cstr/darwin-tts-1.7b-cross-GGUF

Darwin-TTS-1.7B-Cross — GGUF

GGUF conversion of FINAL-Bench/Darwin-TTS-1.7B-Cross for CrispASR.

Training-free TTS improvement: 3% of a Qwen3-1.7B general LLM's FFN weights are linearly interpolated into the Qwen3-TTS-1.7B backbone, adding emotional expressiveness without any fine-tuning. Runs on the qwen3-tts backend.

Architecture

• —Base: Qwen3-TTS-1.7B pipeline (talker + code predictor + speech tokenizer + codec)
• —Talker (modified): 28-layer Qwen3 LM (2048-dim, 6144 FFN, 16Q/8KV heads) — 84 FFN tensors blended at alpha=3% with Qwen3-1.7B LLM weights
• —Code predictor: 5-layer auxiliary net (1024-dim)
• —Speech tokenizer: 12 Hz RVQ codec
• —Languages: Korean, English, Japanese, Chinese + 6 more

Files

Usage

crispasr --backend qwen3-tts \
    -m darwin-tts-1.7b-cross-q4_k.gguf \
    --tts "Hello, world." --tts-output out.wav

License

Apache 2.0 (inherited from Qwen3-TTS and Qwen3-1.7B).

Credits

• —Original model: FINAL-Bench
• —GGUF conversion & runtime: CrispASR

Provenance and EU AI Act Art. 53 note

• —Upstream model: FINAL-Bench/Darwin-TTS-1.7B-Cross — published by FINAL-Bench.
• —Upstream licence: apache-2.0. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• —What was done here: format conversion and/or quantisation only (GGUF). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• —Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository. No training-content summary was found on the upstream model card at the time of writing; that documentation gap is upstream's and is not filled here.
• —Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.