vLLM adds lossless text watermarking based on Gumbel-max
Watermarking in vLLM
AI BriefingvLLM announced support for lossless watermarking based on the Gumbel-max algorithm, integrating it into the sampling pipeline of Model Runner v2. PRs #54053, #56122, and #56233 implement a fused GPU kernel, a dual-key scheme, and context deduplication to support speculative decoding while preserving output diversity.
Source: vLLM official blog · vllm.ai
Event follow-up1
View all event coverage- vLLM adds lossless text watermarking based on Gumbel-maxvLLM Official Blog · September 25