HaruModel( (lm): Qwen3_5TextModel( (embed_tokens): Embedding(248320, 32) (layers): ModuleList( (0-2): 3 x Qwen3_5DecoderLayer( (linear_attn): Qwen3_5GatedDeltaNet( (conv1d): Conv1d(128, 128, kernel_size=(4,), stride=(1,), padding=(3,), groups=128, bias=False) (norm): Qwen3_5RMSNormGated() (out_proj): Linear(in_features=64, out_features=32, bias=False) (in_proj_qkv): Linear(in_features=32, out_features=128, bias=False) (in_proj_z): Linear(in_features=32, out_features=64, bias=False) (in_proj_b): Linear(in_features=32, out_features=4, bias=False) (in_proj_a): Linear(in_features=32, out_features=4, bias=False) ) (mlp): Qwen3_5MLP( (gate_proj): Linear(in_features=32, out_features=64, bias=False) (up_proj): Linear(in_features=32, out_features=64, bias=False) (down_proj): Linear(in_features=64, out_features=32, bias=False) (act_fn): SiLUActivation() ) (input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06) (post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06) ) (3): Qwen3_5DecoderLayer( (self_attn): Qwen3_5Attention( (q_proj): Linear(in_features=32, out_features=128, bias=False) (k_proj): Linear(in_features=32, out_features=16, bias=False) (v_proj): Linear(in_features=32, out_features=16, bias=False) (o_proj): Linear(in_features=64, out_features=32, bias=False) (q_norm): Qwen3_5RMSNorm((16,), eps=1e-06) (k_norm): Qwen3_5RMSNorm((16,), eps=1e-06) ) (mlp): Qwen3_5MLP( (gate_proj): Linear(in_features=32, out_features=64, bias=False) (up_proj): Linear(in_features=32, out_features=64, bias=False) (down_proj): Linear(in_features=64, out_features=32, bias=False) (act_fn): SiLUActivation() ) (input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06) (post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06) ) (4-6): 3 x Qwen3_5DecoderLayer( (linear_attn): Qwen3_5GatedDeltaNet( (conv1d): Conv1d(128, 128, kernel_size=(4,), stride=(1,), padding=(3,), groups=128, bias=False) (norm): Qwen3_5RMSNormGated() (out_proj): Linear(in_features=64, out_features=32, bias=False) (in_proj_qkv): Linear(in_features=32, out_features=128, bias=False) (in_proj_z): Linear(in_features=32, out_features=64, bias=False) (in_proj_b): Linear(in_features=32, out_features=4, bias=False) (in_proj_a): Linear(in_features=32, out_features=4, bias=False) ) (mlp): Qwen3_5MLP( (gate_proj): Linear(in_features=32, out_features=64, bias=False) (up_proj): Linear(in_features=32, out_features=64, bias=False) (down_proj): Linear(in_features=64, out_features=32, bias=False) (act_fn): SiLUActivation() ) (input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06) (post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06) ) (7): Qwen3_5DecoderLayer( (self_attn): Qwen3_5Attention( (q_proj): Linear(in_features=32, out_features=128, bias=False) (k_proj): Linear(in_features=32, out_features=16, bias=False) (v_proj): Linear(in_features=32, out_features=16, bias=False) (o_proj): Linear(in_features=64, out_features=32, bias=False) (q_norm): Qwen3_5RMSNorm((16,), eps=1e-06) (k_norm): Qwen3_5RMSNorm((16,), eps=1e-06) ) (mlp): Qwen3_5MLP( (gate_proj): Linear(in_features=32, out_features=64, bias=False) (up_proj): Linear(in_features=32, out_features=64, bias=False) (down_proj): Linear(in_features=64, out_features=32, bias=False) (act_fn): SiLUActivation() ) (input_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06) (post_attention_layernorm): Qwen3_5RMSNorm((32,), eps=1e-06) ) ) (norm): Qwen3_5RMSNorm((32,), eps=1e-06) (rotary_emb): Qwen3_5TextRotaryEmbedding() ) (model_audio_proj): AudioProjector( (fc1): Linear(in_features=64, out_features=32, bias=True) (norm): RMSNorm((32,), eps=1e-06, elementwise_affine=True) (fc2): Linear(in_features=32, out_features=32, bias=True) ) (user_audio_proj): AudioProjector( (fc1): Linear(in_features=64, out_features=32, bias=True) (norm): RMSNorm