Large Language Model Inference on Prompt length-1
0.05TTFTCPU TEE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CPU TEEModel Architecture=GPT-2, Number of Parameters=124M, Execution Policy=CPU TEE, Prompt Length=12026.06 | 0.05 | 0.33 | 0.02 | |
| CPU TEEModel Architecture=Qwen3, Number of Parameters=0.6B, Execution Policy=CPU TEE, Prompt Length=12026.06 | 0.11 | 0.52 | 0.06 | |
| Bifrost+Model Architecture=GPT-2, Number of Parameters=124M, Execution Policy=Bifrost+, Prompt Length=12026.06 | 3.9 | 62 | 3.9 | |
| BifrostModel Architecture=GPT-2, Number of Parameters=124M, Execution Policy=Bifrost, Prompt Length=12026.06 | 7.8 | 66 | 3.9 | |
| Bifrost+Model Architecture=Qwen3, Number of Parameters=0.6B, Execution Policy=Bifrost+, Prompt Length=12026.06 | 22 | 177 | 22 | |
| BifrostModel Architecture=Qwen3, Number of Parameters=0.6B, Execution Policy=Bifrost, Prompt Length=12026.06 | 45 | 202 | 22 | |
| Pure FHEModel Architecture=GPT-2, Number of Parameters=124M, Execution Policy=Pure FHE, Prompt Length=12026.06 | 68 | 579 | 34 | |
| Pure FHEModel Architecture=Qwen3, Number of Parameters=0.6B, Execution Policy=Pure FHE, Prompt Length=12026.06 | 229 | 1,031 | 115 |