getinclub

practice job before job

[getinclub @getinhq]

[dailysimulation - 04] : Today’s Daily Simulation: “Capacity planning.”

4 GPUs. Model fits. Capacity still sucks.
TP4 ≠ more goodput.
Qualify memory + topology before you rank latency.

try out here : https://t.co/o6Ohi7mTyw

#llm #inference #simulations #getinclub #learn

[getinclub @getinhq]

well on the other hand introducing https://t.co/KHcdgDpL0s where you practice job before job itself,
> study with free resources with challenges
> try out production bugs/problems in the form of simulations
> build your products solo or with a team of your choice.

try it out

[getinclub @getinhq]

made this little intro using gpt-astra ultra.

[getinclub @getinhq]

go check this out ....

[Kartikey Sinha @_iamkartiks_]

well finally did something, now on https://t.co/JPUUzKisc8, you can copy any job description and then find out simulations that you can practice specific to that job role.

it's just a small part of a whole job sprint feature that i have been working on, where users can build small a week or so long learning sprints containing free learning resources, what kind of tasks you might get to work on in the formats of simulations, trying to improve the automated building for the sprints, soon it will be rolled out as well.

but yeah this was a small thing but fun experiment for quick learning.

-> here jev classifies the JD against the skills (for which we have simulations on our platform).
-> compose jev's answers into the skill sets we need to act on and then we used these to search simulations.
-> first we create catalog of simulations using the previous step using skill overlaps, descriptions, titles of simulations, and at the end we create a rankeddd list of simulations you can try out using the relevant score.

[getinclub @getinhq]

[dailysimulation - 03] : Today’s Daily Simulation: “The Timeout That Corrupted the Next Message.”

The request timed out. You retry. Now the next valid message reports an impossible frame size.

Reconnecting fixes it. But why?

[getinclub @getinhq]

more updates are coming meanwhile make sure to checkout our learning directory with all free resources.

[Kartikey Sinha @_iamkartiks_]

@getinhq has some new great updates that are worth checking out [ one is below, make sure to check out other ones as well ]

1) enhancement in acted-scenarios :
we have two types of simulations to actually practice like you do in a job, acted-scenario are complex enhanced version, take it as 3-4 stories problem tickets that a developer work on.

now acted-scenario comes up with internal ai co-pilot that can assist you with some parts of the tasks but its a dumb model so you will be doing the most work anyway.

acted scenario tracks how you approach the problem, how you work on them , time taken and then at the end gives you score once you have completed it.

challenges are github repositories like a developer works on, no coding challenges will be in browser sandbox.

Our goal is to replicate an official setting as much as we can.

try out here : https://t.co/MnRBFcyqCt

[getinclub @getinhq]

[dailysimulation-03]

Pod Ready in 2s. Model ready in 47s.
First request: 45,200ms 💀
HTTP /health ≠ inference-ready.

Fix cold start + readiness probes in 8 min 👇

Try the simulation here :https://t.co/kR2wzJCTtk

#Kubernetes #LLM #PlatformEngineering

[getinclub @getinhq]

[dailysimulation-02] : The Quantization Accuracy Cliff

Your model fits in memory. Then the quality checks fail.

Try it here : https://t.co/reZqrYK0SL

[getinclub @getinhq]

dailysimulation-01.

starting a series where we will unpack role specific production problems one at a time.

today’s is on llm inference, go try it yourself here : https://t.co/bbPImIcdIb

[ web browser recommended to work on this ]

[getinclub @getinhq]

What exactly happens at Day-01,
In most cases you will be confused, because its not writing a algorithm , it's a vague ticket, a foreign repo and a deadline.
That's not a knowledge gap, its a practice gap.

Swipe through, and then practice it before it counts. (thread)

[getinclub @getinhq]

What we are doing here ? ( a thread )

[getinclub @getinhq]

tune in, its here.