Base model: distilbert/distilbert-base-uncased (~67M, generic)
Dataset: cardiffnlp/tweet_eval (sentiment config, 45.6K train)
Training: 3 epochs, lr 2e-5, batch 32, max_len 128, warmup 0.1, weight decay 0.01

Test-set results (TweetEval sentiment, 12,284 tweets)

Metric	Score
Accuracy	0.6888
Macro-F1	0.6877
Macro-Recall	0.6978
Speed (T4)	~2897 tweets/s

Comparison vs twitter-roberta-base

Model	Size	Accuracy	Macro-F1	tweets/s
twitter-roberta-base	125M	0.7155	0.7155	1600
distilbert-base (this)	67M	0.6888	0.6877	2897

This model trades ~2.7 pts accuracy for ~1.8× faster inference and half the size — a good fit for high-throughput or edge deployment.

Usage

python

from transformers import pipeline
clf = pipeline("text-classification", model="Ido-shraga/distilbert-base-tweeteval-sentiment")
clf("I can't believe how good this is 🔥")