Text-to-Speech
MLX
Safetensors
English
dots_tts
tts
quantized
4-bit precision
8-bit precision
apple-silicon
dots.tts
Instructions to use smcleod/dots.tts-soar-mlx with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use smcleod/dots.tts-soar-mlx with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir dots.tts-soar-mlx smcleod/dots.tts-soar-mlx
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
| #!/usr/bin/env python | |
| """Extract the small projection heads the pipeline/solver need from model.safetensors. | |
| All are Linear/LayerNorm (weight (out,in)) - no transpose. CPU/numpy, sandbox-safe.""" | |
| from safetensors import safe_open | |
| from safetensors.numpy import save_file | |
| SNAP = ("/Users/samm/.cache/huggingface/hub/models--rednote-hilab--dots.tts-soar/" | |
| "snapshots/1fd9452e55c2c9f38fe1a8ee09eaf7448c222d35/model.safetensors") | |
| OUT = "/Users/samm/git/sammcj/dots.tts-soar-mlx/heads/model.safetensors" | |
| PREFIXES = ("coordinate_proj.", "hidden_proj.", "latent_proj.", "xvec_proj.", "eos_proj.") | |
| out = {} | |
| with safe_open(SNAP, framework="numpy") as f: | |
| for k in f.keys(): | |
| if k.startswith(PREFIXES): | |
| out[k] = f.get_tensor(k) | |
| import os | |
| os.makedirs(os.path.dirname(OUT), exist_ok=True) | |
| save_file(out, OUT) | |
| for k, v in sorted(out.items()): | |
| print(k, v.shape, v.dtype) | |
| print(f"wrote {OUT} ({len(out)} tensors)") | |