MiniCheck-RoBERTa-Large (safetensors, fp16)

This is lytang/MiniCheck-RoBERTa-Large, converted to safetensors for Chirp, an on-device Mac app. Chirp uses it to check that each point in a summary is supported by the text it came from.

What changed from the original

  • The weights were converted from pytorch_model.bin to model.safetensors in float16. The values are unchanged apart from the float16 rounding.
  • config.json names its two labels: 0 is unsupported and 1 is entailment.
  • The tokenizer files from the original repo are included.

Nothing was retrained or fine-tuned.

Use

Input: a document chunk and a claim, as a RoBERTa sentence pair. The model gives a probability that the claim is supported (label entailment).

Credit and licence

The original model is by Liyan Tang, Philippe Laban and Greg Durrett, released under the MIT licence. It is described in:

Liyan Tang, Philippe Laban, Greg Durrett. MiniCheck: Efficient Fact-Checking of LLMs on Grounding Documents. EMNLP 2024. arXiv:2404.10774

This conversion is also released under the MIT licence.

Downloads last month
-
Safetensors
Model size
0.4B params
Tensor type
F16
ยท
MLX
Hardware compatibility
Log In to add your hardware

Quantized

Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for jakeflavin/minicheck-roberta-large

Finetuned
(3)
this model

Paper for jakeflavin/minicheck-roberta-large