Yashasav Prajapati

building and optimizing LLM inference, evals, software engg, databases, infra
building cool stuff
CS IIT Ropar

[Yashasav Prajapati @infra_engineer4]

@ClementDelangue @nvidia Remote? Inference engineering? Super excited!!!!!

[Yashasav Prajapati @infra_engineer4]

The timeline went from minutes, to an hour, to next morning

[Yashasav Prajapati @infra_engineer4]

Hey @wajo_ai, I tried asking Fo to tweet from my account, connected it, but it is stuck on the login itself since the beginning. Any solutions to this?

[Yashasav Prajapati @infra_engineer4]

Invite for fo - https://t.co/rXozjnqaCM

[Shivani Poddar @shivanipod]

I led engineering at Google DeepMind.

Today, I'm proud to introduce Fo to give personal AI something no lab ever has... Humans.

Other personal AI's pretend AI can do everything. Fo employs humans to do tasks that AI cannot.

- 2x better at real-world task completion (beats other agents by 69%)
- 94% trust rate (4x less likely to leak private info vs Muse, Instinct)

Sign up for free: https://t.co/9ndFhcN5mO

[Yashasav Prajapati @infra_engineer4]

@wafer_ai Are you guys hiring?

[Yashasav Prajapati @infra_engineer4]

@sakshiseninsane @Muse True need @Muse in India!

[Yashasav Prajapati @infra_engineer4]

@mister_whistler The problem is not the building part, its the data part

[Yashasav Prajapati @infra_engineer4]

@mister_whistler This is amazing!

[Yashasav Prajapati @infra_engineer4]

Why are people not talking about this enough??? This is such a big deal!!!

[solidSF @solidSF]

we love to compete

[Yashasav Prajapati @infra_engineer4]

@the_real_ridwan @instantricecook @typesafeai I was talking about the wrapper over jev

[Yashasav Prajapati @infra_engineer4]

@arpit_bhayani btw another new model dropped with a new architecture today, no weights - https://t.co/GoNX2NsGQ7

exciting times ahead.

[solidSF @solidSF]

today we announce something truly special

introducing: Jesse v1

this model architecture is all new, from scratch, built and trained using public, popular datasets

it consumes 0 tokens, and does not have weights

it reasons, it believes, and it's very, very fast

[Yashasav Prajapati @infra_engineer4]

I talked to Jesse V1, it says that its an LLM made by Meta, but its architecture is different internally like a giant lookup table for fast decisions, not sure how this looks internally but excited for the future of this

[solidSF @solidSF]

today we announce something truly special

introducing: Jesse v1

this model architecture is all new, from scratch, built and trained using public, popular datasets

it consumes 0 tokens, and does not have weights

it reasons, it believes, and it's very, very fast

[Yashasav Prajapati @infra_engineer4]

@niveditjain why do you think so? ai infra specifically LLM serving or something different?

[Yashasav Prajapati @infra_engineer4]

@aslambeg84 This depends on the type of bug that was noticed, but usually one would start with logs for a particular trace id, check the info, warn and error logs, then check the database state problems, then code, look for the diff for the request path, then the env variables.

[Yashasav Prajapati @infra_engineer4]

@twaldin @mazeincoding It will always answer P, because that’s what is close to the letter you asked, even if you asked NP it will say P because then it has NP related to P, so P is more favoured.

[Yashasav Prajapati @infra_engineer4]

@hollylawly 1-2 years of exp in software engineering, no opening 😭

[Yashasav Prajapati @infra_engineer4]

@danialhasan Interested

[Yashasav Prajapati @infra_engineer4]

@hackgoofer Applied !

[Yashasav Prajapati @infra_engineer4]

@CompleteSkeptic I am quite interested in the underlying inference engine being used to serve it, how much role does the inference engine have in the structured output, does it parse the output or what?

Even more about how the model looks architecturally and what can be optimized even more!