maybe it's time to start to distill qwen 3.8 27b to a 9b model only to math and software development and run it locally with a rtx 4070 at q4 quantization
Ya know, the second it's #VMAs szn, I ask everyone's fave music vid – and Kathryn Newton came through with "Goodbyes" by Post Malone 🎬
📍: #TheDevilsMouth Premiere
what continues to be interesting as i explore the @wardenprotocol / halo ecosystem is the notion that anyone can serve the models
anyone that has either hardware or existing api access can become an operator and earn stables / compete on price + quality while users pay them for the inference they consume
prompts are encrypted by default and "memory" stays inside your browser rather than a central server, which i think we will continue to see become adopted as a norm for models that people want to input private data in to
halo seems less like another ai interface, and more like an open market forming underneath them with the price of models being influenced by the underlying supply and demand economics
will continue to follow the experiment of global permissionless service built on crypto rails, democratized for all