vLLM team says DeepSeek-V4.1-Flash inference is 5.3x faster
The story so far
Inferact and the vLLM team claim that after three weeks of optimization, DeepSeek-V4.1-Flash has improved service speed by 1.9 times in low-concurrency scenarios; under a constraint of 150 tokens per second, throughput has increased by 5.3 times. The team explains that the optimization methods include bounded replay for sliding window attention, using CUDA graphs and fused operators; their article also mentions that caching reconstruction and skipping pre-filling of some prompts can reduce server-side computation. The performance figures mentioned above are self-reported by the team.
AI-written from the reports · Updated 28 days ago
Coverage timeline
Follow the coverage to see every side of the story.
- vLLM engineeringFeaturedvLLM explains a 5.3× throughput gain for DeepSeek-V4.1-Flash
Inferact and the vLLM team report 1.9× faster low-concurrency serving and 5.3× higher throughput under a 150-tokens-per-second constraint after three weeks of optimization. The changes combine bounded replay of sliding-window attention, CUDA graphs, and fused kernels. Their post explains how cache reconstruction and skipping parts of prompt prefill reduce serving work.