model_family: "index_tts"
adapter: "index_tts"
split_contract_version: 2
graph_contract:
  cache_mode: "prefill_decode"
  cache_layout: "hf_bhsd"
  layers: 24
  dtype: "float32"
  batch: 1
  heads: 20
  cache_sequence_axis: 2
  head_dim: 64
  initial_cache_length: 0
  attention_mask:
    rank: 2
    dtype: "int64"
  logits:
    name: "raw_logits"
    rank: 3
    shape:
      - 1
      - 1
      - 8194
    selection: "runtime"
generation_policy:
  version: 1
  algorithm: "seeded_top_k_top_p"
  num_beams: 1
  source_num_beams: 3
  top_k: 30
  top_p: 0.8
  temperature: 1.0
  repetition_penalty: 10.0
  repetition_scope: "mel_bos_and_full_generated_history"
  default_seed: 0
  max_new_mel_tokens: 600
  start_token: 8192
  stop_token: 8193
  silence_token: 52
sample_rate: 24000
start_token: 8192
stop_token: 8193
max_generate_length: 600
mel_code_size: 8194
vocab_size: 8194
text_vocab_size: 12000
precision: "cpu-fp32"
precision_policy: "fp32-only; cpu-q4 and gpu-fp16 paths are deprecated and ignored"
device: "cpu"
source_model_dir: "D:\\AI_WorkSpace\\lcoal-multimodal-infra\\workdir\\data\\indextts-1.5-source-25851a6\\model"
index_tts_project: "D:\\AI_WorkSpace\\lcoal-multimodal-infra\\workdir\\data\\indextts-1.5-source-25851a6\\code"
source_config: "D:\\AI_WorkSpace\\lcoal-multimodal-infra\\workdir\\data\\indextts-1.5-source-25851a6\\model\\config.yaml"
files:
  - "IndexTTS_A.onnx"
  - "IndexTTS_B.onnx"
  - "IndexTTS_C.onnx"
  - "IndexTTS_D.onnx"
  - "IndexTTS_E.onnx"
  - "IndexTTS_F.onnx"
  - "IndexTTS_E_Prefill.onnx"
  - "bpe.model"
  - "manifest.yaml"
  - "manifest.json"
optional_files:
  []
artifacts:
  -
    path: "IndexTTS_A.onnx"
    size_bytes: 279165593
    sha256: "4a3e2a1019d5045e46c953875bffd5fc88dcc6e0158b149b13aee69f451d1cb3"
  -
    path: "IndexTTS_B.onnx"
    size_bytes: 64531987
    sha256: "220b941f138d67af1aff81e48ed17ab1aef36396b03a8cb49c08d0c560cdcff1"
  -
    path: "IndexTTS_C.onnx"
    size_bytes: 46065900
    sha256: "8b16cfb617c223f87db333b1f487ccabb0ca0a739219eba101f91ea35bb42b20"
  -
    path: "IndexTTS_D.onnx"
    size_bytes: 1158
    sha256: "ed3f221cc853905cc54b85de0383e156ae5faf023fc56fd45fd29303f700b164"
  -
    path: "IndexTTS_E.onnx"
    size_bytes: 1933775727
    sha256: "cacf4e04403105fc2203e3945ad271f4eeb706b83aa649d29e6134fae351806c"
  -
    path: "IndexTTS_F.onnx"
    size_bytes: 501631823
    sha256: "40d0426764c272f4a870585fc0e4fa0f59548a494825f6a7aceb9b340d327aa4"
  -
    path: "IndexTTS_E_Prefill.onnx"
    size_bytes: 1933767798
    sha256: "b048f3b9ad7219e474c72abf741f65515b2276553f26fda8b5f55f68c848d191"
  -
    path: "bpe.model"
    size_bytes: 475997
    sha256: "b2a5ce8090d32da3642cc4f81fdc996376bc6dd3f4cd5e3d165f71120d9f2bc8"
export_provenance:
  script: "scripts/local/indextts_export.py"
  entrypoint: "scripts/indextts_export.py"
  mode: "raw-export"
  opset: 17
  audio_length: 24000
  max_seq_len: 16
  created_unix: 1783867654
  python: "3.11.15"
  platform: "Windows-10-10.0.26200-SP0"
  source:
