@jundotkim Nice! I have quite a few improvements on a private branch, running qwen3.8-next at around 12t/s on my air. I’ll make a new release soon and put together some direct integration if you like. I’ll get back to you.
Qwen3.8-27B can now be run locally! ✨
Run on 17GB RAM via Unsloth Dynamic GGUFs.
Qwen3.8-27B is by far the strongest model for its size. We also uploaded NVFP4 quants.
GGUF: huggingface.co/unsloth/Qwen3.…
Guide: unsloth.ai/docs/models/qw…
Introducing GLM-5.3: Built to Code. Ready for Cyber Defense.
- Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model
- A major leap in cybersecurity, setting a new standard among open models
Tech Blog: z.ai/blog/glm-5.3
Tested on a base M4 Air, 32 GB: Laguna-S 2.1 (oQ2e) at 9 tok/s decode short-context, 4–6 tok/s on opencode at 8k+ context, prefill ~50 tok/s. Speed is SSD-bound, so better chips/SSDs should scale it well, especially for larger models.
Would love to know if someone would like to test larger models on better hardware!
If you like, you can run local MoE models bigger than your RAM on Apple Silicon, streaming the routed experts off SSD with my latest project :)
streamlx: github.com/srcterm/stream…
I Wrote a New Book!!!
Optimization: A Bootcamp for Machine Learning, Inverse Problems, and Control
Pre-Order Now (July 31)
amazon.com/Optimization-B…
Coming Soon:
* Free PDF on website
* YouTube Videos for entire book
* Python code on GitHub
Dear old people, waking up at 6am, tending to a garden, eating dinner at 5pm, reading books, and going to bed at 9:30pm feels amazing.
I was wrong. You were right.
First more technical post documenting torchFW-H, a differentiable GPU acoustics solver in PyTorch. I have been working on this in my spare time so progress is snail-ish.
> This is part 1, it covers some of the early decisions like point clouds over panels, k-NN interpolation from CFD grids, and the reality of memory plumbing on GPU
hubbardian.io/posts/quest__s…
6K Followers 6K FollowingBuilding Ze German HybridClaw🇪🇺 Fo shur loves building things and training LLMs • Serial Founder @DataLion_EN @HybridAI_One • PhD @LMU_Muenchen
7K Followers 523 FollowingSystem cards, risk reports, and misc safety takes at Anthropic; math; puzzles; spaced repetition. Writes with too many caveats for Twitter.
13K Followers 244 FollowingDeep-Tech Semiconductor Analyst, MSEE @JohnsHopkins. Expert-driven system architecture deep-dives of AI datacenter hardware at https://t.co/atebRcUn6C
2K Followers 809 FollowingI build OpenSource things, local LLM engines & UI’s. Created https://t.co/qxuPXH75o1 for Desktop and iOS - https://t.co/nWiKjj7pAt
4K Followers 189 FollowingOpen-Source ML Engineer at 🤗 Hugging Face | Creator of oMLX | I build the tools I wish existed for my Mac, then open-source them. [email protected]
5K Followers 1K FollowingFounder/CEO of Lna-Lab K.K. Former publishing exec building local LLMs, Turn every bit of compute you own into golden tokens.🤖
581 Followers 26 FollowingCanada's open-weight model lab. Training, quantizing, and shipping sovereign
models on B300 silicon. Built in Victoria, BC. https://t.co/a77zsloLBL
220K Followers 0 FollowingOG of: P and T in ChatGPT, recursive self-improvement, neural distillation, GAN/World Model, deepest learning. Co-authored most-cited AI paper of 20th century