Fetching from the wire…
01
02
03
04
05
06
07
08
Infra2026-09-26 · source-backed
arXiv 2609.29032 presents a Swift/MLX runtime keeping quantized expert weights in storage with a byte-budgeted subset in memory. On five recorded 128-token workloads, a 512 MiB LRU cache got 0.00% hits, seeded random eviction at the same budget got 18.80%, and 576 MiB LRU got 38.58%. The apparent memory cliff is an interaction between policy and workload, not a hardware limit. iOS 27 process peaks were 1.87-2.73 GiB. The author reports a thermal stop and says the phone's token sequences disagreed with a resident-Python reference on all five cases, which is an honest disclosure most papers wouldn't make.
Each link below shares sources, entities, or timing with this story.
AlexsJones/llmfit released v1.1.10 today, adding RamaLama runtime discovery to its MCP server, the Qwen3.8 model family and MiniMax M3 vision capability exposure (GitHub). It also merged 32 MLX benchmark results on an Apple M4 Pro, the project's first MLX entries, giving an ap...
MIT, Swift and MLX, created August 28, 220 points on Hacker News (GitHub). It runs Qwen3.8-Flash-Next, 103.8GB across 24 files at 4-bit, by loading only the 3.8GB dense trunk into RAM (about 2 second startup) and reading routed experts with pread into a fixed pool of cache slo...
For about a year, "run your agent locally" meant accepting a model that couldn't reliably call a tool twice in a row. That excuse is gone. Meta Superintelligence Labs published Muse Glimmer today: a 29.6B dense causal transformer, 52 layers, 6,656 hidden dim, with a ~1.8B ViT-...
Blaizzy/nativ (1,163 stars, Swift, MIT, macOS 26+) comes from the mlx-vlm author and bundles that server into a SwiftUI app that discovers MLX models already in your HF cache. It exposes OpenAI-compatible chat, Responses, image, audio and model endpoints plus Anthropic Message...
Alibaba released Qwen3.6-27B on April 22. Dense architecture. Open weights. 77.2% on SWE-bench Verified, within 3.7 points of Claude Opus 4.6. On SkillsBench, it scores 48.2% versus its own 397B MoE predecessor's 30.0%. That's a 77% improvement with 14.8x fewer parameters. Let...
Lily is Apache 2.0 inside pplx-garden, a small Metal inference server built for Qwen3.6-35B-A3B converted to MLX affine 4-bit, exposing a minimal OpenAI-compatible chat API with greedy decoding. The README explicitly rules out dense and smaller Qwen checkpoints, BF16, GGUF, AW...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.