Language Modeling on WikiText-style (PPL, Speedup)
17.42Perplexity (PPL)TinyLlama-1.1B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| TinyLlama-1.1BSeq PPL=17.42, Config=8xF1-h0, K=82026.05 | 17.42 | 0.65 | |
| Qwen2.5-0.5BSeq PPL=31.14, Config=8xF1-h0, K=82026.05 | 31.14 | 0.78 | |
| Gemma-3-1BSeq PPL=229.36, Config=8xF1-h0, K=82026.05 | 229.41 | 0.74 | |
| Gemma-3-1BSeq PPL=229.36, Config=8xF1-h0, K=42026.05 | 233.22 | 0.88 |