Arm/qwen3-tts-0-6b-custom-voice-mix-precision
Full fine-tune (fewer layers) vs Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice. Arm/qwen3-tts-0-6b-custom-voice-mix-precision matches part of Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice (56 of its 402 tensors; the rest of Qwen/Qwen3-TTS-12Hz-0.6B-CustomVoice has no counterpart). On the shared part: Arm/qwen3-tts-0-6b-custom-voice-mix-precision is a GGUF (Q8_0) of different weights. Fine-tune: 100% of the weights changed, across 5 of 5 layers.
Tensors changed: Norms 21/21, MLP 15/15, Attention 20/20