MERA: Model Evolution and Routing with Skill Adaptation for Agentic Systems at Scale

• URL: http://arxiv.org/abs/2608.10333v1
• Date: Tue, 11 Aug 2026 00:28:02 GMT
• Title: MERA: Model Evolution and Routing with Skill Adaptation for Agentic Systems at Scale
• Abstract summary: LLM agents execute heterogeneous sequences of model calls within a single task.<n>Prior routing methods assign easy invocations to a cheaper small model and difficult ones to a large model.<n> MERA instead improves the small model itself, using a single model invocation as the unit of adaptation.
• Score: 23.327794385289952
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: LLM agents execute heterogeneous sequences of model calls within a single task: some invocations require careful reasoning, while others are structured steps such as formatting or tool-argument construction. Prior routing methods exploit this asymmetry by assigning easy invocations to a cheaper small model and difficult ones to a large model. Such policies reduce inference cost, but they leave the small model's capability unchanged, so attainable savings remain bounded by the work the student can already solve. MERA instead improves the small model itself, using a single model invocation as the unit of adaptation. In each cycle, MERA replays failed student invocations to obtain execution-verified teacher demonstrations, distills recurring procedures into an iteratively updated SkillBook, and fine-tunes a student LoRA adapter via supervised learning and optional GRPO. Routing serves as supporting machinery for deployment: the improved student is served behind a cost-calibrated router with verifier-backed fallback, and a candidate SkillBook, adapter, or router is admitted only when joint replay preserves task quality. Empirically, four-cycle adaptation raises Qwen2.5-Coder-1.5B from 28.7% to 49.7% pass on held-out HumanEval+MBPP. Under verifier-backed fallback, the deployed policy retains 88.3% pass at 60.8% of always-Luna cost. On TAU-2, a fine-tuned Qwen3.5-2B improves from 14/35 to 18/35 and matches an unadapted 4B model. These results indicate that verifier-backed multi-cycle adaptation can increase small-model capability, rather than only routing around a fixed student.

