ettin-encoder-32m-imdb-sentiment

jhu-clsp/ettin-encoder-32m (ModernBERT encoder, 8192 context) finetuned for IMDB sentiment (binary) on stanfordnlp/imdb.

Results (held-out test)

metric	value
accuracy	0.9263
macro-F1	0.9263
eval max_length	512

Finetuned on a single RTX 3080 (bf16). See the project for the full training pipeline.

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tok = AutoTokenizer.from_pretrained("vumichien/ettin-encoder-32m-imdb-sentiment")
model = AutoModelForSequenceClassification.from_pretrained("vumichien/ettin-encoder-32m-imdb-sentiment")

inputs = tok("your text here", truncation=True, max_length=512, return_tensors="pt")
pred = model(**inputs).logits.argmax(-1).item()
print(model.config.id2label[pred])