Ahmadzei's picture
added 3 more tables for large emb model
5fa1a76
Try using gradient_accumulation_steps in [TrainingArguments] to effectively increase overall batch size.