoder (Codes -> Audio, Vocoder) --- Latent Dim: 1024 Decoder Dim: 1536 Hidden Size: 512 Intermediate Size: 1024 Num Hidden Layers: 8 Num Attention Heads: 16 Num KV Heads: 16 Head Dim: 64 Codebook Size: 2048 Codebook Dim: 512 Semantic Codebook Size: 4096 Num Quantizers: 16 Num Semantic Quantizers: 1 Upsample Rates: [8, 5, 4, 3] Upsampling Ratios: [2, 2] Sliding Window: 72 VQ Hidden Dim: 512 --- Tokenizer Global --- Valid Num Quantizers: 16 Input Sample Rate: 24000 Output Sample Rate: 24000 Encode Downsample Rate: 1920 Decode Upsample Rate: 1920 ================================================================================ TENSORS: Standalone Tokenizer ================================================================================ Total tensors: 496 Tensor Name Shape Dtype -------------------------------------------------------------------------------------------------------------- decoder.decoder.0.conv.bias (1536,) torch.float32 decoder.decoder.0.conv.weight (1536, 1024, 7) torch.float32 decoder.decoder.1.block.0.alpha (1536,) torch.float32 decoder.decoder.1.block.0.beta (1536,) torch.float32 decoder.decoder.1.block.1.conv.bias (768,) torch.float32 decoder.decoder.1.block.1.conv.weight (1536, 768, 16) torch.float32 decoder.decoder.1.block.2.act1.alpha (768,) torch.float32 decoder.decoder.1.block.2.act1.beta (768,) torch.float32 decoder.decoder.1.block.2.act2.alpha (768,) torch.float32 decoder.decoder.1.block.2.act2.beta (768,) torch.float32 decoder.decoder.1.block.2.conv1.conv.bias (768,) torch.float32 decoder.decoder.1.block.2.conv1.conv.weight (768, 768, 7) torch.float32 decoder.decoder.1.block.2.conv2.conv.bias (768,) torch.float32 decoder.decoder.1.block.2.conv2.conv.weight (768, 768, 1) torch.float32 decoder.decoder.1.block.3.act1.alpha (768,) torch.float32 decoder.decoder.1.block.3.act1.beta (768,) torch.float32 decoder.decoder.1.block.3.act2.alpha (768,) torch.float32 decoder.decoder.1.block.3.act2.beta (768,) torch.float32 decoder.decoder.1.block.3.conv1.conv.bias (768,) torch.float32 decoder.decoder.1.block.3.conv1.conv.weight (768, 768, 7) torch.float32 decoder.decoder.1.block.3.conv2.conv.bias (768,) torch.float32 decoder.decoder.1.block.3.conv2.conv.weight (768, 768, 1) torch.float32 decoder.decoder.1.block.4.act1.alpha (768,) torch.float32 decoder.decoder.1.block.4.act1.beta (768,) torch.float32 decoder.decoder.1.block.4.act2.alpha (768,) torch.float32 decoder.decoder.1.block.4.act2.beta (768,) torch.float32 decoder.decoder.1.block.4.conv1.conv.bias (768,) torch.float32 decoder.decoder.1.block.4.conv1.conv.weight (768, 768, 7) torch.float32 decoder.decoder.1.block.4.conv2.conv.bias (768,) torch.float32 decoder.decoder.1.block.4.conv2.conv.weight (768, 768, 1) torch.float32 decoder.decoder.2.block.0.alpha (768,) torch.float32 decoder.decoder.2.block.0.beta (768,) torch.float32 decoder.decoder.2.block.1.conv.bias (384,) torch.float32 decoder.decoder.2.block.1.conv.weight (768, 384, 10) torch.float32 decoder.decoder.2.block.2.act1.alpha (384,) torch.float32 decoder.decoder.2.block.2.act1.beta (384,) torch.float32 decoder.decoder.2.block.2.act2.alpha (384,) torch.float32 decoder.decoder.2.block.2.act2.beta (384,) torch.float32 decoder.decoder.2.block.2.conv1.conv.bias (384,) torch.float32 decoder.decoder.2.block.2.conv1.conv.weight (384, 384, 7) torch.float32 decoder.decoder.2.block.2.conv2.conv.bias (384,) torch.float32 decoder.decoder.2.block.2.conv2.conv.weight (384, 384, 1) torch.float32 decoder.decoder.2.block.3.act1.alpha (384,) torch.float32 decoder.decoder.2.block.3.act1.beta (384,) torch.float32 decoder.decoder.2.block.3.act2.alpha (384,) torch.float32 decoder.decoder.2.block.3.act2.beta (384,) torch.float32 decoder.decoder.2.block.3.conv1.conv.bias (384,) torch.float32 decoder.decoder.2.block.3.conv1.conv.weight (384, 384, 7) torch.float32 decoder.decoder.2.block.3.conv2.conv.bias (384,) torch.float32 decoder.decoder.2.block.3.conv2.conv.weight (384, 384, 1) torch.float32 decoder.decoder.2.block.4.act1.alpha (384,) torch.float32 decoder.decoder.2.block.4.act1.beta (384,) torch.float32 decoder.decoder.2.block.4.act2.alpha (384,) torch.float32 decoder.decoder.2.block.4.act2.beta (384,) torch.float32 decoder.decoder.2.block.4.conv1.conv.bias (384,) torch.float32 decoder.decoder.2.block.4.conv1.conv.weight (384, 384, 7) torch.float32 decoder.decoder.2.block.4.conv2.conv.bias (384,) torch.float32 decoder.decoder.2.block.4.conv2.conv.weight (384, 384, 1) torch.float32 decoder.decoder.3.block.0.alpha (384,) torch.float32 decoder.decoder.3.block.0.beta (384,) torch.float32 decoder.decoder.3.block.1.conv.bias (192,) torch.float32 decoder.decoder.3.block.1.conv.weight (384, 192, 8) torch.float32 decoder.decoder.3.block.2.act1.alpha (192,) torch.float32 decoder.decoder.3.block.2.act1.beta (192,) torch.float32 decoder.decoder.3.block.2.act2.alpha (192,) torch.float32 decoder.decoder.3.block.2.act2.beta (192,) torch.float32 decoder.decoder.3.block.2.conv1.conv.bias (192,) torch.float32 decoder.decoder.3.block.2.conv1.conv.weight (192, 192, 7) torch.float32 decoder.decoder.3.block.2.conv2.conv.bias (192,) torch.float32 decoder.decoder.3.block.2.conv2.conv.weight (192, 192, 1) torch.float32 decoder.decoder.3.block.3.act1.alpha (192,) torch.float32 decoder.decoder.3.block.3.act1.beta (192,) torch.float32 decoder.decoder.3.block.3.act2.alpha (192,) torch.float32 decoder.decoder.3.block.3.act2.beta (192,) torch.float32 decoder.decoder.3.block.3.conv1.conv.bias (192,) torch.float32 decoder.decoder.3.block.3.conv1.conv.weight (192, 192, 7) torch.float32 