DSA Library
Daily papers
4 picked from 21 new papers
AI engineering 4
-
Read Less, Solve More: Token-Efficient Sparse Reading of External Content for AI Agents
SparseRead cuts agent token use by up to 92.9% while preserving or improving quality in 97% of settings.
-
Evolving Infra: Agent-Adaptive LLM Serving at Scale
Evolving Infra cuts P95 TTFT up to 14× by adapting compute splits and KV eviction to agent workloads.
-
Ensemble Routing for Prompt-Adaptive Cost-Aware LLM Inference
PISA-E lowers inference cost by approximately 36% at selected operating points while preserving higher task performance.
-
SteerDistill: Quantizing LLM Agents via Closed-Loop Teacher Steering
SteerDistill beats quantized agent baselines and uses 67% fewer training GPU hours on τ^2-Bench.