Related papers

• TACIT-Switch: Cost-Aware Model Escalation for LLM Agents from Censored Supervision [2.681967953218225]
  Agents with smaller language-model backbones can drift into persistent failure modes, whereas those with larger backbones are generally more reliable but more costly.<n>This reliability-cost trade-off motivates routing methods that decide when to invoke an agent with a larger backbone.<n>Our method, TACIT-SWITCH, learns permanent handoff policies from accumulated trajectory evidence and Teacher-Annotated Censored Intervention Times.
  arXiv  Detail & Related papers  (2026-08-28T04:33:11Z)
• DuoMem: Towards Capable On-Device Memory Agents via Dual-Space Distillation [26.38620865128242]
  DuoMem is a dual-space distillation framework that transfers procedural problem-solving ability from a large teacher model to compact student models.<n>It is evaluated on ALFWorld, a challenging embodied decision-making benchmark.<n>The DuoMem-enhanced 4B model completes tasks over 3x faster than the 72B teacher in wall-clock time.
  arXiv  Detail & Related papers  (2026-06-29T08:38:40Z)
• Three Roles, One Model: Role Orchestration at Inference Time to Close the Performance Gap Between Small and Large Agents [0.4666493857924357]
  We study whether inference-time scaffolding alone, without any additional training compute, can improve the performance of a small model in complex multi-step environments.<n>We evaluate Qwen3-8B on the AppWorld benchmark under both full-precision and 4-bit quantized configurations.<n>On full-precision inference, our scaffolded 8B model surpasses DeepSeek-Coder 33B Instruct (7.1%) from the original AppWorld evaluation.
  arXiv  Detail & Related papers  (2026-04-13T13:40:33Z)
• Quantization-Robust LLM Unlearning via Low-Rank Adaptation [1.9261138876072244]
  Large Language Model (LLM) unlearning aims to remove targeted knowledge from a trained model.<n>We propose quantization-robust unlearning via low-rank adaptation (LoRA)<n>LoRA improves 4-bit utility by up to 7.93 points (NPO+GDR on BOOKS: 50.17 to 58.10) and yields higher 4-bit utility on NEWS for GA+GDR (40.06 to 44.82, increase of 4.76)
  arXiv  Detail & Related papers  (2026-02-13T18:01:40Z)
• RouteMoA: Dynamic Routing without Pre-Inference Boosts Efficient Mixture-of-Agents [91.0187958746262]
  RouteMoA is an efficient mixture-of-agents framework with dynamic routing.<n>It employs a lightweight scorer to perform initial screening by predicting coarse-grained performance from the query.<n>It refines these scores through lightweight self- and cross-assessment based on existing model outputs, providing posterior correction without additional inference.
  arXiv  Detail & Related papers  (2026-01-26T04:22:22Z)
• Generative Adapter: Contextualizing Language Models in Parameters with A Single Forward Pass [109.34230156454574]
  Large language models (LMs) are typically adapted to improve performance on new contexts. fine-tuning incurs significant training cost and prompting increases inference overhead. We introduce $GenerativeAdapter$, an effective and efficient adaptation method that directly maps new contexts to low-rank LM adapters.
  arXiv  Detail & Related papers  (2024-11-08T00:42:47Z)
• SuperCorrect: Advancing Small LLM Reasoning with Thought Template Distillation and Self-Correction [89.56181323849512]
  SuperCorrect is a novel two-stage framework that uses a large teacher model to supervise and correct both the reasoning and reflection processes of a smaller student model.<n>In the first stage, we extract hierarchical high-level and detailed thought templates from the teacher model to guide the student model in eliciting more fine-grained reasoning thoughts.<n>In the second stage, we introduce cross-model collaborative direct preference optimization (DPO) to enhance the self-correction abilities of the student model.
  arXiv  Detail & Related papers  (2024-10-11T17:25:52Z)
• Reset-free Reinforcement Learning with World Models [6.151562278670799]
  We propose model-based reset-free (MoReFree) agent forReinforcement learning.<n>MoReFree adapts two key mechanisms, exploration and policy learning, to handle reset-free tasks.<n>It exhibits superior data-efficiency across various reset-free tasks without access to environmental reward or demonstrations.
  arXiv  Detail & Related papers  (2024-08-19T08:56:00Z)
• GKT: A Novel Guidance-Based Knowledge Transfer Framework For Efficient Cloud-edge Collaboration LLM Deployment [74.40196814292426]
  We introduce a novel and intuitive Guidance-based Knowledge Transfer (GKT) framework. GKT uses a larger Large Language Models as a ''teacher'' to create guidance prompts, paired with a smaller ''student'' model to finalize responses. It achieves a maximum accuracy improvement of 14.18%, along with a 10.72 times speed-up on GSM8K and an accuracy improvement of 14.00 % along with a 7.73 times speed-up in CSQA.
  arXiv  Detail & Related papers  (2024-05-30T02:37:35Z)
• $\Delta$-Patching: A Framework for Rapid Adaptation of Pre-trained Convolutional Networks without Base Performance Loss [71.46601663956521]
  Models pre-trained on large-scale datasets are often fine-tuned to support newer tasks and datasets that arrive over time. We propose $Delta$-Patching for fine-tuning neural network models in an efficient manner, without the need to store model copies. Our experiments show that $Delta$-Networks outperform earlier model patching work while only requiring a fraction of parameters to be trained.
  arXiv  Detail & Related papers  (2023-03-26T16:39:44Z)
• CHAPTER: Exploiting Convolutional Neural Network Adapters for Self-supervised Speech Models [62.60723685118747]
  Self-supervised learning (SSL) is a powerful technique for learning representations from unlabeled data. We propose an efficient tuning method specifically designed for SSL speech model, by applying CNN adapters at the feature extractor. We empirically found that adding CNN to the feature extractor can help the adaptation on emotion and speaker tasks.
  arXiv  Detail & Related papers  (2022-12-01T08:50:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.