Definition Generation on Biomedical domain tokens
100Similarity ScoreOriginal tokenization
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Original tokenizationBackbone=Llama3-8B2025.05 | 100 | 94.3 | |
| Original tokenizationBackbone=Llama3-8B-i2025.05 | 100 | 98.4 | |
| Original tokenizationBackbone=Llama3.1-8B2025.05 | 100 | 94.7 | |
| Original tokenizationBackbone=Llama3.1-8B-i2025.05 | 100 | 98.4 | |
| Original tokenizationBackbone=Llama3.2-3B2025.05 | 100 | 93.2 | |
| Original tokenizationBackbone=Llama3.2-3B-i2025.05 | 100 | 93.8 | |
| Original tokenizationBackbone=Avg.2025.05 | 99.9 | 95.8 | |
| Original tokenizationBackbone=Mistral-7B2025.05 | 99.8 | 96.5 | |
| Original tokenizationBackbone=OLMo2-7B-i2025.05 | 99.4 | 96.9 | |
| Token DistillationBackbone=OLMo2-7B-i2025.05 | 83.8 | 89.5 | |
| Token DistillationBackbone=Llama3.1-8B-i2025.05 | 81.6 | 89.8 | |
| Token Distillation + αNTPBackbone=OLMo2-7B-i2025.05 | 81.2 | 86.9 | |
| Token Distillation + αNTPBackbone=Llama3.1-8B-i2025.05 | 81.1 | 89.5 | |
| Token Distillation + αNTPBackbone=Llama3-8B-i2025.05 | 80.9 | 91.6 | |
| Token DistillationBackbone=Mistral-7B2025.05 | 79.7 | 85.2 | |
| Token DistillationBackbone=Llama3-8B-i2025.05 | 79.7 | 91 | |
| Token Distillation + αNTPBackbone=Llama3.1-8B2025.05 | 77.5 | 81.2 | |
| Token Distillation + NTPBackbone=Mistral-7B2025.05 | 77.3 | 82.6 | |
| Token Distillation + NTPBackbone=Llama3.1-8B-i2025.05 | 77.3 | 86.1 | |
| Token Distillation + αNTPBackbone=Mistral-7B2025.05 | 76.8 | 82 | |
| Token Distillation + αNTPBackbone=Avg.2025.05 | 76.7 | 83.3 | |
| Token DistillationBackbone=Llama3.1-8B2025.05 | 75.8 | 80.1 | |
| Token Distillation + NTPBackbone=Llama3-8B-i2025.05 | 75.8 | 86.3 | |
| Token DistillationBackbone=Llama3.2-3B-i2025.05 | 75 | 83 | |
| Token DistillationBackbone=Llama3-8B2025.05 | 72.7 | 76.2 | |
| Token Distillation + αNTPBackbone=Llama3.2-3B2025.05 | 72.7 | 79.5 | |
| Token Distillation + NTPBackbone=Llama3.1-8B2025.05 | 72.3 | 79.9 | |
| Token Distillation + αNTPBackbone=Llama3.2-3B-i2025.05 | 71.9 | 80.5 | |
| Token Distillation + αNTPBackbone=Llama3-8B2025.05 | 71.7 | 75.2 | |
| ZeTTBackbone=Llama3-8B-i2025.05 | 70.7 | 80.3 | |
| Token Distillation + NTPBackbone=Avg.2025.05 | 70.5 | 78.5 | |
| ZeTTBackbone=Llama3-8B2025.05 | 69.5 | 73.6 | |
| Token DistillationBackbone=Avg.2025.05 | 68.5 | 74.4 | |
| Token Distillation + NTPBackbone=Llama3-8B2025.05 | 68.4 | 73.8 | |
| Token Distillation + NTPBackbone=OLMo2-7B-i2025.05 | 65.4 | 74.6 | |
