Nishith Jain's picture

In a Training Loop 🔄

Nishith Jain

KingNish

·

AI & ML interests

AI is fun actually.

Recent Activity

posted an update 1 day ago

I tested Muon vs MuonClip vs Muon+AdamW for fine-tuning LLMs Just published a blog on that, Read here 👉 https://huggingface.co/blog/KingNish/optimizer-part1

upvoted an article 1 day ago

Muon vs MuonClip vs Muon+AdamW for Fine-Tuning

published an article 1 day ago

Muon vs MuonClip vs Muon+AdamW for Fine-Tuning

View all activity

Organizations

upvoted an article 1 day ago

Article

Muon vs MuonClip vs Muon+AdamW for Fine-Tuning

1 day ago

•

7

upvoted an article 3 days ago

Article

Tensor Parallelism (TP) in Transformers: 5 Minutes to Understand

6 days ago

•

59

upvoted a paper 6 days ago

PretrainZero: Reinforcement Active Pretraining

Paper • 2512.03442 • Published 8 days ago • 44

upvoted a paper 8 days ago

Critique Fine-Tuning: Learning to Critique is More Effective than Learning to Imitate

Paper • 2501.17703 • Published Jan 29 • 59

upvoted 6 collections 20 days ago

Kandinsky 5.0 Video Lite

Kandinsky 5.0 Video Lite is a lightweight 2B model that generates up to 10-second SD videos from English and Russian prompts with high visual quality. • 9 items • Updated 16 days ago • 8

Kandinsky 5.0 Video Lite Diffusers

Kandinsky 5.0 Video Lite is a lightweight 2B model that generates up to 10-second SD videos from English and Russian prompts with high visual quality. • 8 items • Updated 16 days ago • 4

Kandinsky 5.0 Video Pro Diffusers

Kandinsky 5.0 Video Pro is a 19B model that generates high-quality HD videos from English and Russian prompts with controllable camera motion. • 4 items • Updated about 8 hours ago • 6

Kandinsky 5.0 Video Pro

Kandinsky 5.0 Video Pro is a 19B model that generates high-quality HD videos from English and Russian prompts with controllable camera motion. • 5 items • Updated 16 days ago • 15

Kandinsky 5.0 Image Lite

Kandinsky 5.0 Image Lite is a 6B DiT-based model that generates and edits HD images from English and Russian text prompts with high visual quality. • 4 items • Updated 16 days ago • 13

Kandinsky 5.0 Image Lite Diffusers

4 items • Updated 16 days ago • 8

upvoted a paper 22 days ago

MarsRL: Advancing Multi-Agent Reasoning System via Reinforcement Learning with Agentic Pipeline Parallelism

Paper • 2511.11373 • Published 26 days ago • 12

upvoted 2 papers 27 days ago

Extracting alignment data in open models

Paper • 2510.18554 • Published Oct 21 • 9

TiDAR: Think in Diffusion, Talk in Autoregression

Paper • 2511.08923 • Published 29 days ago • 112

upvoted 2 papers about 1 month ago

Kosmos: An AI Scientist for Autonomous Discovery

Paper • 2511.02824 • Published Nov 4 • 5

INT v.s. FP: A Comprehensive Study of Fine-Grained Low-bit Quantization Formats

Paper • 2510.25602 • Published Oct 29 • 77

upvoted a collection about 2 months ago

interesting architecture

25 items • Updated Oct 20 • 3

upvoted a paper about 2 months ago

Latent Diffusion Model without Variational Autoencoder

Paper • 2510.15301 • Published Oct 17 • 48

upvoted an article about 2 months ago

Article

Sentence Transformers is joining Hugging Face!

Oct 22

•

85

upvoted a collection about 2 months ago

🎆 October 2025 - China Open Source Highlights

30 items • Updated 20 days ago • 13

upvoted a paper 2 months ago

Learning to Reason as Action Abstractions with Scalable Mid-Training RL

Paper • 2509.25810 • Published Sep 30 • 5