Modelpublic

Denizhaan/imdb-roberta-lora-v2

sourceHugging Faceupdated 1mo agoView on Hugging Face

0likes6downloads

Model Card

RoBERTa-base + LoRA — IMDB Sentiment Classification

LoRA-adapted roberta-base for binary sentiment classification on the IMDB Large Movie Review Dataset. One of four base models in a parameter-efficient stacking ensemble described in "Parameter-Efficient Fine-Tuning Ensembles for Sentiment Analysis" (IDAP 2026).

Results (25,000-example IMDB test set)

Metric	Value
Accuracy	95.60%
F1	95.62%
Precision	95.23%
Recall	96.01%
ROC-AUC	99.00%

LoRA Configuration

Hyperparameter	Value
Rank (r)	16
Alpha	32
Dropout	0.1
Target modules	`query`, `value`
Trainable parameters	1,181,954 (0.94% of 125.8M total)
Precision	fp16

Loading

python

from transformers import RobertaTokenizer, RobertaForSequenceClassification
from peft import PeftModel

tokenizer = RobertaTokenizer.from_pretrained("roberta-base")
base = RobertaForSequenceClassification.from_pretrained("roberta-base", num_labels=2)
model = PeftModel.from_pretrained(base, "Denizhaan/imdb-roberta-lora-v2")
model.eval()

Citation

If you use this model, please cite our paper (details in the GitHub repository).

Framework versions

—PEFT 0.19.1