| Token Distillation + NTPBackbone=Llama3.2-3B-i2025.05 | 65.2 | 75.8 | |
| ZeTTBackbone=Mistral-7B2025.05 | 63.5 | 68.6 | |
| Token Distillation + NTPBackbone=Llama3.2-3B2025.05 | 62.3 | 69.1 | |
| NTP (Lampinen et al., 2018)Backbone=Llama3.1-8B-i2025.05 | 60.2 | 69.9 | |
| NTP (Lampinen et al., 2018)Backbone=Llama3.1-8B2025.05 | 58.6 | 65.6 | |
| NTP (Lampinen et al., 2018)Backbone=OLMo2-7B-i2025.05 | 52.1 | 57 | |
| NTP (Lampinen et al., 2018)Backbone=Avg.2025.05 | 52 | 59.4 | |
| NTP (Lampinen et al., 2018)Backbone=Llama3.2-3B2025.05 | 51.6 | 58.4 | |
| NTP (Lampinen et al., 2018)Backbone=Llama3.2-3B-i2025.05 | 50 | 58.8 | |
| NTP (Lampinen et al., 2018)Backbone=Mistral-7B2025.05 | 49.8 | 58.2 | |
| NTP (Lampinen et al., 2018)Backbone=Llama3-8B-i2025.05 | 48.4 | 58 | |
| NTP (Lampinen et al., 2018)Backbone=Llama3-8B2025.05 | 45.3 | 49.6 | |
| NTP (tune all embeddings)Backbone=Llama3.1-8B-i2025.05 | 41.4 | 50 | |
| NTP (tune all embeddings)Backbone=Llama3-8B-i2025.05 | 34.4 | 43.4 | |
| NTP (tune all embeddings)Backbone=Llama3-8B2025.05 | 33.2 | 39.5 | |
| NTP (tune all embeddings)Backbone=Llama3.1-8B2025.05 | 28.5 | 35 | |
| NTP (tune all embeddings)Backbone=Avg.2025.05 | 28.1 | 33.3 | |
| NTP (tune all embeddings)Backbone=Mistral-7B2025.05 | 26.8 | 30.3 | |
| MeanBackbone=Llama3.1-8B2025.05 | 25.6 | 25.6 | |
| MeanBackbone=Llama3.1-8B-i2025.05 | 23.2 | 28.3 | |
| NTP (tune all embeddings)Backbone=OLMo2-7B-i2025.05 | 22.5 | 26.6 | |
| MeanBackbone=OLMo2-7B-i2025.05 | 20.1 | 24.2 | |
| NTP (tune all embeddings)Backbone=Llama3.2-3B2025.05 | 19.7 | 20.7 | |
| NTP (tune all embeddings)Backbone=Llama3.2-3B-i2025.05 | 18.6 | 21.1 | |
| MeanBackbone=Llama3-8B-i2025.05 | 17.6 | 20.7 | |
| MeanBackbone=Llama3-8B2025.05 | 16.8 | 16.2 | |
| MeanBackbone=Avg.2025.05 | 16.6 | 18.6 | |
| MeanBackbone=Llama3.2-3B2025.05 | 15.2 | 15.2 | |
| MeanBackbone=Llama3.2-3B-i2025.05 | 12.1 | 14.6 | |
| MeanBackbone=Mistral-7B2025.05 | 2.5 | 4.3 | |
| RandomBackbone=Mistral-7B2025.05 | 0 | 0.4 | |
| RandomBackbone=Llama3-8B2025.05 | 0 | 0 | |
| RandomBackbone=Llama3-8B-i2025.05 | 0 | 0.2 | |
| RandomBackbone=Llama3.1-8B2025.05 | 0 | 0 | |
| RandomBackbone=Llama3.1-8B-i2025.05 | 0 | 0 | |
| RandomBackbone=Llama3.2-3B2025.05 | 0 | 0 | |
| RandomBackbone=Llama3.2-3B-i2025.05 | 0 | 0 | |
| RandomBackbone=OLMo2-7B-i2025.05 | 0 | 0 | |
| RandomBackbone=Avg.2025.05 | 0 | 0.1 | |
| Token DistillationBackbone=Llama3.2-3B2025.05 | 0 | 0 |