mirror of
https://github.com/hanzoai/tts.git
synced 2026-10-30 18:18:56 +00:00
kadirnar
32a8596aeb
- Add base class for all inference engines (shared config, SNAC decode, token processing) - Replace HQQ backend with SGLang backend - Fix vLLM to use prompt_token_ids (avoids special token loss in decode/re-encode) - Enable CUDA graphs by default in vLLM (3.4x throughput improvement) - Set default config to qwen3.yaml - Add codebook clamp in SNAC decode for robustness - Switch SNAC dependency to fast-snac - Add pyproject.toml with optional deps: pip install vyvotts[vllm] / vyvotts[sglang] - Clean up README
29 lines
679 B
TOML
29 lines
679 B
TOML
[build-system]
|
|
requires = ["setuptools>=64"]
|
|
build-backend = "setuptools.build_meta"
|
|
|
|
[project]
|
|
name = "vyvotts"
|
|
version = "0.1.0"
|
|
description = "LLM-Based Text-to-Speech Training and Inference Framework"
|
|
readme = "README.md"
|
|
license = {text = "MIT"}
|
|
requires-python = ">=3.10"
|
|
dependencies = [
|
|
"transformers",
|
|
"datasets>=3.4",
|
|
"accelerate",
|
|
"snac @ git+https://github.com/kadirnar/fast-snac.git",
|
|
"soundfile",
|
|
"librosa",
|
|
"torchaudio",
|
|
]
|
|
|
|
[project.optional-dependencies]
|
|
vllm = ["vllm>=0.18"]
|
|
sglang = ["sglang[all]", "sgl-kernel"]
|
|
train = ["unsloth", "wandb", "trl", "kernels", "liger_kernel"]
|
|
|
|
[tool.setuptools.packages.find]
|
|
include = ["vyvotts*"]
|