decoder.decoder.3.block.3.conv2.conv.bias (192,) torch.float32 decoder.decoder.3.block.3.conv2.conv.weight (192, 192, 1) torch.float32 decoder.decoder.3.block.4.act1.alpha (192,) torch.float32 decoder.decoder.3.block.4.act1.beta (192,) torch.float32 decoder.decoder.3.block.4.act2.alpha (192,) torch.float32 decoder.decoder.3.block.4.act2.beta (192,) torch.float32 decoder.decoder.3.block.4.conv1.conv.bias (192,) torch.float32 decoder.decoder.3.block.4.conv1.conv.weight (192, 192, 7) torch.float32 decoder.decoder.3.block.4.conv2.conv.bias (192,) torch.float32 decoder.decoder.3.block.4.conv2.conv.weight (192, 192, 1) torch.float32 decoder.decoder.4.block.0.alpha (192,) torch.float32 decoder.decoder.4.block.0.beta (192,) torch.float32 decoder.decoder.4.block.1.conv.bias (96,) torch.float32 decoder.decoder.4.block.1.conv.weight (192, 96, 6) torch.float32 decoder.decoder.4.block.2.act1.alpha (96,) torch.float32 decoder.decoder.4.block.2.act1.beta (96,) torch.float32 decoder.decoder.4.block.2.act2.alpha (96,) torch.float32 decoder.decoder.4.block.2.act2.beta (96,) torch.float32 decoder.decoder.4.block.2.conv1.conv.bias (96,) torch.float32 decoder.decoder.4.block.2.conv1.conv.weight (96, 96, 7) torch.float32 decoder.decoder.4.block.2.conv2.conv.bias (96,) torch.float32 decoder.decoder.4.block.2.conv2.conv.weight (96, 96, 1) torch.float32 decoder.decoder.4.block.3.act1.alpha (96,) torch.float32 decoder.decoder.4.block.3.act1.beta (96,) torch.float32 decoder.decoder.4.block.3.act2.alpha (96,) torch.float32 decoder.decoder.4.block.3.act2.beta (96,) torch.float32 decoder.decoder.4.block.3.conv1.conv.bias (96,) torch.float32 decoder.decoder.4.block.3.conv1.conv.weight (96, 96, 7) torch.float32 decoder.decoder.4.block.3.conv2.conv.bias (96,) torch.float32 decoder.decoder.4.block.3.conv2.conv.weight (96, 96, 1) torch.float32 decoder.decoder.4.block.4.act1.alpha (96,) torch.float32 decoder.decoder.4.block.4.act1.beta (96,) torch.float32 decoder.decoder.4.block.4.act2.alpha (96,) torch.float32 decoder.decoder.4.block.4.act2.beta (96,) torch.float32 decoder.decoder.4.block.4.conv1.conv.bias (96,) torch.float32 decoder.decoder.4.block.4.conv1.conv.weight (96, 96, 7) torch.float32 decoder.decoder.4.block.4.conv2.conv.bias (96,) torch.float32 decoder.decoder.4.block.4.conv2.conv.weight (96, 96, 1) torch.float32 decoder.decoder.5.alpha (96,) torch.float32 decoder.decoder.5.beta (96,) torch.float32 decoder.decoder.6.conv.bias (1,) torch.float32 decoder.decoder.6.conv.weight (1, 96, 7) torch.float32 decoder.pre_conv.conv.bias (1024,) torch.float32 decoder.pre_conv.conv.weight (1024, 512, 3) torch.float32 decoder.pre_transformer.input_proj.bias (512,) torch.float32 decoder.pre_transformer.input_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.0.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.0.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.0.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.0.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.0.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.0.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.0.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.0.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.0.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.0.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.0.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.1.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.1.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.1.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.1.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.1.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.1.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.1.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.1.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.1.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.1.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.1.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.2.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.2.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.2.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.2.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.2.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.2.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.2.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.2.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.2.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.2.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.2.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.3.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.3.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.3.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.3.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.3.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.3.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.3.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.3.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.3.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.3.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.3.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.4.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.4.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.4.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.4.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.4.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.4.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.4.