Basemodel: roBERTa

Configs: Vocab size: 10,000 Hidden size: 512 Max position embeddings: 512 Number of layers: 2 Number of heads: 4 Window size: 256 Intermediate-size: 1024

Results:

Task: glue Score: 57.69 Confidence Interval: [56.75, 58.73]
Task: blimp Score: 59.25 Confidence Interval: [58.78, 59.65]

Downloads last month: 7

Dataset used to train AISE-TUDelft/Custom-Activations-BERT-Adaptive-GELU

Collection including AISE-TUDelft/Custom-Activations-BERT-Adaptive-GELU

BRP Tiny-Transformers

Collection

Models for the 2024-Q4 BSc. Research Project: "Architectural Decisions for Language Modelling with Small Transformers". • 13 items • Updated Mar 2 • 1