GPT2
Benchmarks
TaskDatasetSOTA resultTrendResults
GPT2-1.5B
4.1Throughput
25
GPT2-small
0.394LDS Score
10
GPT2 MLP SAE
40.9Faithfulness Score
8
GPT2 MLP SAE
51.2Faithfulness Score
8
GPT2 Res. SAE
47.2Faithfulness Score
8
GPT2 Res. SAE
60.4Faithfulness Score
8
GPT2-base (124M)
100Usage Fraction
7
GPT2-base (124M)
5.17Embed Inference Time (s)
7
GPT2 Layer 0 match with Layer 11
1.39LLM Eval Score
6
GPT2 Layer 5 match with Layer 11
1.56LLM Eval
6
GPT2 F.t.
74.25ASR (%)
6
GPT2-small Digit Addition
68.12Accuracy
5
GPT2 Layer 5 match with Layer 6
2.53LLM Eval
4
GPT2 Small
74.3Average F1
4
GPT2 Small
0.32MSE
4