Papers for

software infrastructure teams

Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.

KV-streams boost training speed for long-context agentic AI models

KV-streams for Efficient Compaction in Agentic Reinforcement Learning

Abstract: Scaling the horizon of agentic LLMs is bottlenecked by the need to fit ever longer context traces in GPU memory. Context compaction has been the most popular mechanism to alleviate this issue, keeping GPU memory constant for a given trace. Unfortunately, most compaction strategies rely on prefilling the LLM context many times over, hindering training throughput. To alleviate this bottleneck and enable efficient trainable compaction, we propose KV-streams, a plug-and-play strategy compatible with any compaction strategy that substantially increases throughput while showing no evidence of hindering performance. KV-streams enable scalable compaction by streaming the KV cache forward rather than flushing it after each compaction. We show that KV-streams enable three different compaction strategies, achieving a 2.6 to 5x wall-clock speedup in training. Beyond efficiency, we find that the streamed KV cache can act as a recurrent state, carrying forward information that has long since disappeared from the context. Specifically, in a controlled setting we show that, contrary to prior work, RL alone is all that is needed for this behavior to emerge. Overall, we show KV-streams to be an efficient and lightweight plug-and-play addition to any post-training pipeline.

Mon 28 SeptMachine LearningArtificial Intelligence
The gist
Large language models that act like agents need to remember long conversations or experiences, but this uses a lot of GPU memory and slows training down. The authors propose KV-streams, a way to keep memory use steady and speed training by streaming data forward instead of restarting it all the time. This makes training 2.6 to 5 times faster without hurting how well the model learns. They also found that this streamed memory helps the model remember important information for longer, improving its decision-making.
Open → 2609.35750v1

Large language models safely guide compiler optimizations with formal checks

Verified Learning for Compiler Optimization: An LLM-Guided Architecture with Formal Control

Abstract: Compiler optimizations traditionally rely on handcrafted heuristics that often fail to generalize across programs and architectures. We investigate whether large language models can participate in compiler optimization through a verification-centered systems architecture that couples generative rewriting with formal equivalence checking. Using lazification in LLVM IR as a case study, we fine-tune a code-centric LLM on transformations produced by Wyvern and embed Alive2 into a feedback loop that enforces semantic preservation for every generated rewrite. Correctness is enforced externally as a runtime control layer rather than learned implicitly. During inference, candidate transformations are symbolically validated and regenerated when necessary, ensuring accepted rewrites satisfy formal constraints. On the LLVM test suite, the fine-tuned model reproduces core optimization behaviors while applying fewer transformations overall. Although Wyvern remains faster on most benchmarks, 9.8% achieve comparable or improved runtime under the learned system, with no semantic violations observed. Verification overhead remains bounded and convergence stable. These results demonstrate that generative AI components can be safely integrated into compiler pipelines through deterministic validation and structured feedback, offering a scalable architectural pattern for trustworthy AI-driven software infrastructure.

Wed 23 SeptSoftware Engineering
The gist
Compiler optimizations make programs run faster by changing code, but mistakes can break the program. This paper shows how a large language model (LLM) can suggest code changes while a formal system checks each change to keep the program correct. The authors tested this on an existing compiler and found the system often kept changes safe and sometimes improved program speed. This method blends AI creativity with strict safety checks to help build reliable software tools.
Open → 2609.27214v1