vllm

Yes, I've used the v3.2 3B-Instruct model in a Slack app. Specifically using vLLM, with a template: https://github.com/vllm-project/vllm/blob/main/examples/tool... Works as expected if you provide a few system prompts with context.

Hey there! We currently use vllm under the hood and vllm doesn't support Codestral (yet). We're working on expanding our model support. Hence (almost) any model. Thanks for testing! :) https://github.com/vllm-project/vllm/issues/6479 - Billy :)

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 No

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)

I was wondering if it uses something like vLLM[0] or Llama.cpp[1]. Seems to be Llama.cpp via 'Nitro', which was discussed here before [2]. [0] https://github.com/vllm-project/vllm [1] https://github.com/ggerganov/llama.cpp [2] https://news.ycombinator.com/item?id=38887531

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 experimental 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!