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.4.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.4.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.4.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.4.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.5.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.5.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.5.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.5.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.5.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.5.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.5.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.5.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.5.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.5.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.5.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.6.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.6.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.6.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.6.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.6.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.6.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.6.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.6.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.6.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.6.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.6.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.7.input_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.7.mlp.down_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.7.mlp.gate_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.7.mlp.up_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.7.mlp_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.layers.7.post_attention_layernorm.weight (512,) torch.float32 decoder.pre_transformer.layers.7.self_attn.k_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.7.self_attn.o_proj.weight (512, 1024) torch.float32 decoder.pre_transformer.layers.7.self_attn.q_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.7.self_attn.v_proj.weight (1024, 512) torch.float32 decoder.pre_transformer.layers.7.self_attn_layer_scale.scale (512,) torch.float32 decoder.pre_transformer.norm.weight (512,) torch.float32 decoder.pre_transformer.output_proj.bias (1024,) torch.float32 decoder.pre_transformer.output_proj.weight (1024, 512) torch.float32 decoder.quantizer.rvq_first.input_proj.weight (256, 512, 1) torch.float32 decoder.quantizer.rvq_first.output_proj.weight (512, 256, 1) torch.float32 decoder.quantizer.rvq_first.vq.layers.0._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_first.vq.layers.0._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.input_proj.weight (256, 512, 1) torch.float32 decoder.quantizer.rvq_rest.output_proj.weight (512, 256, 1) torch.float32 decoder.quantizer.rvq_rest.vq.layers.0._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.0._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.1._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.1._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.10._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.10._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.11._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.11._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.12._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.12._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.13._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.13._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.14._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.14._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.2._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.2._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.3._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.3._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.4._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.4._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.5._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.5._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.6._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.6._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.7._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.7._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.8._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.8._codebook.embedding_sum (2048, 256) torch.float32 decoder.quantizer.rvq_rest.vq.layers.9._codebook.cluster_usage (2048,) torch.float32 decoder.quantizer.rvq_rest.vq.layers.9._codebook.embedding_sum (2048, 256) torch.float32 decoder.upsample.0.0.conv.bias (1024,) torch.float32 decoder.upsample.0.0.conv.weight (1024, 1024, 2) torch.float32 decoder.upsample.0.1.dwconv.conv.bias (1024,) torch.float32 decoder.upsample.0.1.dwconv.conv.weight (1024, 1, 7) torch.float32 decoder.upsample.0.1.gamma (1024,) torch.float32 decoder.upsample.0.1.norm.bias (1024,) torch.float32 decoder.upsample.0.1.norm.weight (1024,) torch.float32 decoder.upsample.0.1.pwconv1.bias (4096,) torch.float32 decoder.upsample.0.1.pwconv1.weight (4096, 1024) torch.float32 decoder.upsample.0.1.pwconv2.bias (1024,) torch.float32 decoder.upsample.0.1.pwconv2.weight (1024, 4096) torch.float32 decoder.upsample.1.0.conv.bias (1024,) torch.float32 decoder.upsample.1.0.conv.weight (1024, 1024, 2) torch.float32 decoder.upsample.1.1.dwconv.conv.bias (1024,) torch.float32 decoder.upsample.1.1.dwconv.conv.weight (1024, 1, 7) torch.float32 decoder.upsample.1.1.gamma (1024,) torch.float32 decoder.upsample.1.1.norm.bias (1024,) torch.float32 decoder.upsample.1.1.norm.weight (1024,) torch.float32 decoder.upsample.1.1.pwconv1.bias (4096,) torch.float32 decoder.upsample.1.1.pwconv1.weight (4096, 1024) torch.float32 decoder.upsample.1.1.pwconv2.bias (1024,) torch.float32 decoder.upsample.1.1.pwconv2.weight (1024, 4096) torch.float32 encoder.downsample.conv.weight (512, 512, 4) torch.float32 encoder.encoder.layers.0.conv.bias (64,) torch.float32 encoder.encoder.layers.0.conv.weight (64, 1, 7) torch.float32 encoder.encoder.layers.1.block.1.conv.bias (32,) torch.float32 encoder.encoder.layers.1.block.1.conv.weight (32, 64, 3) torch.float32 encoder.encoder.layers.1.block.3.conv.bias (64,) torch.float32 encoder.encoder.layers.1.block.3.conv.weight (64, 32, 1) torch.float32 encoder.encoder.layers.10.block.1.conv.bias (256,) torch.float32 encoder.encoder.layers.10.block.1.conv.weight (256, 512, 3) torch.float32 encoder.encoder.layers.10.block.3.conv.bias (512,) torch.float32 encoder.encoder.layers.10.block.3.conv.weight (512, 256, 1) torch.float32 encoder.encoder.layers.12.conv.bias (1024,) torch.float32 encoder.encoder.layers.12.conv.weight (1024, 512, 16) torch.float32 encoder.encoder.layers.14.conv.bias (512,) torch.float32 encoder.encoder.layers.14.conv.weight (512, 1024, 3) torch.float32 encoder.encoder.layers.3.conv.bias (128,) torch.float32 encoder.encoder.layers.3.conv.weight (128, 64, 8) torch.float32 encoder.encoder.layers.4.block.1.conv.bias (64,) torch.float32 encoder.encoder.layers.4.block.1.conv.weight (64, 128, 3) torch.float32 encoder.encoder.layers.4.block.3.conv.bias (128,) torch.float32 encoder.encoder.layers.4.block.3.conv.weight (128, 64, 1) torch.float32 encoder.encoder.layers.6.conv.bias (256,) torch.float32 encoder.encoder.layers.6.conv.weight (256, 128, 10) torch.float32 encoder.encoder.layers.7.block.1.conv.bias (128,) torch.float32 encoder.encoder.layers.7.block.1.conv.weight (128, 256, 3) torch.float32 encoder.encoder.layers.7.block.3.conv.bias (256,) torch.float32 encoder.encoder.layers.7.block.3.conv.weight (256, 128, 1) torch.float32 encoder.encoder.layers.9.conv.bias (512,) torch.float32 encoder.encoder.layers.9.conv.weight (512, 256, 12) torch.float32 encoder.encoder_transformer.layers.0.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.0.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.0.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.0.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.0.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.0.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.0.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.0.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.0.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.0.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.0.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.0.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.1.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.1.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.1.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.1.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.1.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.1.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.1.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.1.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.1.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.1.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.1.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.1.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.2.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.2.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.2.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.2.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.2.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.2.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.2.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.2.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.2.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.2.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.2.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.2.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.3.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.3.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.3.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.3.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.3.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.3.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.3.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.3.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.3.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.3.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.3.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.3.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.4.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.4.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.4.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.4.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.4.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.4.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.4.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.4.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.4.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.4.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.4.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.4.