Alim Gumran (@gumr4n) 's Twitter Profile
Alim Gumran

@gumr4n

Incoming MSCS @ ENS Lyon

ID: 1461318445400797186

linkhttp://gumran.github.io calendar_today18-11-2021 13:00:57

4 Tweet

4 Followers

74 Following

Charlie O'Neill (@charles0neill) 's Twitter Profile Photo

1/🧵 Thrilled to announce our ICML‑accepted work (done in my undergrad days with @david_klindt and Alim Gumran) that questions the linear‑only dogma in sparse autoencoders👇

Alim Gumran (@gumr4n) 's Twitter Profile Photo

As an exercise, I post-trained the original GPT-2 models on the SFT and DPO datasets from Ai2's best release so far (huge fan of their open-source work 🙏): allenai.org/blog/olmo2-32B. 📎 Models: huggingface.co/gumran 📎 Code: github.com/gumran/post_tr…

As an exercise, I post-trained the original GPT-2 models on the SFT and DPO datasets from <a href="/allen_ai/">Ai2</a>'s best release so far (huge fan of their open-source work 🙏): allenai.org/blog/olmo2-32B.

📎 Models: huggingface.co/gumran
📎 Code: github.com/gumran/post_tr…
Alim Gumran (@gumr4n) 's Twitter Profile Photo

Some of the latest vision-language-action (VLA) models are awesome in how they bring together diverse yet relevant fields of AI. Namely, VLMs, diffusion models (or flow matching), and soon maybe RL.