Log inSign up
Ji Lin
128 posts
user avatar
Ji Lin
@jilin_14
Research @Meta Superintelligence Lab | Prev: Research @OpenAI; PhD @MIT
San Francisco, CA
linji.me
Joined August 2012
1,005
Following
6,034
Followers
  • user avatar
    Ji Lin
    @jilin_14
    Apr 16, 2025
    Exciting to share what i've been working on in the past few months! o3 and o4-mini are our first reasoning models with full tool support, including python, search, imagegen, etc. it also comes with the best VISUAL reasoning performance up-to-date!
    user avatar
    OpenAI
    @OpenAI
    Apr 16, 2025
    Introducing OpenAI o3 and o4-mini—our smartest and most capable models to date. For the first time, our reasoning models can agentically use and combine every tool within ChatGPT, including web search, Python, image analysis, file interpretation, and image generation.
    Image
    00:00
    37K
  • user avatar
    Ji Lin
    @jilin_14
    Jul 24, 2023
    - We present TinyChat, an efficient, lightweight, Python-native serving framework for 4-bit LLMs by AWQ. It delivers 2.3x generation speed up on RTX4090. Check it out here: github.com/mit-han-lab/ll… - I am attending ICML to present SmoothQuant this week. DM if you wanna chat!
    Image
    00:00
    37K
  • user avatar
    Ji Lin
    @jilin_14
    May 14, 2024
    Glad to learn AWQ, one of the final work during my PhD study, was selected for the best paper award at MLSys’24! Congrats to the team: @jmtang42 @haotiant1998 @Shang_mit @songhan_mit and more
    Image
    21K
  • user avatar
    Ji Lin
    @jilin_14
    Oct 19, 2023
    Excited to see SmoothQuant and AWQ being used in the TensorRT-LLM release today. Great work from the NVIDIA team!
    Image
    GitHub - NVIDIA/TensorRT-LLM: TensorRT LLM provides users with an easy-to-use Python API to define...
    From github.com
    21K
  • user avatar
    Ji Lin
    @jilin_14
    Jun 2, 2023
    SmoothQuant is good for W8A8 LLM quantization, what about low-bit weight-only quantization (e.g., W4A16)? We present Activation-aware Weight Quantization (AWQ) for LLM compression and acceleration: github.com/mit-han-lab/ll… 🧵
    Image
    26K
  • user avatar
    Ji Lin
    @jilin_14
    Mar 8, 2021
    Try out the demo and Colab of Anycost GAN: github.com/mit-han-lab/an…. Our method provides consistent outputs at various computational budgets, paving the way for interactive image synthesis and editing. (w/ @rzhang88, Frieder Ganz, @SongHan_MIT, @junyanz89) (1/2)
    Image
    GIF
  • user avatar
    Ji Lin
    @jilin_14
    Jun 14, 2024
    Going to #CVPR2024 next week in Seattle (cannot believe it has been 7 years since last time)! DM me if you want to talk about multimodal, LLM, or anything 😃
    15K
  • user avatar
    Ji Lin
    @jilin_14
    Jul 21, 2023
    We extended our AWQ support for more LLM architectures, including Llama-2, MPT, Falcon, and BLOOM. Checkout our repo if you are interested in efficient 4-bit LLM inference:
    user avatar
    Ji Lin
    @jilin_14
    Jun 2, 2023
    SmoothQuant is good for W8A8 LLM quantization, what about low-bit weight-only quantization (e.g., W4A16)? We present Activation-aware Weight Quantization (AWQ) for LLM compression and acceleration: github.com/mit-han-lab/ll… 🧵
    Image
    Image
    GitHub - mit-han-lab/llm-awq: [MLSys 2024 Best Paper Award] AWQ: Activation-aware Weight Quantiza...
    From github.com
    11K
  • user avatar
    Ji Lin
    @jilin_14
    Aug 31, 2022
    Grateful to receive the @Qualcomm Innovation Fellowship 2022 (w/ @LigengZhu)! Many thanks to my advisor @SongHan_MIT and all collaborators!
    Image
    2022 Qualcomm Innovation Fellowship for North America | US QIF 2022 |
    From qualcomm.com
  • user avatar
    Ji Lin
    @jilin_14
    Sep 12, 2024
    Feeling the AGI! 🍓
    user avatar
    OpenAI
    @OpenAI
    Sep 12, 2024
    We're releasing a preview of OpenAI o1—a new series of AI models designed to spend more time thinking before they respond. These models can reason through complex tasks and solve harder problems than previous models in science, coding, and math. openai.com/index/introduc…
    8.7K
  • user avatar
    Ji Lin
    @jilin_14
    Aug 7, 2025
    🥹
    user avatar
    Yonglong Tian
    @YonglongT
    Aug 7, 2025
    GPT-5 dropped! For *multimodal*, the nice thing is it will use tools way more efficient than o3 (much better than the rendered acc numbers here), making it both better and faster. @jilin_14, efforts baked in.
    Image
    10K
  • user avatar
    Ji Lin
    @jilin_14
    Apr 16, 2025
    Find more visual reasoning samples in this blog. Great team work with @jhyuxm @mckbrando @ZhangZhshuai @bowenc0221 Jamie @dmed256 @hthu2017 and more! Easter egg: I put multiple photos from my Instagram library into the blog 😃 openai.com/index/thinking…
    Image
    4.2K
  • user avatar
    Ji Lin
    @jilin_14
    Dec 17, 2024
    Making intelligence cheaper and more accessible! Fun experience to do some efficiency-related stuff at OpenAI, after PhD.
    user avatar
    Shuchao Bi
    @shuchaobi
    Dec 17, 2024
    10x cheaper realtime voice API. The internet went from text only to multimodal over the last 25 years: blogs + Google → instagram → short-form videos (YouTube Shorts, TikTok). Think about how many human hours are spent on writing / reading text vs talking or watching videos.
    4.7K
  • user avatar
    Ji Lin
    @jilin_14
    May 13, 2024
    GPT is now native multi-modal! Many exciting demos in the blog 👇
    Image
    Hello GPT-4o
    From openai.com
    4K

New to X?

Sign up now to get your own personalized timeline!

Create account

By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.

Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
Don't miss what's happening
People on X are the first to know.
Log inSign up