The sample counts with minimal perplexity at a given 1/vt

(different lines represent different position segments at the 16k-token samples)

0- 2048

2048 - 4096

500

4096 - 6144

6144 - 8192

8192 - 10240

400

10240 - 12288

12288 - 14336

14336 - 16384

300

200

100

1.10

115

1.20

1.25

1.30

1.35

1.40

vt
