from transformers import pipeline
clf = pipeline("text-classification", model="Facececersek/bedrock-sentiment-tiny")
print(clf("this mod is sooo good"))
print(clf("it wouldn't import, it's broken"))

Training config

Parameters: ~4.4 M (bert-tiny backbone)
Training hardware: CPU only (2 vCPU)
Wall-clock training time: ~a few minutes
Val accuracy: 0.790 on a 400-example slice of the SST-2 validation set

epochs: 2
batch size: 32
max length: 64
learning rate: 5e-5
seed: 42
base model: google/bert_uncased_L-2_H-128_A-2

Notes

Trained as a fast demonstration on a CPU-only VM. For production-quality sentiment, fine-tune on the full SST-2 or a larger base model (e.g. distilbert-base-uncased).