Mathematical Reasoning on GSM8k (Throughput and Latency)
39.69Throughput (tokens/s)PACER
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PACERLLM Family/Size=Deepseek 1.3B/33B2026.02 | 39.69 | 2.17 | |
| SpeculativeLLM Family/Size=Deepseek 1.3B/33B2026.02 | 35.08 | 1.92 | |
| AssistLLM Family/Size=Deepseek 1.3B/33B2026.02 | 34.86 | 1.91 | |
| LookaheadLLM Family/Size=Deepseek 1.3B/33B2026.02 | 33.82 | 1.85 | |
| PACERLLM Family/Size=Deepseek 6.7B/33B2026.02 | 31.07 | 1.88 | |
| SpeculativeLLM Family/Size=Deepseek 6.7B/33B2026.02 | 29.36 | 1.78 | |
| LookaheadLLM Family/Size=Deepseek 6.7B/33B2026.02 | 29.09 | 1.76 | |
| AssistLLM Family/Size=Deepseek 6.7B/33B2026.02 | 28.8 | 1.74 | |
| RESTLLM Family/Size=Deepseek 1.3B/33B2026.02 | 23.25 | 1.27 | |
| RESTLLM Family/Size=Deepseek 6.7B/33B2026.02 | 22.22 | 1.34 | |
| PACERLLM Family/Size=Llama-2 7B/70B2026.02 | 19.07 | 2.08 | |
| VanillaLLM Family/Size=Deepseek 1.3B/33B2026.02 | 18.25 | 1 | |
| SpeculativeLLM Family/Size=Llama-2 7B/70B2026.02 | 17.81 | 1.94 | |
| VanillaLLM Family/Size=Deepseek 6.7B/33B2026.02 | 16.51 | 1 | |
| AssistLLM Family/Size=Llama-2 7B/70B2026.02 | 14.02 | 1.53 | |
| RESTLLM Family/Size=Llama-2 7B/70B2026.02 | 12.66 | 1.38 | |
| LookaheadLLM Family/Size=Llama-2 7B/70B2026.02 | 12.07 | 1.31 | |
| VanillaLLM Family/Size=Llama-2 7B/70B2026.02 | 9.19 | 1 |