CEO & co-founder, Mobius Labs (exited to Dropbox). Making Dash multimodal. Trying to make computers see.
space-time continuum
Joined December 2008
- Tweets1.9K
- Following487
- Followers409
- Likes2.7K
appughar retweeted
If you’re at @EurIPSConf this week and looking for your next ML role, ping us. Few of us from @Dropbox including @mobicham @appughar will be there Happy to chat about roles in multimodal, search, and video-AI.
Looking for people with experience in : Building and scaling multimodal ML systems, Retrieval, ranking, and creative search problems, Turning prototypes into polished products
🌍 Remote (Germany or Denmark)
Apply here: jobs.dropbox.com/listing/738… (Please use the link, not DMs!)
After an incredible 7-year roller coaster of building @Mobius_Labs — a journey filled with both amazing excitement and moments of grief — I’m super thrilled to join @Dropbox along with my fellow Mobius team to bring multimodality to Dropbox Dash.
@Dropbox is acquiring our IP, and the Mobius core tech team is joining them. A huge milestone for us!
dropbox.tech/machine-learnin…
A big part of my PhD was on manifold optimization researchgate.net/publication…. Back then, training data were only in the 1000s. I never thought I’d hear about the Stiefel manifold again in the era of stochastic gradients—where massive datasets often make simpler, heuristic methods work just fine. Exciting to see these ideas resurface!
Efficient training of neural networks is difficult. Our second Connectionism post introduces Modular Manifolds, a theoretical step toward more stable and performant training by co-designing neural net optimizers with manifold constraints on weight matrices.
thinkingmachines.ai/blog/mod…
We explore a fundamental understanding of the geometry of neural network optimization.
appughar retweeted
GemLite now supports @vllm_project vllm V1, which brings up to 1.25x faster inference speed vs V0!
github.com/mobiusml/gemlite/…
What fascinates me about Miyazaki’s work is how much empathy it stirs. In Totoro, when Satsuki runs searching for her sister—worried, confused, a bit angry—it felt like I was sharing the experience with her. That, to me, is the core aesthetics behind the shot.
That said, with reasoning models and looping through it, it is not unthinkable ( imho quite plausible ) machine intelligence will have self awareness and a sense of empathy derived from their past. What is not clear is how these sense diverges from our experiences.
appughar retweeted
GemLite now supports bfloat16 on both Hopper and non-Hopper gpus 🫡
github.com/mobiusml/gemlite/…
Gemlite and HQQ comes to @vllm_project 🤘
New releases for hqq ( github.com/mobiusml/hqq/rele……) and gemlite ( github.com/mobiusml/gemlite/…… )!
The most exciting update is that we're bringing support to @vllm_project
My 2 cents on R1 momenttech.eu: Open source and innovation unlocks opportunities for European startups. Speed & innovation now matter most. Deploying models for a few 100€s was once unimaginable. Now we can. Eg: lnkd.in/e-dHQk5T.
tech.eu/2025/01/31/deepseek-…
DeepSeek is a pivotal moment for the open-source movement, bringing innovation & ideas back to the forefront-beyond just capital investment. Here we are, a small team in Berlin, refining R1 into smaller, more accurate variants, each training run costing around €10. Let's go! 🤘
We’ve improved DeepSeek R1 distilled models using logits distillation—delivering +4-14% gains on GSM8K while only spending $3-18 per training run! 🚀
Now available on Hugging Face-e—run them efficiently with HQQ! 🤗
🔗More details at: mobiusml.github.io/r1_redist…
When I saw this scene, I dismissed such a scenario as distant and just another case of anthropomorphizing the future. But with current reasoning models, such scenarios now seem quite plausible."
Sidenote: The acting in this scene is superlative!
youtube.com/watch?v=EYq0_qv7…
More double distilled blend of R1 models.
We are releasing more re-distilled R1 models. Now including a Qwen-7B with a GSM8K score of 83.4%, +4.55% higher than that of the original distilled model.
7B: huggingface.co/mobiuslabsgmb…
1.5B: huggingface.co/mobiuslabsgmb…
appughar retweeted
Our re-distilled @deepseek_ai R1 (1.5B) outperforms the original distilled model! Get it at huggingface.co/mobiuslabsgmb…. We’re distilling more models and look forward to releasing them soon!
appughar retweeted
i'm comically impressed that people are coping on deepseek by spewing bizarre conspiracy theories -- despite deepseek open-sourcing and writing some of the most detail oriented papers ever.
read. replicate. compete.
don't be salty, just makes you look incompetent.
appughar retweeted
Heard you like distillation, so we re-distilled the distilled model
Our re-distilled @deepseek_ai R1 (1.5B) outperforms the original distilled model! Get it at huggingface.co/mobiuslabsgmb…. We’re distilling more models and look forward to releasing them soon!