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.5.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.5.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.5.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.5.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.5.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.5.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.5.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.5.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.5.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.5.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.5.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.5.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.6.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.6.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.6.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.6.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.6.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.6.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.6.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.6.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.6.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.6.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.6.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.6.self_attn_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.7.input_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.7.input_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.7.mlp.fc1.weight (2048, 512) torch.float32 encoder.encoder_transformer.layers.7.mlp.fc2.weight (512, 2048) torch.float32 encoder.encoder_transformer.layers.7.mlp_layer_scale.scale (512,) torch.float32 encoder.encoder_transformer.layers.7.post_attention_layernorm.bias (512,) torch.float32 encoder.encoder_transformer.layers.7.post_attention_layernorm.weight (512,) torch.float32 encoder.encoder_transformer.layers.7.self_attn.k_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.7.self_attn.o_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.7.self_attn.q_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.7.self_attn.v_proj.weight (512, 512) torch.float32 encoder.encoder_transformer.layers.7.self_attn_layer_scale.scale (512,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.input_proj.weight (256, 512, 1) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.1.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.1.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.1.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.10.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.10.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.10.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.11.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.11.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.11.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.12.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.12.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.12.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.13.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.13.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.13.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.14.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.14.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.14.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.15.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.15.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.15.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.16.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.16.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.16.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.17.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.17.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.17.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.18.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.18.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.18.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.19.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.19.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.19.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.2.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.2.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.2.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.20.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.20.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.20.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.21.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.21.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.21.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.22.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.22.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.22.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.23.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.23.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.23.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.24.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.24.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.24.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.25.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.25.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.25.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.26.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.26.