single mention · software implementation
Write-ahead log (WAL)
A write-ahead log here is a token-granular log kept per generation request in a rollout service.1 Each newly generated token is appended immediately, and on preemption or hardware failure the persisted WAL tokens let decoding resume or the KV cache be reconstructed.2
Trade-off Avoids the length bias and re-decoding cost of regenerating unfinished requests from scratch.3
How this definition was written
Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.
we implement a token-granular Write-Ahead Log (WAL) for each generation request
— DeepSeek-V4: Towards Highly Efficient Million-Token Context IntelligenceWhenever a new token is generated for a request, we immediately append it to that request’s WAL
— DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligenceit is mathematically incorrect to regenerate unfinished requests from scratch, as this introduces length bias
— DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence
Also called Write-Ahead Log.
Filed under software implementation › infrastructure service