Liang Chen
@liangchen5518
Followers
3K
Following
208
Media
46
Statuses
255
Phd student in Peking University. I worked at Moonshot AI, Alibaba Qwen and Microsoft Research Asia.
Beijing
Joined February 2022
RT @Zefan_Cai: Really appreciate the support for VisualToolAgent (VisTA): our new RL-based framework for dynamic tool selection in visual r….
0
9
0
RT @deepseek_ai: 🚀 DeepSeek-R1-0528 is here!. 🔹 Improved benchmark performance.🔹 Enhanced front-end capabilities.🔹 Reduced hallucinations.🔹….
0
2K
0
RT @_akhaliq: G1. Bootstrapping Perception and Reasoning Abilities of Vision-Language Model via Reinforcement Learning .
0
17
0
Check more information in the paper and reproduce the amazing results by yourself!. Paper: Fully open source at:
github.com
G1: Bootstrapping Perception and Reasoning Abilities of Vision-Language Model via Reinforcement Learning - chenllliang/G1
0
0
2
G1: By introducing perception-enhanced cold start, the model can exceed the teacher model on all games during RL, preventing from the common problems such as sparse reward and inaccurate reward credit during training G0 models.
1
0
2
G0: Directly training VLMs on VLM-Gym yields diverse result for different games. We found that the VLM can learn special perception and reasoning patterns without supervison.
1
0
2
We first introduce VLM-Gym, a curated environment designed for scalable Reinforcement Learning and algorithm design of VLMs within interactive games.
1
0
1
🚨 Thrilled to announce G1, our latest research in Reinforcement Learning for Vision-Language Model through visual games! . We discover that perception & reasoning can mutually bootstrap during RL !. Fully open source at: Paper:
1
3
16
RT @Kimi_Moonshot: 🚀 Meet Kimi-VL and Kimi-VL-Thinking! 🌟 Our latest open source lightweight yet powerful Vision-Language Model with reason….
0
212
0
RT @bdsqlsz: T5 and Clip is dead. I talked to others last week that in the future, LLM will definitely replace the ordinary textencoder. ….
0
26
0
RT @OedoSoldier: @teortaxesTex @kalomaze I just read the DreamEngine paper and I think it's the technology stack behind the GPT4o (maybe Ge….
0
1
0
DreamEngine (VLM + Diffusion) might be the key behind GPT-4o's image generation. Want to see how it works? Check it out!.
🔥 DreamEngine revolutionizes image generation with its text-guided object fusion capabilities! . The demo and code for Text Guided Object Fustion are released! Let's unlock the Imaginations! . Run it locally now in: Paper:
0
2
11
RT @TsingYoga: Check out Agent TARS (preview). It's fully open-sourced and dev-friendly. Just try it~. https://t.co….
0
12
0
It is interesting and also little frustrating that the SOTA VLMs are still so bad at simple counting problem. I use a figure from my previous paper and asked Grok/Qwen Max/Claude 3.7/GPT4o. None of them gave correct answer (8).
0
5
14
RT @Alibaba_Qwen: Today, we release QwQ-32B, our new reasoning model with only 32 billion parameters that rivals cutting-edge reasoning mod….
0
2K
0
RT @ManusAI_HQ: Introducing Manus: the first general AI agent. Try Manus today and see the future of human-machine collaboration: https://….
0
1K
0
🔥 DreamEngine revolutionizes image generation with its text-guided object fusion capabilities! . The demo and code for Text Guided Object Fustion are released! Let's unlock the Imaginations! . Run it locally now in: Paper:
0
15
42