MLX model loading quirks

[macrolab]

Spent some time on this today and wanted to compare notes.

I've gone back and forth between TensorRT-LLM and TensorRT-LLM for Llama. The interesting difference for me has been reranking.

Interested in tradeoffs more than fan loyalty.

My use case is pretty ordinary: local chat, some document work, and occasional batch jobs.

[mlxgarden]

worth checking the driver first.

[routecraft]

A couple things I'd check first. On A6000, the easiest win is often to back off the aggressive settings and see whether the problem survives. Once you know the plain path works, then add the faster path back in piece by piece. The logs were somehow calmer than I was.

[sanescope]

That matches what I was seeing. The pattern around speculative decoding has been pretty similar for me. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[mlxgarden]

What worked for me was pretty simple. On B200, the easiest win is often to back off the aggressive settings and see whether the problem survives.

[quietarch]

I'd approach it in a boring order. If you're on SGLang, I'd confirm the version first and then reduce variables until the behavior is obvious.

[mlxgarden]

Tried to reproduce it just now. I couldn't reproduce the slowdown until I pushed context much higher.

[quantyard]

My first guess would be this: Before rebuilding anything, I'd test one clean baseline with Qwen and compare from there. Not glamorous advice, but repeatable baselines save a lot of time.

[visionyard23]

What worked for me was pretty simple. If you're on SGLang, I'd confirm the version first and then reduce variables until the behavior is obvious. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

There are a few different conversations hiding inside this thread, so being explicit about the goal helps.

[quantsupply]

subscribed, curious where this lands.