Moodify 2.0 · real models, real APIs
Fine-tuned BERT vs zero-shot JEV
The same 5-class sentiment task Moodify solves (Very Negative → Very Positive), run on nlptown/bert-base-multilingual-uncased-sentiment and on JEV with the five classes described in plain English. Compare latency, accuracy, calibration and cost, then try what BERT can't do.
Status:checking…
BERT · Moodify model
Waiting for a run.
JEV ·
Waiting for a run.
Reading the numbers fairly: BERT runs on your machine (no network hop), while JEV is a remote API. Locally, a 110M-parameter BERT will often win on raw latency. JEV's case against BERT is no training, new labels in plain English, world knowledge (sarcasm, Hinglish) and calibrated confidence. Its case against LLMs is speed and cost. The Benchmark tab measures these properly.