Unverified50% confidenceBenchmarkExact time
TokenSpeed's MLA decode kernel has about half the latency of TensorRT-LLM under typical speculative decoding workloads.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/15/2026
First Seen
Valid until: 10/29/2026
Sources
Related Entities
Related Claims
Unverifiedtokens/s指标可细分为首token延迟（TTFT）和生成速率（Generation Throughput）两个子维度70% similarUnverifiedIn the inference pipeline, the Prefill stage is compute-intensive and time-consuming, while the Decode stage performs token-by-token autoregressive generation and is sensitive to latency.70% similarUnverifiedThe upper speedup limit of the decoding stage (approximately 2.7×) is higher than the upper speedup limit of the encoding stage.69% similarUnverifiedTensorRT-LLM通过算子融合、量化压缩和连续批处理将单token生成延迟降低40%-60%69% similarUnverifiedThrough post-training methods, Swift-Qwen3.8-27B reduces thinking tokens by 58.3% and improves inference speed by 1.95×, while keeping accuracy loss within 1%.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1041083API
curl https://kongchang.com/api/v1/knowledge/claims/1041083MCP
get_claim(id=1041083)