from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

repo = "vyshnav112233/BESSTIE-RoBERTa-en-UK-seed7"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForSequenceClassification.from_pretrained(repo)
model.eval()

inputs = tok("I loved this!", return_tensors="pt", truncation=True, max_length=128)
with torch.no_grad():
    logits = model(**inputs).logits
pred = int(logits.argmax(-1))
print(model.config.id2label[pred])

Training details

Base model: roberta-base
Max sequence length: 128
Task: binary sentiment classification (num_labels=2)
Variety: en-UK (one of en-AU, en-UK, en-IN, mixed)
Random seed: 7

The raw training-state checkpoint is also included as best_model.pt for users who want to reload the state_dict directly into a RobertaForSequenceClassification instance.