Benchmark of DeepSeek-R1-Distill-Qwen Model Performance
Test Conditions
-
Test Board：S100P。
-
Performance Data Acquisition: Test a single prompt and record the metrics of TTFT (Time to First Token) and TPS (Average Tokens Per Second).
-
Python version：Python3.10。
-
Runtime Environment：Linux。
Measured data
| model | platform | dtype | seqlen | max context | TTFT(ms) | TPS | memory(GB) |
|---|---|---|---|---|---|---|---|
| DeepSeek-R1-Distill-Qwen-1.5B | S100P | q8 | 256 | 1024 | 109 | 27.08 | 1.7 |
| DeepSeek-R1-Distill-Qwen-1.5B | S100P | q4 | 256 | 1024 | 108 | 39.49 | 1.1 |
| DeepSeek-R1-Distill-Qwen-1.5B | S100P | q8 | 256 | 4096 | 226 | 23.80 | 1.8 |
| DeepSeek-R1-Distill-Qwen-1.5B | S100P | q4 | 256 | 4096 | 224 | 32.35 | 1.2 |
| DeepSeek-R1-Distill-Qwen-7B | S100P | q8 | 256 | 1024 | 544 | 6.76 | 7.4 |
On This Page