affectively-ai/twitter-roberta-base-sentiment-onnx

Twitter RoBERTa Sentiment (ONNX, int8 Quantized)

Production-ready ONNX conversion of cardiffnlp/twitter-roberta-base-sentiment-latest for in-browser sentiment analysis — zero server cost, zero latency, complete privacy.

Highlights

• 3-class sentiment — negative, neutral, positive
• ~124 MB quantized — trained specifically on social media text
• transformers.js compatible — drop-in pipeline('text-classification')
• TweetEval benchmark leader — RoBERTa fine-tuned on ~124M tweets

Quick Start

import { pipeline } from '@huggingface/transformers';

const classifier = await pipeline(
  'text-classification',
  'affectively-ai/twitter-roberta-base-sentiment-onnx',
  { dtype: 'q8' }
);

const result = await classifier('This new feature is absolutely amazing!');
// [{ label: 'positive', score: 0.96 }]

Labels

Label	Description
negative	Negative sentiment (criticism, complaints, frustration)
neutral	Neutral or factual statements
positive	Positive sentiment (praise, excitement, satisfaction)

Conversion Details

Property	Value
Base model	cardiffnlp/twitter-roberta-base-sentiment-latest
Pre-training	~124M tweets
Export	PyTorch → ONNX via Optimum
Quantization	int8 dynamic (ORTQuantizer, avx512_vnni)
Quantized size	~124 MB

Use Cases

This model powers sentiment analysis in Edgework.ai — bringing fast, cheap, and private inference as close to the user as possible. Best for:

• Social media monitoring and brand sentiment
• Customer feedback triage (positive/negative/neutral)
• Community health dashboards
• Real-time mood tracking from short-form text

About

Published by AFFECTIVELY · Managed by @buley

We convert, quantize, and publish production-ready ONNX models for edge and in-browser inference. Every release is tested for correctness and stability before publication.

• All models · GitHub · Edgework.ai