Topic / Trend Rising

AI Agents, Memory and Inference Optimization

New tools and benchmarks focus on agent memory management, token reduction for coding agents, and non-blocking multimodal inference. Together Compiler, AMD Loom and ICARE highlight the shift toward continuous optimization and lifelong learning for AI agents.

Detected: 2026-11-20 ยท Updated: 2026-11-20

Related Coverage

2026-11-20 โ€ข ArXiv cs.CL

ICARE: continual learning without forgetting previous tasks

ICARE is a framework for cross-domain text classification that mitigates catastrophic forgetting in NLI systems. It combines instance selection based on representation learning and knowledge distillation at feature and logit levels, using pruning and...

#LLM On-Premise #Fine-Tuning #DevOps
2026-11-18 โ€ข ArXiv cs.LG

AdaFlowLite: non-blocking inference for asynchronous mobile sensor data

AdaFlowLite shifts from waiting for all multimodal data to starting inference as soon as partial data arrives. Tests show up to 80.4% lower latency, up to 62.1% higher accuracy, and nearly 50% lower energy consumption. Flexible imputation without ret...

#Hardware #LLM On-Premise #DevOps
2026-11-17 โ€ข Phoronix

AMD Loom 0.2 Brings Broader C++ Support and Better Targeting for AI Agents

AMD today released Loom 0.2 within ROCm HRX, updating the compiler, runtime, and serving infrastructure for AI agent workflows. The new version expands C++ support and improves targeting. The update matters mainly for teams building AI stacks on AMD ...

#Hardware #LLM On-Premise #DevOps
2026-11-16 โ€ข Microsoft Research

TokenEconomizer: An Open-Source Plugin to Cut Token Use in AI Coding Agents

An open-source plugin from UC Berkeley and Microsoft Research cuts token consumption of AI coding agents by 18.4 to 30.9 percent across 277 real tasks, without touching models or agents. It learns from past repository sessions, prunes unused tools, a...

#LLM On-Premise #Fine-Tuning #DevOps
2026-11-14 โ€ข ArXiv cs.AI

Together Compiler: continuous optimization accelerates AI inference up to 2.6x

Together AI introduces TKCC, a continuous compilation system built on agents for inference. It tackles the measurement throughput-fidelity trade-off with a tiered persistent engine and a cost model. In tests it outperforms the best open source engine...

#Hardware #LLM On-Premise #DevOps
โ† Back to All Topics