KV-streams trains SWE agents 2x faster by preserving KV cache across compaction
burny_tech · x · 2026-10-01
Researchers introduce KV-streams, which preserves the KV cache during agentic compaction instead of flushing it, avoiding re-prefill cost on both the inference engine and the trainer.
Key points:
- Unlike works assuming a fixed compaction strategy, KV-streams works with any compaction method on agentic tasks
- Validated on text-based game experiments, then scaled to SWE tasks
- Matches the performance of re-prefill compaction and full context at half the training time
More from coding & agent
- Generic API-catalog MCP server vs one tool per job: a token-efficiency tradeoff — MandeepKaur9407 · 2026-11-16
- sessionwiki turns your past coding-agent sessions into a searchable local wiki for agents — Cold_Past7184 · 2026-11-16
- Developer hits 700+ commits/day with AI, asks how to manage parallel agent work — blixt · 2026-11-16
- Alibaba's G-CAST: governed continual self-training keeps e-commerce agents fresh in production — Yuechun Yu · 2026-11-16
- harness-evals: open-source library to compare agent versions with success-vs-cost plots — slavadubrov · 2026-11-16
- Developers poll real costs of running coding agents on remote VMs for work — radim11 · 2026-11-16