sucrette/go-emotions-coreai

roberta-base-go_emotions for Core AI

SamLowe/roberta-base-go_emotions (revision d7504834, MIT, by Sam Lowe) converted to Apple's Core AI (.aimodel, macOS 27). Classifies English text into the 28 GoEmotions labels; Asa uses it to pick a reaction for each sentence it says (AsaFilter.EmotionClassifier).

Files

• go_emotions.aimodel: fp16, dynamic batch 1-64 and sequence 2-128
• tokenizer/: the RoBERTa BPE tokenizer and config.json (its id2label names the 28 outputs, pad_token_id the padding)

Inputs and outputs

• input_ids int32 [B, S], attention_mask int32 [B, S] (B 1-64, S 2-128)
• out logits float [B, 28]. Multi-label: apply a sigmoid, not a softmax.

Checked on

Apple silicon Mac, macOS 27, coreai-core 1.0.0b3, against the fp32 PyTorch model on 20 mixed sentences padded to 64 tokens: max logit difference 0.087 on the GPU and 0.050 on the CPU; every label at p = 0.5 the same (560 of 560 on each).

How it was made

uv run tools/export_go_emotions.py OUT_DIR

from Asa's tools/. Pins: coreai-core 1.0.0b3, coreai-torch 0.4.3, transformers 4.57.3.

License

MIT, as the original model (LICENSE, Copyright (c) 2023 SamLowe). Dataset: GoEmotions (Demszky et al., 2020).