The main findings are that diverse, high-quality SFT establishes a strong capability floor and difficulty filtering keeps RL prompts within a productive learning range, and reward reliability provides a practical principle for ordering stages.
Abstract
Rufus-Air is an open and reproducible post-training recipe on GLM-4.5-Air-Base (106B-A12B), organized as a serial pipeline of eight stages: SFT, Reasoning RL, Coding RL, Instruction-Following RL, General Agent, Coding Agent, Search Agent, and RLHF. We document the data, reward design, infrastructure, stage order, and stagewise results needed to reproduce the recipe. Stages progress from basic to advanced capabilities and from hard, verifiable rewards to softer judge-based signals. Training builds on open-source components and public data, much of it used as released, without new human annotation or an in-house distillation teacher. Our main findings are that (i) diverse, high-quality SFT establishes a strong capability floor; (ii) difficulty filtering keeps RL prompts within a productive learning range; (iii) reward reliability provides a practical principle for ordering stages; and (iv) infrastructure and engineering choices are part of the recipe, not just an implementation detail. Rufus-Air improves over the official GLM-4.5-Air post-trained release and is competitive with similarly sized open models.
This work characterize OPD across nine student-teacher pairs spanning 2x to 53x parameter ratios and shows that OPD effectiveness depends on student-teacher compatibility rather than teacher scale alone, and suggests that each post-training stage should be chosen not only for the capability it adds, but for the learnin...
Emre Can Acikgoz, Yang Li, Z. Liu et al.· 0 citations
Miles v0.1 is presented, a full-stack, production-ready system for frontier post-training reinforcement-learning training, which supports LoRA RL, on-policy distillation, supervised fine-tuning, and true-on-policy rollout-training alignment, and extends the same architecture to diffusion models.
Decision models such as Jev answer questions with probabilities, which are only useful if they are calibrated. Open-source reproductions rely on supervised fine-tuning plus temperature scaling, while reinforcement learning from verifiable rewards (RLVR) makes reasoning models overconfident. We present a working impleme...
Many useful language-model tasks cannot be evaluated by exact outcome verification. Rubric-based reinforcement learning (RL) addresses this issue by scoring open-ended responses against explicit criteria. However, because the reward is assigned after the complete response, the training signal does not directly identify...
Xinpeng Wang, Wei-Han Shi, Yu-Chia Chen et al.· 1 citation
Gnani-Evon-v3.3-30B-A3B is a long-context hybrid Mamba-Transformer mixture-of-experts model for reasoning, instruction following, and tool use in English and eight Indian languages. Starting from NVIDIA's Nemotron-3-Nano-30B-A3B checkpoint, it retains the 52-layer backbone and changes the vocabulary, training data, and...
An open midtraining recipe is presented that primes base models for agentic post-training, and its effectiveness on two Nemotron models of different sizes is shown, improving robustness across agent harnesses on both models.
Soumye Singhal, Joseph Jennings, Ka-Shun Shum et al.· 0 citations
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.
Machine-learning networks can learn without detailed instructions from a designer. Those that do may offer insight into how the brain reorganizes itself during learning.
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.