Linguistic Evaluation on T-Story Cloze
84.6T-Story Cloze AccuracySpiritLM-base-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| SpiritLM-base-7BTrain Data Size (Hours)=570K, Tokenizer=SSL2025.09 | 84.6 | |
| Moshi-7B (Base)Train Data Size (Hours)=7,000K, Tokenizer=RVQ2025.09 | 81.8 | |
| SpiritLM-exp-7BTrain Data Size (Hours)=570K, Tokenizer=SSL2025.09 | 75.7 | |
| TWIST-1.3BTrain Data Size (Hours)=150K, Tokenizer=SSL2025.09 | 69.9 | |
| Flow-SLM-1B-extTrain Data Size (Hours)=110K, Tokenizer=RVQ2025.09 | 65.6 | |
| Llama-Mimi-1.3BTrain Data Size (Hours)=240K, Tokenizer=RVQ2025.09 | 64 | |
| CSM-1.3BTrain Data Size (Hours)=240K, Tokenizer=RVQ2025.09 | 58 |