Language Modeling on LAMBADA 500 examples (Accuracy)
23Accuracy (LAMBADA 500)INT4
Evaluation results
| Method | Links | |
|---|---|---|
| INT4Model=Mistral, Bits=5122026.04 | 23 | |
| INT4Model=GPT-2, Bits=2562026.04 | 7.2 | |
| Rank-32 FP16Model=Mistral, Bits=5122026.04 | 0.4 | |
| Rank-16 FP16Model=GPT-2, Bits=2562026.04 | 0.2 |