Skip to content

Removed local inference category from the blog #15

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
allexistence merged 1 commit into from
Sep 10, 2026
Merged

Removed local inference category from the blog #15

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 3 additions & 5 deletions _posts/2026-09-10/2026-09-10-muse-glimmer-m4-air-poc.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
---
title: "Running Meta's Muse Glimmer 30B on a 24 GB MacBook Air M4"
date: 2026-09-10 10:00:00 +0800
categories: [AI/ML, Local Inference]
categories: [AI/ML]
tags: [muse-glimmer, llama.cpp, apple-silicon, benchmarking, m4]
---

Expand All @@ -12,8 +12,7 @@ on my own machine — a MacBook Air M4 with 24 GB of unified memory. No cloud
GPU, no rented compute, just the laptop I already have.

This is a first attempt at local LLM benchmarking for me, so I kept the
scope small: get it loaded, and measure three things well rather than
measure ten things poorly.
scope small: get it loaded, and measure three things well.

![Terminal showing the model load and server startup](/assets/img/posts/muse-glimmer-m4-air/01-server-startup.png)
_llama-server loading the model and coming up on port 8080, n_ctx_slot = 8192_
Expand Down Expand Up @@ -218,5 +217,4 @@ n_tokens = 5495, t = 115.45 s / 47.60 tokens per second
see it; four runs may just not be enough to trigger it.

This was a first pass at benchmarking a local model, and I'm sure there's
room to tighten the methodology. Vision projector and speculative decoding
tests are next.
room to tighten the methodology.
Loading