Repository navigation
Releases
Releases · darylalim/siebert-mlx
v0.7.1
Full Changelog: v0.7.0...v0.7.1
v0.7.0
A Streamlit app for English-text sentiment classification with SiEBERT, running locally on Apple Silicon via MLX.
Highlights
- Batched MLX inference in float16 on Apple Silicon — no PyTorch/CUDA, no device management
- Upload a CSV or load built-in samples; auto-detects the text column with manual override
- Binary sentiment (positive/negative) with confidence scores
- Dashboard UI: metric cards, sentiment-distribution chart, styled results table, CSV download
- Results persist across reruns; one-click Reset; light/dark themes
- Robust handling of empty, whitespace-only, and malformed input
Requirements: Apple Silicon Mac · Python ≥ 3.12 · a Hugging Face token (HF_TOKEN) for model download
License: MIT (app code). The SiEBERT model is downloaded at runtime under its own terms.