audio.cpp package (qwen3_tts) converted from Qwen/Qwen3-TTS-12Hz-0.6B-Base with audiocpp_gguf.
Quantized files keep the speaker encoder in BF16. Embeddings and codebooks stay F16. Other tensors that cannot be quantized stay at the source dtype.
| File | Type |
|---|---|
qwen3-tts-12hz-0.6b-base-q2_k.gguf |
Q2_K |
qwen3-tts-12hz-0.6b-base-q3_k.gguf |
Q3_K |
qwen3-tts-12hz-0.6b-base-q4_0.gguf |
Q4_0 |
qwen3-tts-12hz-0.6b-base-q4_1.gguf |
Q4_1 |
qwen3-tts-12hz-0.6b-base-q4_k.gguf |
Q4_K |
qwen3-tts-12hz-0.6b-base-q5_0.gguf |
Q5_0 |
qwen3-tts-12hz-0.6b-base-q5_1.gguf |
Q5_1 |
qwen3-tts-12hz-0.6b-base-q5_k.gguf |
Q5_K |
qwen3-tts-12hz-0.6b-base-q6_k.gguf |
Q6_K |
qwen3-tts-12hz-0.6b-base-q8_0.gguf |
Q8_0 |
qwen3-tts-12hz-0.6b-base-f16.gguf |
F16 |
qwen3-tts-12hz-0.6b-base-bf16.gguf |
BF16 |
qwen3-tts-12hz-0.6b-base-f32.gguf |
F32 |
audio.cpp has tested this family on BF16 and Q8_0. The lower-bit files are convertible, and the runtime has not been tested on them.
