Reaching out from the HF SLM community!
Hey Pinkstack! I've been following your work since late 2024, back from the early PARM releases all the way to Fijik 2.0.
Just wanted to drop by and let you know there is a massive, thriving SLM and micro-model community here on Hugging Face right now (if you haven't heard of them already). Groups like SupraLabs, AtomixLabs (me!), AxiomicLabs, Glint-Research, and others are all actively building sub-1B edge models, running open leaderboards, and sharing datasets.
Your edge and reasoning research fits right into the ecosystem. If you ever want to benchmark together, share datasets, or join the community discussions, we'd love to have you in the loop!
P.S. I was reading through the Fijik 2.0 card and seeing it trained on a single 70W RTX 2000 Ada with Ademamix and 65K YARN scaling is legendary. Squeezing that level of agentic performance out of a single low-power card takes serious dedication.
As an independent builder who also focuses on resource-constrained training, I really respect your stack and philosophy. If you ever want to chat DMs, share dataset pipelines, or brainstorm compute setups, feel free to reach out anytime!
Hey @MultivexAI ! Thank you very much for your continued, prolonged support! ❤️ :3
it is much appreciated.
More is to come soon and more models are being trained! Next release will include 2-3 models hopefully (if no complications arise), using a different methodology.
Yes, I am aware of all of the small/micro llm (and similar) research groups popping up! They are quite awesome.
You're free to reach out via X/Twitter DMS to @Pinkstack_ (same as the one listed on the HF page)
Appreciate the reply! Really excited to see what you do with the new methodology on those 2-3 upcoming models.
I'll definitely drop you a DM on X (@Pinkstack_) shortly. Looking forward to connecting and chatting research!