model.safetensors: MiniLM/BERT encoder weights readable by MLX.
reward_head.safetensors: clamped linear reward head in MLX safetensors format.
tokenizer files copied from the source reward model.

Usage

python

from neuraltxt import NeuralTxtReward

reward = NeuralTxtReward(backend="mlx")
score = reward.score(
    response="Paris is the capital of France.",
    reference="The capital of France is Paris.",
)

To load this repo explicitly:

reward = NeuralTxtReward("paperbd/neuraltxt-reward-tiny-mlx", backend="mlx")