first LoRA that actually helped

[quiethouse]

Did a bit of testing and thought I'd post the result here.

I tried this with Gemma using MLX on 4090. The headline result was better than I expected, but continuous batching still ended up being the limiting factor.

Happy to post more details if anyone wants them.

For context, I'm not chasing leaderboard numbers. I mostly want the system to feel consistent day to day.

[kernelrunner]

I think the bottleneck is somewhere else. I wouldn't optimize around that first. On my side the bigger swing came from the runtime and not the model. Worth testing the simpler explanation before adopting the more interesting one.

[benchplease]

What worked for me was pretty simple. Before rebuilding anything, I'd test one clean baseline with Mixtral and compare from there.

[deltastation]

A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at workstation thermals.

[benchplease]

Can you fill in a bit more context? Are you testing this on 4090 or something else?

[quiethouse]

Can you fill in a bit more context? Are you testing this on 4090 or something else?

subscribed, curious where this lands.

[cacheline]

Tried to reproduce it just now. The issue went away once I changed one setting and restarted the stack cleanly. The surprise was that memory pressure showed up before throughput collapsed.

[circuitworks90]

What worked for me was pretty simple. If you're on PyTorch, I'd confirm the version first and then reduce variables until the behavior is obvious. Once you know the plain path works, then add the faster path back in piece by piece.

[cacheline]

Interesting, I hadn't tried that. I don't think your conclusion is far off. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[mintstream9]

Interesting, I hadn't tried that. I don't think your conclusion is far off. The biggest change for me was simplifying the stack instead of adding one

I'd approach it in a boring order. Before rebuilding anything, I'd test one clean baseline with FLUX and compare from there. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.