Log inSign up
Shangshang Wang profile banner
user avatar

Shangshang Wang

@UpupWang
Phd @CSatUSC | Ex Intern @Alibaba_Qwen, @bespokelabsai | RL, LLMs, Agents
Los Angeles
shangshang-wang.github.io
Joined December 2024
236
Following
612
Followers
RepliesMedia
  • Pinned
    user avatar
    Shangshang Wang
    @UpupWang
    Apr 23, 2025
    😋 Want strong LLM reasoning without breaking the bank? We explored just how cost-effectively RL can enhance reasoning using LoRA! [1/9] Introducing Tina: A family of tiny reasoning models with strong performance at low cost, providing an accessible testbed for RL reasoning. 🧵
    Image
  • user avatar
    Shangshang Wang
    @UpupWang
    Dec 3, 2025
    Many thanks to @thinkymachines for the Tinker grant! Our plan: we know that LoRA matches full-param RL by exploiting a tiny subspace—now we’re testing whether exploration can push it further. Stay tuned.
    Image
  • user avatar
    Shangshang Wang
    @UpupWang
    Oct 8, 2025
    We now know that LoRA can match full-parameter RL training (from x.com/thinkymachines… and our Tina paper arxiv.org/abs/2504.15777), but what about DoRA, QLoRA, and more? We are releasing a clean LoRA-for-RL repo to explore them all. github.com/shangshang-wan…
    Image
    Image
    user avatar
    Thinking Machines
    @thinkymachines
    Sep 29, 2025
    LoRA makes fine-tuning more accessible, but it's unclear how it compares to full fine-tuning. We find that the performance often matches closely---more often than you might expect. In our latest Connectionism post, we share our experimental results and recommendations for LoRA.
  • user avatar
    Shangshang Wang
    @UpupWang
    Jun 12, 2025
    Sparse autoencoders (SAEs) can be used to elicit strong reasoning abilities with remarkable efficiency. Using only 1 hour of training at $2 cost without any reasoning traces, we find a way to train 1.5B models via SAEs to score 43.33% Pass@1 on AIME24 and 90% Pass@1 on AMC23.
    Image