LittleDesignSolution/Voxtral-4B-TTS-2603
01
1{2 "dim": 3072,3 "n_layers": 26,4 "head_dim": 128,5 "hidden_dim": 9216,6 "n_heads": 32,7 "n_kv_heads": 8,8 "fp8_matmul": false,9 "use_biases": false,10 "causal": true,11 "rope_theta": 1000000.0,12 "norm_eps": 1e-05,13 "init": "NO_INIT",14 "dropout": 0.0,15 "vocab_size": 131072,16 "model_parallel": 1,17 "is_sequence_parallel": false,18 "context_parallel": 1,19 "tied_embeddings": true,20 "shard_on_vocab_dim": false,21 "model_pipelining": 1,22 "virtual_model_pipelining": 1,23 "fused_rms_norm": true,24 "checkpoint": false,25 "use_cache": false,26 "max_concurrent_tokens": 65536,27 "learnable_sinks": false,28 "rms_norm": "PRE",29 "cust_bwd": false,30 "recompute_w1_every": 0,31 "recompute_w3_every": 0,32 "recompute_attn_every": 0,33 "freeze_nonembedding": false,34 "fsdp2": true,35 "dp_replicate_size": 1,36 "zero2": true,37 "fsdp_optimize_backward_concat_if_pp": true,38 "attention_type": "FLASH_ATTN_3",39 "multimodal": {40 "bos_token_id": 1,41 "audio_model_args": {42 "semantic_codebook_size": 8192,43 "acoustic_codebook_size": 21,44 "n_acoustic_codebook": 36,45 "audio_encoding_args": {46 "codebook_pattern": "parallel",47 "interleave_audio_tokens_per_segment": 8192,48 "interleave_text_tokens_per_segment": 8192,49 "single_trailing_segment": false,50 "num_codebooks": 37,51 "sampling_rate": 24000,52 "frame_rate": 12.553 },54 "audio_token_id": 24,55 "begin_audio_token_id": 25,56 "input_embedding_concat_type": "sum",57 "acoustic_transformer_args": {58 "input_dim": 3072,59 "dim": 3072,60 "n_layers": 3,61 "head_dim": 128,62 "hidden_dim": 9216,63 "n_heads": 32,64 "n_kv_heads": 8,65 "use_biases": false,66 "rope_theta": 10000.0,67 "sigma": 1e-05,68 "sigma_max": 1.069 },70 "p_uncond": 0.0,71 "text_feature_bugged": false,72 "condition_dropped_token_id": 4273 },74 "audio_tokenizer_args": {75 "channels": 1,76 "sampling_rate": 24000,77 "pretransform_patch_size": 240,78 "patch_proj_kernel_size": 7,79 "semantic_codebook_size": 8192,80 "semantic_dim": 256,81 "acoustic_codebook_size": 21,82 "acoustic_dim": 36,83 "conv_weight_norm": true,84 "causal": true,85 "attn_sliding_window_size": 16,86 "half_attn_window_upon_downsampling": true,87 "dim": 1024,88 "hidden_dim": 4096,89 "head_dim": 128,90 "n_heads": 8,91 "n_kv_heads": 8,92 "qk_norm_eps": 1e-06,93 "qk_norm": true,94 "use_biases": false,95 "norm_eps": 0.01,96 "layer_scale": true,97 "layer_scale_init": 0.01,98 "decoder_transformer_lengths_str": "2,2,2,2",99 "decoder_convs_kernels_str": "3,4,4,4",100 "decoder_convs_strides_str": "1,2,2,2",101 "voice": {102 "casual_female": 0,103 "casual_male": 1,104 "cheerful_female": 2,105 "neutral_female": 3,106 "neutral_male": 4,107 "pt_male": 5,108 "pt_female": 6,109 "nl_male": 7,110 "nl_female": 8,111 "it_male": 9,112 "it_female": 10,113 "fr_male": 11,114 "fr_female": 12,115 "es_male": 13,116 "es_female": 14,117 "de_male": 15,118 "de_female": 16,119 "ar_male": 17,120 "hi_male": 18,121 "hi_female": 19122 }123 }124 },125 "torch_compile_swiglu_noncust_bwd": false,126 "override_parameters_str": "",127 "max_seq_len": 65536,128 "model_type": "voxtral_tts",129 "max_position_embeddings": 128000130}