Brian Edwards bedwards

GLM-4.6 on Mac Studio: Complete Guide

The biggest, baddest GLM-4.6 running locally on Metal GPU

This guide covers everything from zero to serving GLM-4.6 via OpenAI-compatible API on your Mac Studio, plus uploading your merged model to Hugging Face.

Part 1: Local Setup (Nothing → Serving)

Portfolio projects

Crime Lab - web app with Shopify integration

Skills: Shopify, PostgreSQL, Node.js, React, HTML, CSS, JavaScript, GraphQL
Repo: https://github.com/bedwards/imaginary-crime-lab
Demo: https://bedwards.github.io/imaginary-crime-lab/
Bonus: CloudFlare Workers AI

Shopify app with post-purchase-ui extension

Skills: Shopify app and CLI, Node.js, React

Project Architecture Overview (Hobby / Portfolio App)

Overall Goals
- Full control and visibility (minimal platform magic)
- Stack combines static front end + edge logic + SaaS commerce + multiple data stores
- Transparent architecture where internals are visible in the UI
- No user auth; treat every visit as one global user

Claude Sonnet 4.5 (extended-thinking) - 94/100 Exceptional literary quality with natural consciousness flow. The details serve psychological purposes—the "mint green pencil case...too expensive but bought anyway" reveals character through gesture. The final paragraph's existential weight ("you could do everything correctly and still feel like you had somehow made a series of wrong turns") achieves genuine emotional complexity. Reads like accomplished literary fiction.
Claude Opus 4.1 (extended-thinking) - 89/100 Strong narrative consciousness with details that emerge naturally from attention: "emergency twenty-kroner note," the cutting board's "dark stains that wouldn't come out." Excellent use of temporal markers ("four thousand times, five thousand") creates rhythm. The Andersens' window observation adds psychological depth. Ends appropriately mid-scene without false closure.
mistral-large-123b - 83/100 Solid literary execution with good sensory grounding. The "not quite affect

Reading lists

Current

John Hersey - Hiroshima
China Miéville - The City & The City
W. G. Sebald - Austerlitz
David Graeber - Dawn of Everything
Susan Sontag - Illness as Metaphor
László Krasznahorkai - Satantango

Your 2025 Summer Reading Analysis

May 24 - September 23, 2025

📊 EXECUTIVE SUMMARY

Total Books: 15
Total Pages: 5,016
Reading Period: 121 days (17.3 weeks)

ML Framework Performance Comparison on Mac M1: Tabular Data Benchmark

Objective

Compare training performance of popular machine learning frameworks on Mac Studio M1 for tabular regression tasks, focusing on CPU vs GPU acceleration capabilities.

Raw timing results:

10k samples

	Familiarize yourself with this repository.

	We are creating a new album. You are acting as Rick Rubin.

	My two favorite albums and the inspirations for this album we are creating are:

	- Black Country, New Road - Ants from Up There
	- Songs: Ohia - The Magnolia Electric Co.

	The concept is knowing people deeply, the people who have been off and on through out our lives and/or are currently close to us. The lyrics are poetic and cryptic.

	codestral:22b
	0.20 seconds

	qwen3-coder:30b
	0.21 seconds

	gemma2:27b
	0.29 seconds

	command-r-plus:104b

	Note: Click the "Raw" button for a wider view.

	Sorted by Tok/sec (descending).

	🚀 Ollama LLM Benchmark Results
	┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━━━━━━┳━━━━━━┓
	┃ Model ┃ TTFT (s) ┃ TPOT (s) ┃ Tok/sec ┃ Params (B) ┃ Size (GB) ┃ VRAM (GB) ┃ Runs ┃
	┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━━━━━━╇━━━━━━┩
	│ tinyllama:1.1b │ 0.052 ±0.008 │ 0.0038 ±0.0001 │ 266.2 │ 1.1 │ 0.59 │ 0.71 │ 3 │
	├──────────────────────────────────────────────────┼───────────────┼────────────────┼─────────┼────────────┼───────────┼───────────┼──────┤