title

Takes a passage of text and returns a short factual title and a one- to two-sentence description.

Files

File	Format	Contents
`title.pt`	PyTorch checkpoint	Merged bfloat16 weights, the full-precision source (for retraining / other runtimes)
`config.json`	JSON	Model config for `title.pt`
`generation_config.json`	JSON	Default decoding parameters
`chat_template.jinja`	Jinja	Chat template used to format input
`tokenizer.json`, `tokenizer_config.json`	JSON	Tokenizer
`title-mlx-6bit.safetensors`	MLX (6-bit)	Quantized derivative of `title.pt` (group size 64, affine)
`title-mlx-6bit.safetensors.index.json`	JSON	Weight index for the MLX build
`title-mlx-6bit.config.json`	JSON	Model config for the MLX build, including its quantization block

The MLX build is a quantized derivative kept alongside the source checkpoint. Loading it with mlx_lm requires the three title-mlx-6bit.* files renamed to model.safetensors, model.safetensors.index.json and config.json in a directory with the tokenizer files.

Core ML and LiteRT builds are not published yet.

Built on

ibm-granite/granite-4.0-350m (Apache 2.0): the pretrained backbone and its tokenizer.

See THIRD_PARTY_NOTICES.md.

License

Desert Ant Labs Source-Available License. Free for most apps; a commercial license is required at scale. Full terms are at the link. Licensing: licensing@desertant.com.

Downloads last month: -