@liliana_hotsko: How do you give a code LLM knowledge of an entire repository without paying for it at every single query? We introduce …
Summary
Introduces Code2LoRA, a hypernetwork that converts an entire code repository into a LoRA adapter for code LLMs, eliminating inference-time token overhead for repository-level knowledge.
View Cached Full Text
Cached at: 06/13/26, 02:46 PM
How do you give a code LLM knowledge of an entire repository without paying for it at every single query?
We introduce Code2LoRA: a hypernetwork that turns a repository into its own LoRA adapter. Repo knowledge baked into weights → zero inference-time token overhead. https://t.co/T3hnxCVkna
Similar Articles
What evidence should AI coding agents leave before saying “done”?
Discusses the need for AI coding agents to provide evidence of their work before marking tasks as complete, exploring verification strategies and best practices.
@gdb: Codex has gotten very good
Corey Quinn apologizes for underestimating OpenAI's Codex, praising its significant improvement and capabilities.
Learning to Select, Not Relearn: Hard-Routed Mixtures of Reasoning LoRAs
Proposes Hard-Routed MoR-LoRA, a two-stage framework that composes frozen reasoning LoRA experts via hard top-1 routing, preserving expert behavior with fewer trainable parameters than soft-routing baselines.
A Three-Phase Foundation Model for Tax-Aware Personalized Portfolio Management
A three-phase deep reinforcement learning system for personalized portfolio management that addresses ticker lock-in, monolithic objectives, and static user models, using a cross-asset encoder pretrained with self-supervised learning and the Chronos time series foundation model, fine-tuned with Mixture of Experts and PPO, and personalized via LoRA.
Clinically Structured Rank-Gated LoRA for Cross-Benchmark Medical Question Answering
The paper proposes BiRG-LoRA, a rank-gated LoRA method for medical question answering that uses clinically structured priors to select sparse rank subsets, achieving 69.31% macro-average accuracy across four benchmarks while using fewer parameters than mixture-of-experts approaches.