Update README.md

README.md

@@ -53,7 +53,7 @@ This variant is trained synthetic data and on the RAGTruth dataset for hallucina
 ## Training Procedure
 
 - Tokenizer: AutoTokenizer; DataCollatorForTokenClassification; label pad −100
-- Max length: 8k; batch size: 8; epochs: 3–6
+- Max length: 8k; batch size: 16; epochs: 5
 - Optimizer: AdamW (lr 1e‑5, weight_decay 0.01)
 - Hardware: Single A100 80GB