method · first seen 9 Oct 2026
prefill
Also: prompt prefill, prefill acceleration for LLM inference
The initial stage of LLM inference that processes the full input prompt before token generation.
Mentions per week
PapersCodeDiscussion
Earliest sources
- RAMP: Recurrent Adaptive Mixed Precision for Efficient Prefill in Recurrent-Depth Language ModelsPapers · 9 Oct 2026 · Ke Wan, Chen Chen
- RidgeFill: Training-Free Late-Layer Cache Fill for Faster PrefillPapers · 6 Nov 2026 · Dylan Johnston, Robert P. Smith
Latest sources
- RidgeFill: Training-Free Late-Layer Cache Fill for Faster PrefillPapers · 6 Nov 2026 · Dylan Johnston, Robert P. Smith
- RAMP: Recurrent Adaptive Mixed Precision for Efficient Prefill in Recurrent-Depth Language ModelsPapers · 9 Oct 2026 · Ke Wan, Chen Chen