    path: "D:\\AI_WorkSpace\\lcoal-multimodal-infra\\workdir\\data\\indextts-1.5-source-25851a6\\provenance.json"
    sha256: "3bfb39cc326d834be4fda72000e4cc53ebbb6c52e154150f1fdbe7323d1e909c"
    model_repository: "IndexTeam/IndexTTS-1.5"
    model_revision: "25851a6036dfd3095bb70fb3c8f49217104672c3"
    code_repository: "https://github.com/index-tts/index-tts.git"
    code_tag: "v1.5.0"
    code_commit: "9098497272d5803bae46cbaf5154cf2ba48f6866"
    code_tree: "aa0335ccaba54ac42d6d209dac56bb9a8b2e80a7"
    model_license:
      hugging_face_metadata: "apache-2.0"
      license_file_in_model_snapshot: null
      note: "The model snapshot contains no standalone license file. The pinned official code tree contains INDEX_MODEL_LICENSE, a custom bilibili Index-TTS model license with non-commercial and other restrictions; preserve and review it rather than relying only on Hub metadata."
      companion_license_path: "code/INDEX_MODEL_LICENSE"
      companion_license_sha256: "550cf1882b0186cffe689ee6ca22de7ee16a8ce064fae9dce9fbd92be74e7ca9"
status: "ready"
notes:
  - "A-F graph contract: A(reference audio conds), B(text ids), C(GPT token embedding), D(concat embeddings), E autoregressive KV loop, F BigVGAN vocoder."
  - "loaded official IndexTTS from D:\\AI_WorkSpace\\lcoal-multimodal-infra\\workdir\\data\\indextts-1.5-source-25851a6\\code with cfg_path=D:\\AI_WorkSpace\\lcoal-multimodal-infra\\workdir\\data\\indextts-1.5-source-25851a6\\model\\config.yaml, device=cpu, fp16=False, use_cuda_kernel=False where supported"
  - "patched 109 BigVGAN alias-free Activation1d resampler pairs in memory for ONNX export: pre-expanded grouped convolution filters with static channel count and replicate padding; official source files were not modified"
  - "IndexTTS_A uses a project-local ONNX-safe conv1d mel frontend mirroring official MelSpectrogramFeatures(sample_rate=24000,n_fft=1024,hop=256,win=1024,n_mels=100,power=1,center=True)."
  - "exporting IndexTTS_A.onnx with opset=17, external_data_threshold_mb=512 (PyTorch handles threshold support by version)"
  - "exporting IndexTTS_B.onnx with opset=17, external_data_threshold_mb=512 (PyTorch handles threshold support by version)"
  - "exporting IndexTTS_C.onnx with opset=17, external_data_threshold_mb=512 (PyTorch handles threshold support by version)"
  - "exporting IndexTTS_D.onnx with opset=17, external_data_threshold_mb=512 (PyTorch handles threshold support by version)"
  - "exporting IndexTTS_E_Prefill.onnx with opset=17, external_data_threshold_mb=512 (PyTorch handles threshold support by version)"
  - "exporting IndexTTS_E.onnx with opset=17, external_data_threshold_mb=512 (PyTorch handles threshold support by version)"
  - "exporting IndexTTS_F.onnx with opset=17, external_data_threshold_mb=512 (PyTorch handles threshold support by version)"
  - "raw PyTorch export produced IndexTTS_A.onnx ... IndexTTS_F.onnx"
  - "cpu-fp32 selected; no quantization or fp16 transform applied"
  - "copied tokenizer D:\\AI_WorkSpace\\lcoal-multimodal-infra\\workdir\\data\\indextts-1.5-source-25851a6\\model\\bpe.model"