GSV.CPP — GPT-SoVITS V5 weights for gsv.cpp

Converted model files for gsv.cpp, a C++/ggml inference engine for GPT-SoVITS V5 only — v5turbo (4-step DMD) and v5dev (32-step). v2 / v3 / v4 checkpoints are not supported.

The engine runs the whole pipeline in C++ (AR → cond → DiT/CFM → vocoder) with a ggml Vulkan or CPU backend, and an ONNX Runtime WebGPU/CPU vocoder. A small HTTP service exposes the same API as the upstream api_v2.py, so the GAG GUI can use it directly.

Files

file	role	notes
gsv-ar-f16.gguf	AR (s1v3)	fp16
gsv-ar-q8.gguf	AR (s1v3)	Q8_0 attention/FFN, ~2× smaller
gsv-cond-f16.gguf	cond (enc_p / ref_enc / wns1 / rvq)	v5turbo
gsv-dit-f16all.gguf	DiT / CFM	v5turbo
gsv-refcode.gguf	reference encoder projection	v5turbo
gsv-v5dev-cond-f16.gguf	cond	v5dev
gsv-v5dev-dit-f16all.gguf	DiT / CFM	v5dev
gsv-v5dev-refcode.gguf	reference encoder projection	v5dev
gsv-bert-f16.gguf	BERT (Chinese RoBERTa)	shared
gsv-hubert-f16.gguf	CN-HuBERT	shared
vocoder_fp32.onnx	HiFi-GAN vocoder, 48 kHz, fp32, dynamic T	shared
resampy_kaiser_best.bin	resampling filter table	shared

AR is provided in two precisions; gsv-ar-q8.gguf is the default recommendation (audio is perceptually indistinguishable from fp16).

Layout

Put the files into the engine's model directory (v5turbo setup):

models/
  gsv-ar-q8.gguf
  gsv-cond-f16.gguf
  gsv-dit-f16all.gguf
  gsv-refcode.gguf
  gsv-bert-f16.gguf
  gsv-hubert-f16.gguf
  vocoder_fp32.onnx
  resampy_kaiser_best.bin

Additional voices (v5dev or your own fine-tunes) are described by a manifest, e.g. models/voices/v5dev/manifest.json:

{
  "name": "v5dev",
  "version": "v5dev",
  "ar": "gsv-ar-q8.gguf",
  "cond": "gsv-v5dev-cond-f16.gguf",
  "dit": "gsv-v5dev-dit-f16all.gguf",
  "refcode": "gsv-v5dev-refcode.gguf",
  "bert": "gsv-bert-f16.gguf",
  "hubert": "gsv-hubert-f16.gguf",
  "vocoder": "vocoder_fp32.onnx"
}

Usage

See the engine repository for build instructions, the HTTP API and how to convert your own fine-tuned checkpoints:

• https://github.com/KakaruHayate/gsv.cpp
• docs/usage.md — artifacts, models, conversion, running the service
• docs/benchmark.md — full-pipeline timings

# text frontend
python tools/text_sidecar.py --port 9882
# engine
./gsv_server --device vulkan --port 9880

License

The engine code and these conversions are MIT licensed (see LICENSE). The weights are derived from GPT-SoVITS checkpoints and remain under the upstream MIT license; see licenses/GPT-SoVITS-LICENSE.txt and THIRD_PARTY_NOTICES.md.