phenomenon · first seen 14 Oct 2026
prefill phase
The initial LLM inference stage that processes all input tokens in parallel before generating output.
Mentions per week
PapersCodeDiscussion
Earliest sources
- The Cost of Context: Profiling the Energy Footprint of Input Tokens in Large Language ModelsPapers · 14 Oct 2026 · Boris Ruf, Marcin Detyniecki
Latest sources
- The Cost of Context: Profiling the Energy Footprint of Input Tokens in Large Language ModelsPapers · 14 Oct 2026 · Boris Ruf, Marcin Detyniecki