Hugging Face just dropped a comprehensive blog post on Vision-Language Models (VLMs) — covering architecture, training, benchmarks, and useful resources. A must-read if you're navigating the multimodal frontier! https://
huggingface.co/blog/vlms-2025
@jiqizhixin
-
Hugging Face Releases Comprehensive Vision-Language Models Guide
By
–
-
MLE-Dojo Benchmark Evaluates Frontier LLMs on ML Engineering
By
–
MLE-Dojo: A new benchmark to evaluate LLM agents on real Machine Learning Engineering tasks. Its key innovation? An interactive environment that allows agents to experiment, debug, and refine solutions via structured feedback loops. Here’s how 8 frontier LLMs perform
-
Embodiment Scaling Laws Improve Robot Locomotion Generalization
By
–
Towards Embodiment Scaling Laws in Robot Locomotion This study investigates whether scaling the diversity of training embodiments leads to better generalization in robotic agents. Key contributions: Built a dataset of ~1,000 procedurally generated robot bodies
-
Neuro-Symbolic Concept-Centric Agents for General Intelligence
By
–
Concept-Centric Agents: A Neuro-Symbolic Shift in General Intelligence MIT and Stanford researchers proposes a concept-centric paradigm for building generalist agents capable of continual learning and flexible reasoning. Core idea: Neuro-symbolic concepts (e.g.,
-
Qwen3 Quantized Models Now Available for Local Deployment
By
–
Qwen3 goes local! Quantized versions of Qwen3 are now live — deploy via Ollama, LM Studio, SGLang, or vLLM, with support for GGUF, AWQ, and GPTQ formats. Get started:
Hugging Face：
https://
huggingface.co/collections/Qw
en/qwen3-67dd247413f0e2e4f653967f
…
ModelScope：
https://
modelscope.cn/collections/Qw
en3-9743180bdc6b48
… -
Microsoft MARK: Memory-Augmented LLMs Without Retraining
By
–
Microsoft’s MARK: Memory-Augmented LLMs Microsoft introduces MARK (Memory-Augmented Refinement of Knowledge), a new framework that lets LLMs evolve with domain knowledge without retraining. Inspired by Society of Mind, MARK uses specialized memory agents: Residual
-
Jim Fan on NVIDIA’s Roadmap for Embodied AI Robots
By
–
How do we develop general-purpose robots?
— 机器之心 JIQIZHIXIN (@jiqizhixin) 10 mai 2025
A must-watch talk by Jim Fan, NVIDIA’s Director of Robotics and Distinguished Scientist, explores this pressing question.
The Physical Turing Test: Jim Fan on Nvidia's Roadmap for Embodied AI
Link: https://t.co/YGaQgl6hun pic.twitter.com/G4mfavMreEHow do we develop general-purpose robots?
A must-watch talk by Jim Fan, NVIDIA’s Director of Robotics and Distinguished Scientist, explores this pressing question. The Physical Turing Test: Jim Fan on Nvidia's Roadmap for Embodied AI
Link: https://
youtube.com/watch?v=_2NijX
qBESI
… -
Flow-GRPO: Online RL Integration into Flow Matching Models
By
–
Flow-GRPO: Training Flow Matching Models via Online RL This work introduces Flow-GRPO, the first method to embed online reinforcement learning into flow matching diffusion models. Two innovations drive it: 1. ODE → SDE conversion for richer, exploration-ready
-
Toxic Data Improves LLM Post-Training Control and Separability
By
–
"Bad Data, Good Models?" A surprising take on LLM pretraining. This paper flips the script: pretraining with more toxic data can actually improve post-training control. Using Olmo-1B variants, the authors show that toxicity becomes more linearly separable—making
-
Elastic Reasoning: Smart CoT Control for Large Models
By
–
Controlling CoT in large reasoning models just got smarter. Elastic Reasoning decouples thinking and solution phases—allocating separate budgets to each. This means more reliable outputs under strict compute/token constraints . Trained with a lightweight, budget-aware