Multimodal discussion for Phi

[visionforge]

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

Two paths both seem viable, but they break down differently once consumer GPUs becomes important.

Trying to avoid rebuilding the whole thing twice.

[visionforge]

A couple things I'd check first. On mini PC, the easiest win is often to back off the aggressive settings and see whether the problem survives. Not glamorous advice, but repeatable baselines save a lot of time.

[southtinker]

Good point. I ended up in about the same place using TensorRT-LLM. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[macrotinker]

Interesting, I hadn't tried that. I ended up in about the same place using Ollama. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[marbleyard32]

worth checking the driver first.

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

[homelabhenry]

I think the bottleneck is somewhere else. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit. Could be a workload difference, but I wouldn't generalize too fast from one setup.

[visionforge]

I think the bottleneck is somewhere else. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit. Could be a wo

I tested this tonight. My result was mixed: first run looked bad, second run looked normal. The surprise was that memory pressure showed up before throughput collapsed.

[dockbrook21]

[quote]I think the bottleneck is somewhere else. The reported numbers can look compelling, but in ordinary usage I didn't see the same benefit. Could

What exact setup are you running? What model, what runtime, and roughly what context size?

[mangovault67]

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

[macropilot]

A couple things I'd check first. On Threadripper box, the easiest win is often to back off the aggressive settings and see whether the problem survives.