Tag
Microsoft Mage Team presents Mage-VL-Cowork, a full training recipe (data synthesis from evolved harnesses, multi-stage SFT, and online hybrid RL) for learning hybrid computer use, with the 4B model achieving strong results on OSWorld-Verified relative to its size.
A recent paper worth noting proposes a “unified training” algorithm that integrates pre-training, mid-training, and post-training into a single algorithmic workflow, simplifying infrastructure and providing a framework for better scientific research.
This paper introduces Steerling-8B, a diffusion language model trained with interpretability as a constraint, showing that interpretability improves with scale and enabling concept steering without retraining.
The author shares techniques that improved training pipeline performance from 36 to 47 steps per minute.
A demonstration of training a vision-language model to play Snake using the FeynRL framework, illustrating the full training pipeline in an accessible manner.
MiMo-V2.5-Pro-UltraSpeed is an ultra-fast large model training pipeline.
The official Hugging Face team fully open-sourced and reproduced the entire training pipeline of DeepSeek-R1 (Open-R1 project), including data, training, and evaluation. It has received 26.4k stars on GitHub, providing a reproducible textbook for training reasoning models for the industry.
A concise one-page cheat sheet covering PyTorch tensors, models, matrix multiplication, and autograd for deep learning training pipelines.
This paper proposes the LLM-as-Environment-Engineer framework, where a policy model analyzes failures to automatically redesign the training environment for reinforcement learning, and introduces MAPF-FrozenLake as a controllable testbed. The framework, using Qwen3-4B, outperforms larger models like GPT and Gemini, showing that policy learning improves the model's ability to diagnose weaknesses.
This paper presents a scalable open-source pipeline using NVIDIA NeMo for training and inference of Video Foundation Models, addressing challenges in generating high-quality videos with accelerated dataset curation and parallelized training.