PG
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
PG-19
5Perplexity
244
PG-19 32K context length
329.61Throughput
32
30-page PG (test)
96.9Precision
8
PG-19 20K context length
4Throughput Speedup (micro)
6
PG-19 10K context length
2.92Throughput Speedup (micro-averaged)
6
30-page PG (test)
1.05CER (%)
4
PG19
8.8Tie Percentage
2
30-page PG (test)
98Precision
1
30-page PG (test)
98.3Precision
1