Manish Tiwari

AI Engineer | I build the plumbing LLM apps run on: sandboxes, guardrails, inference benchmarks. 41 merged OSS PRs, saferun-ai on PyPI. Open to remote worldwide

[Manish Tiwari @compmanish]

Every "we have 600 GB of pooled VRAM" spreadsheet is fiction.

A model has to fit on ONE node before it can train on any of them. So the ceiling is your largest single GPU, not the sum.

I built a planner to compute this properly. Here's what it says. 🧵

[Manish Tiwari @compmanish]

An opinion by someone at OpenAI

[Manish Tiwari @compmanish]

@itsOmSarraf_ @thericebowlgirl

[Manish Tiwari @compmanish]

@NaceAI what about hidden texts into documents? is it only OCR based or the older parsing methods are used too for docs?

[Manish Tiwari @compmanish]

@AleeeM1 @RoboticswithV in delhi-NCR by any chance, builder community here is also strong and some of the promising AI labs of India are based out of this region!

[Manish Tiwari @compmanish]

@AsadSiddiiqui @kunalstwt @pratykumar yes the amount of work indus by sarvam is doing for free, none of agents platform at that scale along with that level of answers are giving it for free, deepseek, Z ai etc are good but very very slow there one answer equals 10 answers from this!!

[Manish Tiwari @compmanish]

@itsOmSarraf_ @thericebowlgirl when it will be publicly available!!

[Manish Tiwari @compmanish]

@joelniklaus @SarvamAI testing for sarvam-code internal testing might be best at this stage!

[Manish Tiwari @compmanish]

Finally a rare addition into the syllabus of Traditional ML!!

[Swann Bessa @swann_bessa]

Introducing e-means, an algorithm beating k-means in 10x fewer epochs.

With order of magnitude wall-clock speedups, e-means is released in FAISS and needs no hyperparameter tuning beyond k. It is robust to various scales and datasets. Still skeptical ? Just try e-means [1/10]🧵

[Manish Tiwari @compmanish]

THE ULTRA-SCALE PLAYBOOK!! Nice platform to invest time and get more deeper for my "mesh" !!

#huggingface

[Manish Tiwari @compmanish]

just fast changing world!

[Anthropic @AnthropicAI]

Claude Fable 5.2 and Claude Mythos 5.2 are available now:

[Manish Tiwari @compmanish]

@prakhar210106 @MahekPatel24 aapne agree kar diya tab to ekdum sahi h

[Manish Tiwari @compmanish]

@jindal_nayanx Thanks!!

[Manish Tiwari @compmanish]

what a beautiful short film @anusinghc

[Manish Tiwari @compmanish]

@FIR31415 🙂🙂F

[Manish Tiwari @compmanish]

@goyalaman03 @Reidoerei looking for some good indic models

[Manish Tiwari @compmanish]

@goyalaman03 @Reidoerei will the models be available publicly? I just wanted to ran some experiments with them, few benchmarking and recent one WSP!!

[Manish Tiwari @compmanish]

@goyalaman03 @Reidoerei I somehow founded your and kashi labs
They still haven't updated anything but you guys are killing!!