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.26.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.27.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.27.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.27.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.28.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.28.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.28.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.29.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.29.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.29.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.3.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.3.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.3.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.30.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.30.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.30.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.4.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.4.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.4.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.5.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.5.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.5.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.6.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.6.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.6.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.7.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.7.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.7.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.8.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.8.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.8.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.9.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.9.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.layers.9.codebook.initialized (1,) torch.float32 encoder.quantizer.acoustic_residual_vector_quantizer.output_proj.weight (512, 256, 1) torch.float32 encoder.quantizer.semantic_residual_vector_quantizer.input_proj.weight (256, 512, 1) torch.float32 encoder.quantizer.semantic_residual_vector_quantizer.layers.0.codebook.cluster_usage (2048,) torch.float32 encoder.quantizer.semantic_residual_vector_quantizer.layers.0.codebook.embed_sum (2048, 256) torch.float32 encoder.quantizer.semantic_residual_vector_quantizer.layers.0.codebook.initialized (1,) torch.float32 encoder.quantizer.semantic_residual_vector_quantizer.output_proj.weight (512, 256, 1) torch.float32 -------------------------------------------------------------------------------- TENSOR CATEGORIES: Standalone Tokenizer -------------------------------------------------------------------------------- decoder.decoder (118 tensors): - decoder.decoder.0.conv.bias - decoder.decoder.0.conv.weight - decoder.decoder.1.block.0.alpha - decoder.decoder.1.block.0.beta - decoder.decoder.1.block.1.conv.bias ... and 113 more decoder.pre_conv (2 tensors): - decoder.pre_conv.conv.bias - decoder.pre_conv.conv.weight decoder.pre_transformer (93 tensors): - decoder.pre_transformer.input_proj.bias - decoder.pre_transformer.input_proj.weight - decoder.pre_transformer.layers.0.input_layernorm.weight - decoder.pre_transformer.layers.0.mlp.down_proj.weight - decoder.pre_transformer.layers.0.mlp.gate_proj.weight ... and 88 more decoder.quantizer (36 tensors): - decoder.quantizer.rvq_first.input_proj.weight - decoder.quantizer.rvq_first.output_proj.weight - decoder.quantizer.rvq_first.vq.layers.0._codebook.cluster_usage - decoder.quantizer.rvq_first.vq.layers.0._codebook.embedding_sum - decoder.quantizer.rvq_rest.input_proj.weight ... and 31 more decoder.upsample (22 tensors): - decoder.upsample.0.0.conv.bias - decoder.upsample.0.0.conv.weight - decoder.upsample.0.1.dwconv.conv.bias - decoder.upsample.0.1.dwconv.conv.weight - decoder.upsample.0.1.gamma ... and 17 more encoder.downsample (1 tensors): - encoder.downsample.conv.weight encoder.encoder (28 tensors): - encoder.encoder.layers.0.conv.bias - encoder.encoder.layers.0.conv.weight - encoder.encoder.layers.1.block.1.conv.bias - encoder.encoder.layers.1.block.1.conv.weight - encoder.encoder.layers.1.block.3.conv.bias ... and 23 more encoder.encoder_transformer (96 tensors): - encoder.encoder_transformer.layers.0.input_layernorm.bias - encoder.encoder_transformer.layers.0.input_layernorm.weight - encoder.encoder_transformer.layers.0.mlp.fc1.weight - encoder.encoder_transformer.layers.0.mlp.fc2.weight - encoder.encoder_transformer.layers.0.mlp_layer_scale.scale ... and 91 more encoder.quantizer (100 tensors): - encoder.quantizer.acoustic_residual_vector_quantizer.input_proj.weight - encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.cluster_usage - encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.embed_sum - encoder.quantizer.acoustic_residual_vector_quantizer.layers.0.codebook.initialized - encoder.quantizer.acoustic_residual_vector_quantizer.layers.1.codebook.cluster_usage ... and 95 more ################################################################################ SUMMARY ################################################################################ TTS Base Main Model: 478 tensors TTS Base Speech Tokenizer: 496 tensors Standalone Tokenizer: 496 tensors --- Key Architecture Parameters --- TTS Talker: 28 layers, 1024 hidden, 16 heads, 16 codebooks Code Predictor: 5 layers, 1024 hidden, 16 codebooks Tokenizer Encoder: 8 layers, 512 hidden, 32 quantizers Tokenizer Decoder: 8 layers, 1536 decoder dim, 16 quantizers