akg_67

Interesting read, your experience gave me some ideas to reduce dependency on Apple ecosystem. A few questions:

1. "Don't use safari" Why?

2. Is there an alternative to "Find My" that I can use with AirTag?

Some of the things, I have done since this experience:

- Exported all my passwords from Apple Password Manager to BitWarden. Also, now I better understand why my wife was keeping her passwords in a notebook (no service can be trusted).

- Looking at ways to export Contacts and Notes and sync.

- Keeping Dropbox storage subscription instead of considering migration to iCloud storage.

- Dropped the idea of upgrading my iPhone13 and wife's MacBook (light user) to Apple devices.

Yes, learnt this the hard way. Now, I am trying to figure out how to spread out to not get stuck again in a walled garden.

Context compaction? I noticed llm seem to forget partially or completely the original task when context compaction happens. The problem is more serious with local llm with low context size.

The above tests were done with 24k context window. Testing was mostly driven by ChatGPT analyzing oMLX server logs and suggesting changes.

Finally, I settled on Qwen3.6-35B-A3B-4bit with 32,768 context window and 16,384 max tokens.

---

Additional results from Qwen3.6-35B-A3B-4bit (Can't edit previous comment)

Qwen3.6-35B-A3B-4bit, 329.7 PP, 41.3 TG

What system? If on *M1 Max 32GB* or weaker, I will be interested in learning more about your setup.

Recent performance data on my M1 Max 32GB MacBook using oMLX. I have been working on identifying suitable model and config for my use case and system. Using a refactor and suggest improvements prompt for a specific Django code block using VSCode Cline extension.

---

Qwen3.8-27B-4bit, Prompt Processing (PP) 66.3 tok/s, Token Generation (TG) 11.8 tok/s

Ornith-1.5-35B-A3B-MLX-4bit, PP 379.7, TG 45.8

Ornith-1.5-35B-A3B-MLX-4bit, PP 381.5, TG 46.4

Qwen3.6-35B-A3B-mxfp4, PP 389.6, TG 47.6

Qwen3.6-35B-A3B-OptiQ-4bit, PP 342.6, TG 44.4

---

Qwen3.8-27B-4bit generally runs out of output token before completing the task though excellent partial results.

Ornith-1.5-35B-A3B-MLX-4bit seems to get in the loop often specially with tool calls.

Qwen3.6-35B-A3B-mxfp4 seems to be optimal with speed and quality output.

I am going to test Qwen3.6-35B-A3B-4bit soon with same code block just to check my intuition that any derivatives don't seem to perform better than the originals.

RAM 32GB, tokens are less than 20, may be about 12-15. Main issue is context window, I have set to 64k but practically beyond 32k, too much swap and slow down.

Macs fan control set to full blast and lifting the Mac above table for air circulation across bottom.

I use M1 Max with qwen3.8 27b mlx. The gpu temp can reach easily to 95°C as fan doesn’t kick in automatically until 90°C. With Macs fan control at full blast and MBP off table, temp usually hover around 85°C.

Are you using MLX version on M5 Max?

qwen3.8:27b-mlx

Though almost a year old (may be too long in LLM age), just sharing a meta-article on LLM applications in different areas of finance.

Large Language Models in equity markets: applications, techniques, and insights, Frontiers in Artificial Intelligence, A. Jadhav and V. Mirza, 27 August 2025, DOI: 10.3389/frai.2025.1608365.

Personally, I am using LLM, mostly for analyzing and shortlisting companies for in-depth analysis, for investing in Japanese equity market for over a year with relatively decent results.

Andy Rubin, chief executive of Genki Robotics, told Nikkei Asia that the company plans to open up its operating system to outside developers, allowing them to create and sell new robot behaviors through an app-store-like marketplace.

NHK TV is showing extensive damage including fallen smokestacks at Nippon Paper Factory facility, bridge road separation, fallen bridge lane, road cracks, fire, water coming out of drainage etc. It was a huge earthquake and will take some time to assess the damage.