talk VS vllm

Also curious to hear about your setup. Using whisper too? When I was experimenting with it there was still a lot of annoyance about hallucinations and I was hard coding some "if last phrase is 'thanks for watching', ignore last phrase"

I was just googling a bit to see what's out there now for whisper/llama combos and came across this: https://github.com/yacineMTB/talk

There's a demo linked on the github page that seems relatively fast at responding conversationally, but still maybe 1-2 seconds at times. Impressive it's entirely offline.

Here’s a project attempting to do just this!

https://github.com/yacineMTB/talk

GitHub repository: https://github.com/vllm-project/vllm

I guess the root cause of my claim is that OpenAI won't tell us whether or not GPT-3.5 is an MoE model, and I assumed it wasn't. Since GPT-3.5 is clearly nondeterministic at temp=0, I believed the nondeterminism was due to FPU stuff, and this effect was amplified with GPT-4's MoE. But if GPT-3.5 is also MoE then that's just wrong.

What makes this especially tricky is that small models are truly 100% deterministic at temp=0 because the relative likelihoods are too coarse for FPU issues to be a factor. I had thought 3.5 was big enough that some of its token probabilities were too fine-grained for the FPU. But that's probably wrong.

On the other hand, it's not just GPT, there are currently floating-point difficulties in vllm which significantly affect the determinism of any model run on it: https://github.com/vllm-project/vllm/issues/966 Note that a suggested fix is upcasting to float32. So it's possible that GPT-3.5 is using an especially low-precision float and introducing nondeterminism by saving money on compute costs.

Sadly I do not have the money[1] to actually run a test to falsify any of this. It seems like this would be a good little research project.

[1] Or the time, or the motivation :) But this stuff is expensive.

The easiest is to use vllm (https://github.com/vllm-project/vllm) to run it on a Couple of A100's, and you can benchmark this using this library (https://github.com/EleutherAI/lm-evaluation-harness)

Great question! scheduling workloads onto GPUs in a way where VRAM is being utilised efficiently was quite the challenge.

What we found was the IO latency for loading model weights into VRAM will kill responsiveness if you don't "re-use" sessions (i.e. where the model weights remain loaded and you run multiple inference sessions over the same loaded weights).

Obviously projects like https://github.com/vllm-project/vllm exist but we needed to build out a scheduler that can run a fleet of GPUs for a matrix of text/image vs inference/finetune sessions.

disclaimer: I work on Helix

FYI, vLLM also just added experiment multi-lora support: https://github.com/vllm-project/vllm/releases/tag/v0.3.0

Also check out the new prefix caching, I see huge potential for batch processing purposes there!