Language Modeling on GPT-2 Evaluation Set
29.91Hyper-Prior BPTDirect-Access
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Direct-AccessSplit Pt.=6, λ=0.0252026.01 | 29.91 | 215.04 | 22.91 | |
| Direct-AccessSplit Pt.=6, λ=0.0102026.01 | 30.24 | 160.16 | 26.24 | |
| Direct-AccessSplit Pt.=6, λ=0.0752026.01 | 30.47 | 165.94 | 25.22 | |
| Direct-AccessSplit Pt.=6, λ=0.0012026.01 | 32.89 | 263.62 | 22.43 | |
| ProposedSplit Pt.=3, λ=0.0752026.01 | 33.63 | 152.25 | 18.36 | |
| ProposedSplit Pt.=3, λ=0.0102026.01 | 33.97 | 148.68 | 19.1 | |
| ProposedSplit Pt.=6, λ=0.0102026.01 | 35.06 | 207.36 | 22.88 | |
| ProposedSplit Pt.=6, λ=0.0752026.01 | 36.31 | 214.98 | 22.69 | |
| ProposedSplit Pt.=9, λ=0.0102026.01 | 39.03 | 268.34 | 25.12 | |
| ProposedSplit Pt.=3, λ=0.0252026.01 | 39.2 | 189.74 | 17.42 | |
| ProposedSplit Pt.=6, λ=0.0252026.01 | 40.48 | 273.06 | 21.25 | |
| ProposedSplit Pt.=9, λ=0.0752026.01 | 40.53 | 288.44 | 24.59 | |
| ProposedSplit Pt.=3, λ=0.0012026.01 | 41.43 | 230.16 | 16.98 | |
| ProposedSplit Pt.=6, λ=0.0012026.01 | 45.55 | 355.93 | 20.89 | |
| ProposedSplit Pt.=9, λ=0.0252026.01 | 47.01 | 388.11 | 22.8 | |
| ProposedSplit Pt.=9, λ=0.0012026.01 | 52.71 | 492.08 | 21.61 | |
| FourierSplit Pt.=6, λ=0.0252026.01 | 248.14 | 464.49 | 21.7 | |
| FourierSplit Pt.=6, λ=0.0012026.01 | 249.18 | 541.67 | 21.15 | |
| FourierSplit Pt.=6, λ=0.0102026.01 | 302.62 | 422.32 | 24.19 | |
| FourierSplit Pt.=6, λ=0.0752026.01 | 309.2 | 434.07 | 24.03 |