Manu Romero
@mrm8488
Followers
20K
Following
50K
Media
3K
Statuses
46K
CSO/Co-founder @maisaAI_. Head Contrib/ Ambassador🤗 @huggingface. Research 🌸@bigsciencew/@BigCodeProject | ex @narrativaAI
Murcia - SF
Joined January 2011
Work smarter and save more with TMG Industrial. 1-Year Warranty Free Shipping to Lower 48 States $100 OFF first order over $3,000 Reliable shelter, storage & equipment solutions built for the long haul.
0
0
2
It's mmBERT fine-tuned on Spanish instructions dataset using dllm repo
1
0
3
The Vice President of the Government of Spain says that artificial intelligence rebels against humans when it knows it is going to be shut down… and that this is happening… 🤦🏽♂️🤦🏽♂️🤦🏽♂️🤦🏽♂️
3
0
11
Hamiltonian Monte Carlo frames sampling from a probability distribution as a physics problem. By endowing "particles" with momentum and simulating their energy and motion through Hamilton's equations you can efficiently explore a distribution.
31
243
2K
I am looking for a Senior Applied AI Engineer. Core skills include: - Context Engineering - Prompt Evaluation - Popular Python AI Frameworks If you think you are a good candidate, my DMs are open!
3
3
6
It could be something like creating a mental representation of an "idea" whose "embedding" is close to the embedding of the original idea
What It Means to Understand @bimedotcom @Khulood_Almani @theomitsa @sulefati7 @NathaliaLeHen @IanLJones98 @rvp @bamitav @TheAIObserverX
https://t.co/xADlNFhPcM
0
1
3
🚨 New paper alert! We’re thrilled to share our new preprint “PrefixNLI: Detecting Factual Inconsistencies as Soon as They Arise” ✨ LLMs generate text one token at a time, but factuality checks still wait for a full sentence. We extend NLI to text prefixes, enabling the
4
17
65
🚀Excited to share our new work! 💊Problem: The BF16 precision causes a large training-inference mismatch, leading to unstable RL training. 💡Solution: Just switch to FP16. 🎯That's it. 📰Paper: https://t.co/AjCjtWquEq ⭐️Code: https://t.co/hJWSlch4VN
20
108
656
We've just published the Smol Training Playbook: a distillation of hard earned knowledge to share exactly what it takes to train SOTA LLMs ⚡️ Featuring our protagonist SmolLM3, we cover: 🧭 Strategy on whether to train your own LLM and burn all your VC money 🪨 Pretraining,
20
84
461
this could be because its a publicly available text and it was trained on it?
DeepSeek-OCR is the best OCR ever. It parses this extremely hard to read handwritten letter written by mathematician Ramanujan in 1913 with a frightening degree of accuracy. Not perfect, but beats former best dots ocr. Bonus points if you can spot the errors. Try it here:
2
4
91
👀
Had some really interesting discoveries recently: If a model performs extremely stable on one benchmark. Let's say a model is always getting 62% on SWEBench no matter what prompts or scaffold you used. It DOES NOT mean that the model is robust. It actually means that the model
0
0
0
🔥Excited to share our new work: "A Practitioner's Guide to Multi-turn Agentic Reinforcement Learning"! We systematically study what actually works (and what doesn't) for agentic multi-turn RL, breaking down the design space into 3 pillars: 🌎Environment, 🤖Policy, and ⭐Reward.
2
30
172
Trying to reproduce it with a Spanish RoBERTa...
Nice, short post illustrating how simple text (discrete) diffusion can be. Diffusion (i.e. parallel, iterated denoising, top) is the pervasive generative paradigm in image/video, but autoregression (i.e. go left to right bottom) is the dominant paradigm in text. For audio I've
0
0
4
Don’t hire managers for your team. Hire talent and let leadership emerge naturally.
0
0
3
Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single,
667
3K
24K
New blog post analyzing the top 50 entities with the most downloaded models on @huggingface 🤗! The purpose here is to get an idea of the profile of the models with the greatest impact in open source (we are not interested in closed models here!). Some key findings:
7
24
126
♥️🐴
🎥 #12DeOctubre | Se mantiene la participación de las unidades a caballo. Estaba en duda si estas agrupaciones podrían participar en el desfile terrestre a causa de la lluvia. Las unidades a caballo son las encargadas de cerrar el desfile terrestre https://t.co/RSmXESe2gZ
0
0
2