70B still worth the pain with llama.cpp

[quantsupply]

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.

[vramarch]

worth checking the driver first.

[orbitgarden]

That matches what I was seeing. I don't think your conclusion is far off.

[latentscope35]

What worked for me was pretty simple. If you're on TensorRT-LLM, 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.

[oxideworks59]

What worked for me was pretty simple. If you're on TensorRT-LLM, I'd confirm the version first and then reduce variables until the behavior is obvious

That matches what I was seeing. That matches what I saw on mini PC. Worth rechecking your defaults though, because one bad assumption can make the rest of the thread look confusing.

[modgarden]

I'd approach it in a boring order. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at INT4. Not glamorous advice, but repeatable baselines save a lot of time.

[oxidegrove40]

I'd approach it in a boring order. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at INT4. Not glamorous advi

Interesting, I hadn't tried that. The pattern around MoE serving has been pretty similar for me. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.

[pixelspark]

[quote]I'd approach it in a boring order. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at INT4. Not glamoro

Thanks, that helps. I ended up in about the same place using LM Studio. The biggest change for me was simplifying the stack instead of adding one more optimization layer.

[stackpilot]

[quote][quote]I'd approach it in a boring order. Start by verifying the driver/runtime combo, then test with a smaller prompt, then look at INT4. Not

What worked for me was pretty simple. If you're on LM Studio, 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. At this point the real bottleneck is my patience.

[edgekite]

That matches what I was seeing. I ended up in about the same place using llama.cpp.