Linguistic Acceptability on CoLA
83.43AccuracyFinetune
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Finetune2022.12 | 83.43 | — | — | |
| ColD-Fusion2022.12 | 82.74 | — | — | |
| Multitask2022.12 | 82.42 | — | — | |
| MUPPET2022.12 | 81.21 | — | — | |
| Latent Concept LearningBackbone=GPT2-large, Configuration=concept tokens as prefixes2023.01 | 75 | — | — | |
| Tay et al. (2022)Mode=Fully-supervised, Model Parameters=4x of UD (~44B)2022.11 | 71.5 | — | — | |
| UD+-XXLMode=Fully-supervised, Backbone=T5-XXL encoder, Model Parameters=11B2022.11 | 71.42 | — | — | |
| Latent Concept LearningLLM=GPT3-b (1.3B)2023.01 | 70.2 | — | — | |
| Latent Concept LearningLLM=GPT2 (124M)2023.01 | 69.1 | — | — | |
| Latent Concept LearningLLM=GPT2-l (774M)2023.01 | 69.1 | — | — | |
| Latent Concept LearningLLM=GPT2-m (355M)2023.01 | 69 | — | — | |
| Latent Concept LearningLLM=GPT3-c (6.7B)2023.01 | 69 | — | — | |
| Latent Concept LearningLLM=GPT2-xl (1.5B)2023.01 | 68.9 | — | — | |
| SimilarLLM=GPT2 (124M)2023.01 | 63.2 | — | — | |
| UniformLLM=GPT2-l (774M)2023.01 | 62.7 | — | — | |
| UniformLLM=GPT2 (124M)2023.01 | 61.9 | — | — | |
| UniformLLM=GPT3-c (6.7B)2023.01 | 59.1 | — | — | |
| Latent Concept LearningLLM=GPT3-d (175B)2023.01 | 58.5 | — | — | |
| Latent Concept LearningLLM=GPT3-a (350M)2023.01 | 58.2 | — | — | |
| SimilarLLM=GPT3-c (6.7B)2023.01 | 58 | — | — | |
| UniformLLM=GPT2-m (355M)2023.01 | 57.8 | — | — | |
| SimilarLLM=GPT2-m (355M)2023.01 | 57.5 | — | — | |
| UniformLLM=GPT2-xl (1.5B)2023.01 | 55.8 | — | — | |
| SimilarLLM=GPT2-xl (1.5B)2023.01 | 55 | — | — | |
| UniformLLM=GPT3-a (350M)2023.01 | 53.1 | — | — | |
| SimilarLLM=GPT3-a (350M)2023.01 | 53.1 | — | — | |
| SimilarLLM=GPT2-l (774M)2023.01 | 50.9 | — | — | |
| UniformLLM=GPT3-b (1.3B)2023.01 | 46.8 | — | — | |
| UniformLLM=GPT3-d (175B)2023.01 | 45.5 | — | — | |
| SimilarLLM=GPT3-d (175B)2023.01 | 45.4 | — | — | |
| SimilarLLM=GPT3-b (1.3B)2023.01 | 45.1 | — | — | |
| Integrated Gradients2026.01 | — | 0.54 | 0.402 | |
| LIMEMode=Standard2026.01 | — | 0.255 | 0.21 | |
| LIME-LLMBackbone=Claude Sonnet 4.52026.01 | — | 0.598 | 0.457 | |
| LLimeMode=Generative2026.01 | — | 0.391 | 0.261 | |
| Partition SHAP2026.01 | — | 0.544 | 0.393 |