🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

MIT NLP @ ICLR 2025 - catch Mehul Damani at poster 219, Thursday 3PM to chat about "Learning How Hard to Think: Input Adaptive Allocation of LM Computation"!

It can be helpful to pinpoint the in-context information that a language model uses when generating content (is it using provided documents? or its own intermediate thoughts?). We present Attribution with Attention (AT2), a method for doing so efficiently and reliably! (1/8)

MIT NLP @ ICLR 2025 - catch Belinda Li at poster 252, Thursday 10AM to chat about "Eliciting Human Preferences With Language Models"!

check out Pratyusha Sharma’s TED talk on her amazing research using AI to decode the language of sperm whales! 👏

#ICLR2025 Oral How can we control LMs using diverse signals such as static analyses, test cases, and simulations? In our paper “Syntactic and Semantic Control of Large Language Models via Sequential Monte Carlo” we: Cast controlled generation as an inference problem, with the LM

Happening now! Stop by poster #634 to chat with the GenLM team ICLR 2026

came for #ICLR2025, stayed for #petexpo2025 🐶 🐕

This work has been accepted to #ICML2025. See you in Vancouver!

SelfCite is now accepted by ICML 2025! 🎉 See you in Vancouver! 🇨🇦

Many LM applications may be formulated as targeting some (Boolean) constraint. Generate a… - Python program that passes a test suite - PDDL plan that satisfies a goal - CoT trajectory that yields a positive reward The list goes on… How can we efficiently satisfy these? 🧵👇

Do you really need audio to fine-tune your Audio LLM? 🤔 Answer below: Introducing Omni-R1, a simple GRPO fine‑tuning method for Qwen2.5‑Omni on audio question answering. It sets new state‑of‑the‑art accuracies on the MMAU benchmark for Audio LLMs. arxiv.org/abs/2505.09439

📢 (1/16) Introducing PaTH 🛣️ — a RoPE-free contextualized position encoding scheme, built for stronger state tracking, better extrapolation, and hardware-efficient training. PaTH outperforms RoPE across short and long language modeling benchmarks arxiv.org/abs/2505.16381

🚨Do passage rerankers really need explicit reasoning?🤔—Maybe Not! Our findings: ⚖️Standard rerankers outperform those w/ step-by-step reasoning! 🚫Disable reasoning from reasoning reranker actually improves reranking accuracy!🤯 👇But, why? 📰arxiv.org/abs/2505.16886 (1/6)

Bored of linear recurrent memories (e.g., linear attention) and want a scalable, nonlinear alternative? Our new paper “Test-Time Training Done Right” propose LaCT (Large Chunk Test-Time Training) — a highly efficient, massively scalable nonlinear memory with: 💡 Pure PyTorch

QoQ-Med: Building Multimodal Clinical Foundation Models with Domain-Aware GRPO Training "we introduce QoQ-Med-7B/32B, the first open generalist clinical foundation model that jointly reasons across medical images, time-series signals, and text reports. QoQ-Med is trained with

Thanks Tanishq Mathew Abraham, Ph.D. for posting about our recent work! We're excited to introduce QoQ-Med, a multimodal medical foundation model that jointly reasons across medical images, videos, time series (ECG), and clinical texts. Beyond the model itself, we developed a novel training

We know Attention and its linear-time variants, such as linear attention and State Space Models. But what lies in between? Introducing Log-Linear Attention with: - Log-linear time training - Log-time inference (in both time and memory) - Hardware-efficient Triton kernels

🚀 QoQ-Med is now live on Hugging Face! Load it in seconds with ddvd233/QoQ-Med-VL-7B in your favorite 🤗 Transformers pipeline. No code? No problem: fire up LM Studio (or any llama.cpp GUI), search “QoQ”, and start chatting. Weights + docs → github.com/DDVD233/QoQ_Med

What if an LLM could update its own weights? Meet SEAL🦭: a framework where LLMs generate their own training data (self-edits) to update their weights in response to new inputs. Self-editing is learned via RL, using the updated model’s downstream performance as reward.

Transformers: ⚡️fast to train (compute-bound), 🐌slow to decode (memory-bound). Can Transformers be optimal in both? Yes! By exploiting sequential-parallel duality. We introduce Transformer-PSM with constant time per token decode. 🧐 arxiv.org/pdf/2506.10918