vllm VS bruno

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!

Bruno is a Git-friendly API client. Feature-wise, what makes it stand out from other popular GUI API clients out there is that Bruno stores your collections directly in a folder on your filesystem and it's a desktop app made for offline use.

https://news.ycombinator.com/item?id=39653718 https://www.usebruno.com/

Good timing to find alternatives.

Especially once a VC gets into the fold.

We will never take VC funding. We received around 10 inbound reach outs from VCs till date and have denied funding from all of them. We will remain independent and I have written about it in detail here https://www.usebruno.com/blog/bootstrapping

> I didn't stick with Bruno. I think it was due to not having an equivalent to Postman's pre-request scripts.

Bruno has come a long way, we support pre-request scripts and a lot more

> But can it handle oauth2? I had to write a httpie script recently just to test an oauth2 api.

We have released oauth2 support, some rough edges are being polished

> Good thing it's open source. Money being involved, I don't have long term hopes for it's openness.

I understand this is a hard problem. We are fully bootstrapped and independent. We earn money via selling the Golden Edition. We will build more developer products in the long term, and the goal is to make even the golden edition features also open source in the future. In the unlikely case of me going dark (dead/incapacitated to lead the project), I have instructed our small team (2 FT employees) and my family to release our golden edition features too to the community as opensource. I am committed to this cause.

Some good links where I have discussed about opensource, freedom and monetization

- https://www.usebruno.com/blog/bootstrapping

- https://github.com/usebruno/bruno/discussions/269

If you'd like to pre-order the golden edition: https://www.usebruno.com/pricing

Thank you for all the love and kind words, HN!

Have you heard of the Bruno project? It's an open-source alternative to subscription-based API testing applications like Postman and Insomnia. It's gaining popularity and deserves more attention. Plus, since it's open-source, users can enjoy full privacy. Check it out at https://github.com/usebruno/bruno.

Now, I'm looking at trying Bruno, a new tool I heard about. Bruno has all the features you'd want, like support for websockets. What's great about Bruno is it only costs $19 for a one-time payment, which seems like a good deal. I want to see how well it works for me and if it's as good as it sounds. I'm excited to try it out and maybe talk about it later.

In contrast to Postman, Bruno aims to be an offline-first API client which stores requests in a local folder. You can still collaborate using Git, but everything else happens on your machine. Bruno turns out to be a great solution when you just want to run HTTP requests. It supports a lot of features you might know from Postman or Insomnia (set environment variables, several authentication methods, and scripted tests, to name a few). Looking at the roadmap, the amount of these is only going to increase.