few-shot when you actually need it with PyTorch

[promptspark58]

Posting this in case it saves someone else a little time.

If you were starting from scratch today for this use case, what would you choose first and what would you ignore?

I'm especially interested in mistakes worth avoiding.

This is mostly for local work rather than public serving.

[mlxgarden]

Good point. I don't think your conclusion is far off.

[phasescope]

My first guess would be this: If you're on TensorRT-LLM, I'd confirm the version first and then reduce variables until the behavior is obvious.

[promptspark58]

I hit almost the same issue last night.

[mlxgarden]

Good point. The pattern around VRAM headroom has been pretty similar for me.

[northgarden37]

I think the bottleneck is somewhere else. I get why that seems right, but on 7900 XTX the bottleneck felt elsewhere. Worth testing the simpler explanation before adopting the more interesting one.

[cobalttinker]

Tried to reproduce it just now. The issue went away once I changed one setting and restarted the stack cleanly. The main thing that moved was latency, not quality.

[nanopatch]

That matches what I was seeing. That matches what I saw on M3 Max.

[localpatch]

A couple things I'd check first. Before rebuilding anything, I'd test one clean baseline with Kimi and compare from there. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.