Chatterbox ONNX
KitsuMate layout of the English ResembleAI Chatterbox model using the pinned ONNX Community export at 3cab09af388d3f02bba43443fce88c1f4525ac43.
The repository includes the upstream FP32, FP16, Q4, and Q4F16 language-model
variants. The conditional decoder, token embedding, and speech encoder published
without a suffix are shared by the upstream precision variants. The additional
q4f16-suffixed copies provide a uniformly named four-graph profile for the
KitsuMate model catalog.
Every imported ONNX file is byte-for-byte unchanged from ONNX Community
revision 3cab09af388d3f02bba43443fce88c1f4525ac43. Each graph must remain beside
its matching .onnx_data file. The English tokenizer and default voice are
also redistributed unchanged under the upstream MIT license. This is the
English checkpoint; it intentionally excludes the unrelated Cangjie mapping.