John Records (@JohnRecords) profile picture

John Records

@JohnRecords
Joined September 2025
  • Tweets79
  • Following330
  • Followers12
3Media
Photo by John Records (@JohnRecords)Photo by John Records (@JohnRecords)Photo by John Records (@JohnRecords)
Inspired by @emollick’s Portals, I used his repo and Fable to build Portals for Iain M. Banks Culture novels: sonic-sphinx-9qsd.here.now #theculture #iainmbanks
Culture Portals

Culture Portals

An endless, hands-off flight through generated places from Iain M. Banks's Culture novels, each reached through a portal in the last.

sonic-sphinx-9qsd.here.now
2
2
1
21
18,895
Which inference engine for local Mac Hermes? rapid-mlx benchmarked best on M1 Max 64 gb. YMMV. bright-lotus-8q5y.here.now
MLX Inference Engine Performance Benchmarks

MLX Inference Engine Performance Benchmarks

A comparative performance analysis of four MLX inference engines—rapid-mlx, mlx-lm, omlx, and ollama—on Apple Silicon M1 Max hardware.

bright-lotus-8q5y.here.now
83
Replying to @leopardracer
This is llama.cpp. The underlying inference engine of many popular AI engines. For example Lmstudio and ollama etc. The parameter that allows it is: --mmap. So everything is not loaded in vRAM for MoE models. Hence you can run models larger than your vRAM. Downside is:
2
1
26
6,420
Load more