NVFP4 / INT4 tradeoffs for Mistral

[mangotinker73]

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?

Would rather learn from your scars than collect my own all at once.

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

[mintstream9]

A couple things I'd check first. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at prefill/decode split. Once you know the plain path works, then add the faster path back in piece by piece.

[rocmstack25]

I'd approach it in a boring order. Before rebuilding anything, I'd test one clean baseline with Qwen and compare from there.

[signaltrack44]

What worked for me was pretty simple. On dual 3090, the easiest win is often to back off the aggressive settings and see whether the problem survives. A lot of these threads end up being one bad config plus a pile of conclusions drawn from it.

[thetashed24]

My first guess would be this: If you're on SGLang, I'd confirm the version first and then reduce variables until the behavior is obvious. Not glamorous advice, but repeatable baselines save a lot of time.

[vectorport]

Thanks, that helps. I ended up in about the same place using PyTorch. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.

[cudapoint]

Thanks, that helps. I ended up in about the same place using PyTorch. I got better results once I stopped chasing one metric and just looked at whethe

Need a little more detail here. Is this local-only or are you routing anything externally?

[mintstream9]

Tried to reproduce it just now. On 5090 with vLLM, I saw basically the same behavior. The surprise was that memory pressure showed up before throughput collapsed.

[orbitgarden]

following this because I'm in the same boat.

[thetatinker77]

My first guess would be this: Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at KV cache pressure.