Text Generation 벤치마크
Text Generation on OpenWebText
eval_loss 낮을수록 좋음
- 2019-02-14 — GPT2-124M: eval_loss 3.12
- 2024-09-21 — GPT2-81M-LOOP: eval_loss 3.11
- 2025-02-03 — GPT2-Hermite: eval_loss 2.91
| Rank | Model | eval_loss | Paper | Code | Year |
|---|---|---|---|---|---|
| 1 | GPT2-Hermite | 2.91 | Polynomial, trigonometric, and tropical activations | K-H-Ismail/torchortho | 2025 |
| 2 | GPT2-81M-LOOP | 3.11 | Loop Neural Networks for Parameter Sharing | 2024 | |
| 3 | GPT2-124M | 3.12 | Language Models are Unsupervised Multitask Learners | huggingface/transformers · openai/gpt-2 · PaddlePaddle/PaddleNLP · +18 | 2019 |