Language Modeling on C4 300K-token sample
12.24PerplexityFP16
Evaluation Results
| Method | Links | |
|---|---|---|
| FP16Bit-width=FP162026.05 | 12.24 | |
| QTIP 2-bit PTQBit-width=2-bit2026.05 | 14.16 | |
| BCJR-QAT-N4Bit-width=2 bpw, Configuration=N42026.05 | 14.8 | |
| BCJR-QAT-v2Bit-width=2 bpw, Configuration=v22026.05 | 16.38 |