Secure LLM Inference on Prompt Lengths (16, 64)
4.269TTFTBifrost+
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Bifrost+Model=GPT-2 (124M), Prompt=162026.06 | 4.269 | 63.399 | 3.942 | |
| Bifrost+Model=GPT-2 (124M), Prompt=642026.06 | 5.459 | 63.659 | 3.88 | |
| Bifrost+Model=Qwen3 (0.6B), Prompt=162026.06 | 23.023 | 177.261 | 22.034 | |
| Bifrost+Model=Qwen3 (0.6B), Prompt=642026.06 | 26.585 | 182.335 | 22.25 | |
| BifrostModel=GPT-2 (124M), Prompt=162026.06 | 62.199 | 120.579 | 3.892 | |
| BifrostModel=GPT-2 (124M), Prompt=642026.06 | 249.849 | 308.859 | 3.934 | |
| BifrostModel=Qwen3 (0.6B), Prompt=162026.06 | 352.302 | 506.631 | 22.047 | |
| Pure FHEModel=GPT-2 (124M), Prompt=162026.06 | 579.19 | 1,090.24 | 34.07 | |
| BifrostModel=Qwen3 (0.6B), Prompt=642026.06 | 1,419.378 | 1,574.631 | 22.179 | |
| Pure FHEModel=Qwen3 (0.6B), Prompt=162026.06 | 1,946.84 | 2,748.48 | 114.52 | |
| Pure FHEModel=GPT-2 (124M), Prompt=642026.06 | 2,214.55 | 2,725.6 | 34.07 | |
| Pure FHEModel=Qwen3 (0.6B), Prompt=642026.06 | 7,443.8 | 8,245.44 | 114.52 |