Sparse Probing on SAEBench
81.9Average F1 Scorefmxcoder (p = 0)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| fmxcoder (p = 0)LLM=Gemma2-2b2026.05 | 81.9 | 198.7 | |
| fmxcoder (p = 0)LLM=Pythia-1.4b2026.05 | 78.4 | 36.1 | |
| fmxcoder (p = 0)LLM=GPT2-Small2026.05 | 78.3 | 120.2 | |
| fmxcoder (p = 0)LLM=Pythia-410m2026.05 | 74.3 | 10.2 | |
| fmxcoder (p = 0.05)LLM=GPT2-Small2026.05 | 73.5 | 68.4 | |
| fmxcoder (p = 0.05)LLM=Gemma2-2b2026.05 | 73 | 39.1 | |
| fmxcoder (p = 0.05)LLM=Pythia-1.4b2026.05 | 71.2 | 13.1 | |
| fmxcoder (p = 0.05)LLM=Pythia-410m2026.05 | 68.1 | 4.5 | |
| CrosscoderLLM=GPT2-Small2026.05 | 67.6 | 11.6 | |
| SAELLM=GPT2-Small2026.05 | 65.7 | 8.4 | |
| SAELLM=Pythia-410m2026.05 | 65 | 0.5 | |
| SAELLM=Gemma2-2b2026.05 | 64.8 | 2.7 | |
| SAELLM=Pythia-1.4b2026.05 | 64.6 | 0.8 | |
| CrosscoderLLM=Gemma2-2b2026.05 | 51.2 | 0.2 | |
| CrosscoderLLM=Pythia-410m2026.05 | 47.4 | 0.1 | |
| CrosscoderLLM=Pythia-1.4b2026.05 | 47 | 0.2 |