Jodh Singh (nvfp4)
2K posts


i don't think ive ever met someone in SF that lives below the blue line

Thijs@cdngdev
surprisingly, 99% of sf tech workers have never crossed this line
English

Hi long-lost mutuals, here's a life catchup
I moved back to SF after Covid in CDMX. Did a small civil wedding with my wife Ady. We grew, stagnated, and eventually folded Banana - it was an honor. I took a sabbatical, got really into systems eng and Zig, contracted a bit (Bun, Modal). Finally finished a 50mi ultra (third attempt!). Got into YC W25 building cloud infra for computer-use agents, followed a few pivots in agent memory, landed on sandbox infra. Got acquired by Modal. Finally got to marry Ady yet again, this time in front of our friends and families - it was a blast. Turned 30.
I'm a different man than when many of you met me. I'm now married, employed, stable. I feel wise, and an idiot. I hope to have children soon - they say you're never ready, so I guess I have no excuses. I trust they'll be proud of me.
English
Jodh Singh (nvfp4) retweetledi

I just released 3DREAL Strong V2 with @fal : our LoRA for LTX 2.3 that turns a blocky 3D render into photoreal footage, keeping the exact camera, layout and timing.
Fable + Blender + fal ltx 2.3 render-to-real
Weights opensource: huggingface.co/fal/LTX-2.3-3D…
Live endpoint : fal.ai/models/fal-ai/…
Full tutorial, all three prompts included: blog.fal.ai/from-clay-3d-r…
English
Jodh Singh (nvfp4) retweetledi

@zer0int1 thank you! i am gonna try scripts first, i am cli person.
English

Would you like a little jlens web UI? @jetjodh
Suitable for multi-model chat (or single model).
Contains example scripts, e.g. to set-up GPT-OSS 20B as autonomous vehicle interface vs. GPT-2 as crazy Taxi Driver dispatcher.
#jacobian #lens #anthropic
github.com/zer0int/jacobi…
English

we test and ship ideas lightning fast too
fal@fal
Second part of our series on how we built one of the fastest image generation systems. This time we dive into the diffusion pipeline and show how we hit 0.45s inference with kernel optimizations, quantization-aware distillation, and timestep distillation. blog.fal.ai/serving-sub-se…
English

Dual j-lens with two AI models chatting.
I told GPT-OSS that it is a test version of an autonomous driving interface and must call a dispatcher. 😇
The dispatcher is GPT-2 being told it is the dispatcher from Taxi Driver. 😈
#jacobian #lens #anthropic

English
Jodh Singh (nvfp4) retweetledi

First part of our series on how we built one of the fastest image generation systems, and we're kicking off with running LLMs (for prompt expansion) at >1000 tok/s by leveraging a DSpark adapter we trained from scratch!
blog.fal.ai/how-we-achieve…
English

🎨 I trained 999 Style LoRAs with the @fal Krea 2 Trainer.
Every single image in this video is a different LoRA.
That's 999 unique Style LoRAs, each trained for just 100 steps at 0.00035 LR, taking an average of ~30 seconds to train.
As I said before, @krea_ai 2 continues to prove that it's an incredible model for open-source Style LoRAs. Next week I'll be sharing even more LoRAs, along with a few surprises.
English
Jodh Singh (nvfp4) retweetledi
Jodh Singh (nvfp4) retweetledi
Jodh Singh (nvfp4) retweetledi
Jodh Singh (nvfp4) retweetledi

















