Table 8 Model Statistics of pretrained transformer models
Architecture | Training | Canonical | Enumerated | Time (hh:mm) | Size | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
Character accuracy | Sequence accuracy | Character accuracy | Sequence accuracy | ||||||||
Encoder only | C2C | 1.000 | 1.000 | 1.000 | 1.000 | 01:16 | 6.7M | ||||
R2C | 0.202 | 0.000 | 0.195 | 0.000 | 00:42 | 6.7M | |||||
E2C | 1.000 | 1.000 | 1.000 | 1.000 | 03:33 | 6.7M | |||||
MC2C | 0.999 | 0.993 | 0.994 | 0.819 | 01:25 | 6.7M | |||||
MR2C | 0.202 | 0.000 | 0.196 | 0.000 | 00:46 | 6.7M | |||||
ME2C | 1.000 | 0.991 | 0.999 | 0.962 | 10:31 | 6.7M | |||||
Encoder-decoder | C2C | 0.998 | 0.995 | 0.994 | 0.994 | 0.998 | 0.996 | 0.994 | 0.994 | 01:45 | 15.9M |
R2C | 0.813 | 0.089 | 0.000 | 0.000 | 0.554 | 0.082 | 0.000 | 0.000 | 02:44 | 15.9M | |
E2C | 0.998 | 0.995 | 0.995 | 0.995 | 0.998 | 0.995 | 0.994 | 0.994 | 08:43 | 15.9M | |
MC2C | 0.998 | 0.996 | 0.993 | 0.987 | 0.976 | 0.965 | 0.463 | 0.452 | 02:53 | 15.9M | |
MR2C | 0.812 | 0.040 | 0.000 | 0.000 | 0.549 | 0.040 | 0.000 | 0.000 | 05:56 | 15.9M | |
ME2C | 0.998 | 0.994 | 0.983 | 0.958 | 0.998 | 0.994 | 0.979 | 0.936 | 24:04 | 15.9M | |