Please welcome K2-65B🏔️, the most performant fully-open LLM released to date. As a blueprint for open-source AGI, we release all model checkpoints, code, logs, and data. About K2: 🧠65 billion parameters 🪟Fully transparent & reproducible 🔓Apache 2.0 📈Outperforms Llama 2 70B

🚨Llama 3.1 405B eval just dropped🚨 🥇 in instruction following 🥈 in reasoning On par with GPT-4o in math and coding It’s a great day for the open-source community!! Full evals on the challenging, contamination-free benchmark ➡️ livebench.ai

✨ Check out our revamped repo! Analysis360: Open Implementations of LLM Analyses 🔗 github.com/LLM360/Analysi… Featuring tutorials on: 💾 Data memorization 🧠 LLM unlearning ⚖️ AI safety, toxicity, & bias 🔍 Mechanistic interpretability 📊 Evaluation metrics

Quantifying the Value of Information is generally intractable, and prior work uses heuristic approximations that are still quite expensive. We propose PS-BAX, which extends posterior sampling to the Bayesian Algorithm Execution setting: arxiv.org/abs/2410.20596 (appearing at

SuperCharged Euclid is on 🤗 Hugging Face Also, this is the best paper heading I’ve seen in quite some time. The 'en tête' looks fantastic. (⚡Llama 3.3) Chat with the paper: huggingface.co/spaces/hugging… 🤗 Model: huggingface.co/euclid-multimo… 🤗 Dataset: huggingface.co/datasets/eucli… 🤗 Paper:

excited to share our first scientific foundation model release in collab with Willie Neiswanger Ollie Liu and more from USC.

[1/11] Many recent studies have shown that current multimodal LLMs (MLLMs) struggle with low-level visual perception (LLVP) — the ability to precisely describe the fine-grained/geometric details of an image. How can we do better? Introducing Euclid, our first study at improving

An awesome set of resources on LLM reasoning and test-time compute, compiled by @UpUpWang — check it out!

Tina: Tiny Reasoning Models via LoRA "the best Tina model achieves a >20% reasoning performance increase and 43.33% Pass@1 accuracy on AIME24, at only $9 USD post-training and evaluation cost (i.e., an estimated 260x cost reduction). Our work reveals the surprising effectiveness

😋 Want strong LLM reasoning without breaking the bank? We explored just how cost-effectively RL can enhance reasoning using LoRA! [1/9] Introducing Tina: A family of tiny reasoning models with strong performance at low cost, providing an accessible testbed for RL reasoning. 🧵

Presenting our spotlight paper on LLMs for decision making at ICLR 2026, Apr 25, 10–12:30PM, Hall 3 #113. Come say hi!

Is LoRA (Low Rank Adaptation) relevant in 2025 for reasoning models? I recently read "Tina: Tiny Reasoning Models via LoRA (arxiv.org/abs/2504.15777)", and it made me pause for a moment: when was the last time I heard someone excitedly talk/write about LoRA? LoRA (Low-Rank

Textual steering vectors can improve visual understanding in multimodal LLMs! You can extract steering vectors via any interpretability toolkit you like -- SAEs, MeanShift, Probes -- and apply them to image or text tokens (or both) of Multimodal LLMs. And They Steer!

Sparse autoencoders (SAEs) can be used to elicit strong reasoning abilities with remarkable efficiency. Using only 1 hour of training at $2 cost without any reasoning traces, we find a way to train 1.5B models via SAEs to score 43.33% Pass@1 on AIME24 and 90% Pass@1 on AMC23.