Conditional Pseudo-Arithmetic on Fine-tuned GPT-2 (test)
80.08ICL ScoreCoQE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CoQEToken Embedding=Llama-3.2-1B2026.03 | 80.08 | 97.34 | |
| CoQEToken Embedding=Llama-3.2-3B2026.03 | 75.01 | 96.57 | |
| CoQEToken Embedding=Llama-3.1-8B2026.03 | 65.35 | 93.96 | |
| TransformerToken Embedding=Llama-3.1-8B2026.03 | 50.12 | 96.65 | |
| TransformerToken Embedding=Llama-3.2-1B2026.03 | 49.95 | 99.12 | |
| TransformerToken Embedding=Llama-3.2-3B2026.03 | 49.33 | 98.81 |