Engineering
WUSH-KV: KV Cache Quantization with Data-Adaptive Transforms
Researchers introduce WUSH-KV, a method for low-bit quantization of the key-value cache used in long-context LLM inference. The technique builds…
Engineering
SeLMRoute: Probabilistic Semantic Evidence for Large Language Model Routing
Researchers introduce SeLMRoute, a framework for routing queries to the most suitable large language model among several candidates. Instead of…
Engineering
Mid-Harness: Scaling Actions Between Model and Harness for Terminal Agents
Researchers present Mid-Harness, a method that allocates test-time compute at the boundary between a language model and its execution harness…
Engineering
Decoding Looped Transformers Better for (Almost) Free
Researchers present LoopCD, a training-free contrastive decoding framework for looped transformers that contrasts a model’s final prediction against an earlier…
Engineering
Beyond Memory: Harnessing Long-Horizon Agents with Explicit Belief States
A new paper introduces PoS, an inference-time framework that constructs and continually maintains explicit belief states as an LLM agent’s…
Engineering
ActiveSaddler: Automated Curriculum Learning for Agent Harness Optimization
Researchers introduce ActiveSaddler, a framework that treats the training curriculum for LLM agent harness optimization as a non-stationary bandit problem…
Engineering
Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It
Researchers studied how far pretrained transformers actually propagate reference chains through their depth, finding that thirteen base models reliably followed…
Engineering
AgSpec: Pushing the Limits of Retrieval-Based Speculative Decoding in Coding Agent Pipelines
AgSpec is a new speculative decoding framework for coding agent pipelines that addresses the problem of reusable draft text being…
Engineering
Pretrain Once, Route Anywhere: Towards a Foundation Model for LLM Routing
The paper proposes treating LLM routing, the task of assigning incoming queries to the best-fit model from a heterogeneous pool,…
Engineering
Prefill-Free Cross-Family KV Cache Transfer for Heterogeneous Multi-Agent LLMs
Researchers propose HeteroFold, a prefill-free method for transferring key-value cache between heterogeneous large language models in multi-agent systems, letting a…
Engineering
GraphForge: Training Working Agents with Graph-Anchored Workspace Synthesis
Researchers introduce GraphForge, an evidence-graph based framework for synthesizing agent training data that grounds both tasks and their verification rubrics…
Engineering
MILO: Automated Harness Discovery via Orchestrated Multi-Agent Evolution
Researchers introduce MILO (Meta-evolutionary Island Orchestration), a framework that co-evolves agent harnesses alongside the search strategy used to discover them,…