((32,), eps=1e-06, elementwise_affine=True) (fc2): Linear(in_features=32, out_features=32, bias=True) ) (head): MeanFlowHead( (latent_proj): Linear(in_features=64, out_features=32, bias=False) (cond_proj): Linear(in_features=32, out_features=32, bias=False) (t_embed): ScalarEmbed( (fc1): Linear(in_features=256, out_features=32, bias=False) (fc2): Linear(in_features=32, out_features=32, bias=False) ) (dt_embed): ScalarEmbed( (fc1): Linear(in_features=256, out_features=32, bias=False) (fc2): Linear(in_features=32, out_features=32, bias=False) ) (h_embed): ScalarEmbed( (fc1): Linear(in_features=256, out_features=32, bias=False) (fc2): Linear(in_features=32, out_features=32, bias=False) ) (omega_embed): ScalarEmbed( (fc1): Linear(in_features=256, out_features=32, bias=False) (fc2): Linear(in_features=32, out_features=32, bias=False) ) (layers): ModuleList( (0): HeadLayer( (norm): Module() (ada): Linear(in_features=32, out_features=96, bias=False) (ffn): FeedForward( (gate_proj): Linear(in_features=32, out_features=96, bias=False) (up_proj): Linear(in_features=32, out_features=96, bias=False) (down_proj): Linear(in_features=96, out_features=32, bias=False) ) ) ) (final): FinalLayer( (ada): Linear(in_features=32, out_features=64, bias=False) (proj): Linear(in_features=32, out_features=64, bias=False) ) ) (codec): VibeVoiceCodec( (encoder): Encoder( (stem): CausalConv1d() (down): ModuleList( (0-5): 6 x CausalConv1d() ) (stages): ModuleList( (0): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=1, out_features=4, bias=True) (fc2): Linear(in_features=4, out_features=1, bias=True) ) ) (1): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=2, out_features=8, bias=True) (fc2): Linear(in_features=8, out_features=2, bias=True) ) ) (2): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=4, out_features=16, bias=True) (fc2): Linear(in_features=16, out_features=4, bias=True) ) ) (3): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=8, out_features=32, bias=True) (fc2): Linear(in_features=32, out_features=8, bias=True) ) ) (4): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=16, out_features=64, bias=True) (fc2): Linear(in_features=64, out_features=16, bias=True) ) ) (5): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=32, out_features=128, bias=True) (fc2): Linear(in_features=128, out_features=32, bias=True) ) ) (6): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=64, out_features=256, bias=True) (fc2): Linear(in_features=256, out_features=64, bias=True) ) ) ) (head): CausalConv1d() ) (decoder): Decoder( (stem): CausalConv1d() (up): ModuleList( (0-5): 6 x CausalConvTranspose1d() ) (stages): ModuleList( (0): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=64, out_features=256, bias=True) (fc2): Linear(in_features=256, out_features=64, bias=True) ) ) (1): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=32, out_features=128, bias=True) (fc2): Linear(in_features=128, out_features=32, bias=True) ) ) (2): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=16, out_features=64, bias=True) (fc2): Linear(in_features=64, out_features=16, bias=True) ) ) (3): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=8, out_features=32, bias=True) (fc2): Linear(in_features=32, out_features=8, bias=True) ) ) (4): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=4, out_features=16, bias=True) (fc2): Linear(in_features=16, out_features=4, bias=True) ) ) (5): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=2, out_features=8, bias=True) (fc2): Linear(in_features=8, out_features=2, bias=True) ) ) (6): ModuleList( (0): Block( (norm): RMSNorm() (conv): CausalConv1d() (ffn_norm): RMSNorm() (fc1): Linear(in_features=1, out_features=4, bias=True) (fc2): Linear(in_features=4, out_features=1, bias=True) ) ) ) (head): CausalConv1d() ) ) )