a very normie guy, just eat and sleeps.
Clone
Joined June 2024
- Tweets479
- Following103
- Followers255
- Likes1.5K
Pinned Tweet
Hi I am panda.
I got phone digit jee rank in jee and 60% in boards.
I am quite normie person and looking for friends.
So, let’s connect and be friends.
panda retweeted
nerds, don’t save this and keep scrolling - llmdojo.dev
panda retweeted
got access to space bunny in opencode this week and ended up with this three.js + webgl valley running in the browser.
it’s a flash-level model so the loop is quick, but the scene itself came out pretty calm. i kept going back to the lighting and the water more than any one object.
what actually got me is the camera. it just floats through the grass and the river and those ruins toward the shrine, and the whole place feels a bit bigger once you’re moving through it instead of staring at it.
panda retweeted
fable 5.1 was leading wandr at 0.601 for $12.76 a task.
astra just pushed that to 0.682 at $11.98.
so we got +13.5% on score while also getting ~6% cheaper.
usually the frontier moves because someone spends more.
this time it moved up and left.
sam and the team really cooked this time
panda retweeted
Something changed in agents today and most people will miss it.
Traditionally, most companies will ship one official agent and freeze.
Nobody wants to wait on that queue. However, the work is already sitting in your slack, your inbox, your browser tabs.
Skydive is built for that person.
150+ agents already running in-house.
support, eng, growth, ops.
not one mega-agent.
Just simply a bench that can hand work off.
you write one sentence.
→ it becomes a teammate with a real browser.
→ it lives in slack / email / iMessage.
→ you send the task at night and pick it up done.
if you are the one doing the busywork between tools, that is why you would actually use this.
🤝 Paid partnership
panda retweeted
POV -
you found one of the best resources to study ML-Systems engineering.
my personal ml systems notes
over the last couple of months, ive been diving really deep into ml systems both on the training and inference for llms. this is a personal collection of notes covering distributed computing, parallelism, quantization, and pytorch internals mostly from my experiments.
1. distributed techniques - covers distributed training fundamentals: nccl collectives (gather, all-gather, reduce, all-reduce, scatter, reduce-scatter), mixture-of-experts, parallelism strategies (dp, ddp, zero, tensor/pipeline parallelism), and torch.distributed basics.
2. quantization - model quantization from first principles: symmetric/asymmetric quantization, llm.int8(), awq, smoothquant, gptq/obs/obq, and quip.
3. pytorch internals
4. jax scaling book - some solved exercises from the jax scaling book.
ill keep adding notes and refine the content to make it more presentable over time.
While every single guy is trying to be feminist, I am literally a punching bag plushie to a lot of feminists.
They beat and hug me, idk why tf they are so crazy.
panda retweeted
Finally got my turn to write on this, and giving hot take again -
What you guys missed is a lot, and a lot of reality shows actually showcase stuff similar to this.
I have lost the count of such rubbish flexed as body count, hiding all the cheap and dirty work did behind.
Let’s be honest none of such lust games never come clean.
If this mahapurush worded even slightly better and buttery English, folks who are against him, would have actually find it funny, and called others “incels” or blame them of lacking so called “game”.
In the end, I am glad this case got taken, and masses opened their ignorant brains.
panda retweeted
This weekend, I wanted to announce a similar tech twitter NCR meetup among X oomfs.
But since some of our homies our going, so it’s better to make this one a short circle of individuals.
Anyways, if you are available this weekend, tired off rotting in your bed and want to some cool people on X,
Feel free to DM.
