The paper reports on training and calibration experiments for three open models on the ARC‑AGI‑2 benchmark, using offline runs on four L4 GPUs for up to 12 hours. For the NVARC Qwen3‑4B grid model, a curated corpus of 12,000 executable rules and 95,993 tasks was built, and fine‑tuning on 1,920 real task records raised the semi‑private score from 28.96 to 33.06, with no synthetic‑task checkpoint outperforming the real‑task model. For the Qwen3.8‑27B reasoning model, fine‑tuning on 1,036 of its own reasoning traces and a specialized inference procedure increased the score from 26.81 to 44.03, while the Qwen3.8‑Flash‑Next mixture‑of‑experts model achieved a best score of 56.67 after expert pruning, routing statistics adaptation, and an FP8 indexer cache.
By Nguyen Minh Thang, Trinh Thi Phuong
UpgradeBench is a decision‑centric longitudinal benchmark that evaluates how fine‑tuned language‑model specialists should be handled when new base‑model releases occur. It covers four consecutive Qwen releases, a continuation checkpoint, six tasks, two model sizes, and OLMo checkpoints with known training lineage, and examines whether retraining, adapter transfer, or other recovery strategies improve specialist performance. The benchmark reveals that upgrade gains vary by task and release interval, that direct adapter copying is sensitive to pretraining distance, and that teacher relabeling can recover specialists without new annotations.
"whyItMatters":"The study provides actionable insights into the cost‑effective management of specialist models across model releases, showing how to balance retraining effort with performance gains."
By Ye Chen, Weining Zhang
The paper introduces RSOD, a framework for creating telecom‑specific language models that blend broad networking knowledge with specialized routing, broadband, and diagnostic capabilities. It generates over 212,000 synthetic training records from networking taxonomies, trains six frozen specialist LoRA teachers, and distills their behavior into a single deployable student adapter, Llama‑CalixCLM v1. The student achieves near‑teacher performance (within 0.05 micro‑score points) while maintaining a validated, auditable training‑to‑deployment path.
By Anil Srirangapatna Nagesh, Nitin Gupta, Anuj Taneja, Vivek Singh, Shane Eleniak, Theo Zhao, Moulidharan N, Shashank Keshetty
The paper presents a post‑training method for dynamically routed mixture‑of‑experts (MoE) models that integrates independently optimized specialists efficiently. It introduces calibrated multi‑teacher on‑policy distillation (C‑MOPD) with teacher‑wise loss normalization and clipping, combined with a post‑capacity bounded‑width dispatch, final‑route replay, and asynchronous rollout–training overlap. On the Turing‑20B‑A2B model, the approach achieves a high integration score (0.9988), retains most specialist gains, outperforms shared SFT and Qwen3‑8B on reasoning‑enabled benchmarks, and reduces consolidation time by 2.86×.
By Yue He, Jiahui Hu, Shaman Tang, Hua Zhou, Qiman Wu, Menglin Li, Yizhao Wang, Yuheng Zhang, Da Zhu, Zhou Wang, Anhua Liu, Hanlin Chen, Mingyuan Wang, Rui Xin, Yuhua Wei, Shuang Su, Siqi Liu, Kai Yang, Weichao Huang, Hang Zhang, Honggou Yang, Xianming Liu
arXiv:2606. 26671v1 Announce Type: new Abstract: Post-training alignment determines the reasoning and human preference following capabilities of large language models, yet most existing works withhold detailed data construction, filtering rules and training recipes, which hinders community reproducibility and lightweight model optimization.
By Qiaobo Hao, Yangqian Wu, Shunyi Wang, Zhongjian Zhang, Ziqun Li, Yayin He, Muqing Li, Chen Zhong
arXiv:2606. 07810v1 Announce Type: cross Abstract: Large language models (LLMs) are widely used as judges for evaluating model outputs, but their high cost, latency, and opacity limit scalability.
By Anish Laddha, Nitesh Pradhan, Gaurav Srivastava
arXiv:2607. 05904v1 Announce Type: new Abstract: Training a language model against its own reference-free judgments (the premise of self-rewarding, self-play, and LLM-as-a-judge pipelines) assumes a model's verdict on a shown answer tracks correctness.
By Chenyu Zhou
arXiv:2608.21382v1 Announce Type: new
Abstract: Multiple-choice benchmarks fix the questions and the correct answers, but not the harness: the order of the options, the wording of the prompt, and whe...
By V. S. Raghu Parupudi
The paper presents a unified evaluation of seven open reasoning language models across four benchmarks (ARC-Challenge, GSM8K, MATH levels 1–3, and TruthfulQA MC1) using a consistent 238-example subset and three prompting strategies (zero-shot, chain-of-thought, few-shot CoT). It reports not only accuracy but also Wilson confidence intervals, latency, VRAM usage, weighted aggregate performance, Pareto-efficient points, prompt-sensitivity, and compatibility diagnostics, revealing that Gemma-4-26B-A4B tops the weighted score while Gemma-4-E4B offers a strong practical trade-off. The study emphasizes that model rankings shift with prompting strategy and that deployment trade-offs remain crucial, advocating for a deployment-aware, multi-objective evaluation framework rather than a single-score leaderboard.
By Md Motaleb Hossen Manik, Ge Wang
arXiv:2504. 15610v4 Announce Type: replace Abstract: Fine-tuning a 7B language model for specialized advising is attractive in resource-constrained settings, but multi-epoch runs routinely exceed the wall-clock limits of the free-tier GPUs (Kaggle, Colab) such users rely on.
By Md Millat Hosen
The paper evaluates a manager‑worker scaffold that uses a shared filesystem workspace to orchestrate multi‑agent large language model (LLM) coding tasks without training or tuning. Across nine models—including five open‑weight and four closed‑weight systems—the scaffold yields statistically significant accuracy gains for some models (e.g., Qwen3.8‑27B, GPT‑5.6‑Luna, GPT‑5.6‑Terra, Kimi‑K3, Minimax‑M3) while producing null or negative effects for others (e.g., Qwen3.6‑35B). The study shows that the manager can triple token usage but still achieves higher accuracy at a fraction of the cost compared to larger single‑pass models, with key mechanisms identified as context management and problem decomposition.
By Victor Gao (Sang Won), Vida Khosrowshahi (Sang Won), Ali Khosrowshahi (Sang Won), Xihao Sun (Sang Won), Juhyun Lee (Sang Won), Simon (Sang Won), Lee
arXiv:2606. 03650v1 Announce Type: cross Abstract: Choosing or ranking language models for a specific application is hardest when no task-specific labeled data exists, and standard public benchmarks cannot be trusted, their items having likely leaked into pretraining, so scores reflect memorization rather than fitness.
By Alexander Apartsin, Yehudit Aperstein