Neural Network Lab · Digit model · arhamamin.com
Review sentiment GRU
Load the model, write a movie review, and see whether it reads as positive or negative. It was trained from scratch on IMDb, SST-2, and balanced compositional examples. It scored 85.86% on the untouched 25,000-review IMDb test set and 97.44% on 39 short behavioral checks.
Not loaded. No model bytes downloaded yet.
0 tokens. The first 256 tokens are analyzed.
Load the model, then enter a review.
How it works
The two-layer bidirectional attention GRU has 3,629,569 parameters. Vercel serves the 13.9 MiB model, vocabulary, and WebAssembly runtime as static files. Negation scope is marked during tokenization, then the neural network classifies the text on your CPU in this tab. Your review is not posted to an inference server. The percentages are model scores, not calibrated certainty.