📢Thrilled to introduce ATLAS 🗺️: scaling laws beyond English, for pretraining, finetuning, and the curse of multilinguality. The largest public, multilingual scaling study to-date—we ran 774 exps (10M-8B params, 400+ languages) to answer: 🌍Are scaling laws different by

Trying this Face-Swap LoRA for Qwen Image Edit (difficult task w/o a LoRA) Incredible how far we can push this model with a bit of finetuning

New website is up (some finetuning will still be made the coming days) and those two will become available this sunday 3pm (UTC+1) here: https://t.co/8rmLrJxFnu

the biggest problems of LLM's are, they hallucinate so bad especially when they see unseen data > prompt engineering, RAG, and finetuning are three very common AI engineering techniques that you can use to adapt a model to your needs. > Prompt engineering : a technique of

Super excited to have gotten early access to Tinker by @thinkymachines! Tinker is great at making it easy to post-train, but also meant I wound up with lots of model checkpoints 😅 This is why I built a finetuning journal to keep track of each model iteration, with the help of

[ml grind] 👩‍💻optimized tokenizer encoding 📖 read anatomy of a modern finetuning api https://t.co/V8hfpsqfUb (very pathetically passed verbal part of gre practice)

🌟 Excited to be at #NeurIPS2025 (Dec 1–8)! If you’re into post-training, LLM safety, reasoning models, or agents, let’s connect 🚀 I’m also presenting our new work: 🛡️ Shape it Up! Restoring LLM Safety during Finetuning ShengYun Peng, Pin-Yu Chen, Jianfeng Chi, Seongmin Lee,

PhD application season is here! The CLEAR Lab @ MIT CSAIL is recruiting students excited about human-centered robot learning and algorithmic HRI. If you're interested in working on: 🤝 IRL & preference learning 🎛️ steering & finetuning large behavior models (diffusion

🚀 New Paper: We’ve just released HalluDesign – a novel framework for protein optimization and de novo design using AlphaFold3-style models, without finetuning or gradient backpropagation! 🎉 https://t.co/tmqKqPMp2b

I spent the last ~2 weeks recreating the @thinkymachines LoRA without Regret experiments from scratch. SFT: Qwen3-4B on No Robots dataset RL: Qwen3-1.7B on MATH dataset so cool to see rank 1 LoRA matching the performance of full finetuning! 🤯

if you want to get into ai engineering, understand this first : it is basically of 3 layers - 1) application layer : > building ai products > fullstack + agents, agentic stuff 2) model layer: > training and finetuning models > LLM's, CV 3) Infrastructure layer: > deploying

this GitHub repo is a hands-on guide to building ChatGPT-style language models in PyTorch from fundamentals to full implementation it contains the code for developing, pretraining, and finetuning a GPT-like LLM

New EMNLP main paper: “Finetuning LLMs for Human Behavior Prediction in Social Science Experiments” We built SocSci210—2.9M human responses from 210 social science experiments. Finetuning Qwen2.5-14B on SocSci210 beats its base model by 26% & GPT-4o by 13% on unseen studies.🧵

What timeline are we in where pewdiepie is finetuning an LLM??? Someone write this man a check

If your finetuning doesn’t have the same problems as your pretraining, you’re not doing the finetuning right.

teensy tiny game update! title screen ui :> it just needs some music and a bit of finetuning pixel art: @moqidesu #momijiiro

Thinking Machines to Raise $5B | Download full report = https://t.co/VG5DYk67DL • Targeting $4–5B raise at $50B+ valuation, after earlier $2B at $10B • Only nine months old; capital used for compute + top researcher hiring • Released Tinker API for finetuning models on

Presenting our paper at @pals_nlp_wrkshp at @emnlpmeeting "Is Active Persona Inference Necessary for Aligning Small Models to Personal Preferences?" TLDR; We find model inferred preference to be a better finetuning prefix for generalization, faithful attribution, and less bias

i've spent a lot of time finetuning and customizing this model so far~ it's so cute and expressive! #PrismCustomizableVtuber