🔥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

📢 (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

🚨 Don’t trust your ASR benchmarks. LibriSpeech and Common Voice are contaminated. Your SpeechLLM might be cheating—because it already saw the test answers during pretraining. Stop evaluating on leaked data. Here's why 1/n

Qwen benefits from random rewards on math 🧮 but that doesn't hold for general instruction following (IF). That said, noisy rewards aren't useless. In [🫐 arxiv.org/pdf/2505.11080], we show that simple string-matching metrics like BLEU can work surprisingly well for general IF!

This year, there have been various pieces of evidence that AI agents are starting to be able to conduct scientific research and produce papers end-to-end, at a level where some of these generated papers were already accepted by top-tier conferences/workshops. Intology’s

.Alibaba Group introduces TransBench, a benchmark designed to evaluate 🔍 how well #AI #translation systems perform in real-world industry settings — starting with international e-commerce 🛍️🌐 #xl8 #t9n #ecommerce Minghao Wu Chenyang Lyu 吕晨阳 Longyue Wang slator.com/alibaba-launch…

"𝘐𝘴 𝘈𝘊𝘓 𝘢𝘯 𝘈𝘐 𝘤𝘰𝘯𝘧𝘦𝘳𝘦𝘯𝘤𝘦?" A possible perspective is to examine: "𝘞𝘩𝘪𝘤𝘩 𝘈𝘊𝘓 𝘵𝘰𝘱𝘪𝘤𝘴 𝘳𝘦𝘴𝘰𝘯𝘢𝘵𝘦 𝘣𝘦𝘺𝘰𝘯𝘥 𝘢𝘤𝘢𝘥𝘦𝘮𝘪𝘢 𝘢𝘯𝘥 𝘤𝘢𝘱𝘵𝘶𝘳𝘦 𝘱𝘶𝘣𝘭𝘪𝘤 𝘢𝘵𝘵𝘦𝘯𝘵𝘪𝘰𝘯?" Our #ACL2025 main conf paper delves into this issue. (1/n)

🧵 MoDoMoDo: smarter data mixing → stronger reasoning for multimodal LLMs 🚀 New preprint! We show how right mixing of multi-domain data can super‑charge multimodal LLM RL reasoning. 🌐 Site: modomodo-rl.github.io 📄 Paper: arxiv.org/abs/2505.24871 details in thread. 👇

Anyone working on adaptive optimization methods and replacements for Adam should check this paper.

should be related to the pretraining data distribution across languages

ah this is interesting, language model is easily being contaminated by some data that's already included in the training process. I am curious about what if extending this to multilingual ASR setting especially involving some correlation analysis with the training data

✨Introducing ComfyUI-R1: a large reasoning model for automated workflow generation🌺 huggingface.co/papers/2506.09…

If any Irish speakers are interested in helping annotate some of the BLEnD examples to Irish, please let me know (can be as little as a few examples). We aim to have Irish included in the next release. huggingface.co/datasets/nayeo…

The Diffusion Duality "The arg max operation transforms Gaussian diffusion into Uniform-state diffusion" Adapts consistency distillation to diffusion language models, unlocking few-step generation by accelerating sampling by two orders of magnitude. Introduces a curriculum

PMF-CEC: Phoneme-augmented Multimodal Fusion for Context-aware ASR Error Correction with Error-specific Selective Decoding. arxiv.org/abs/2506.11064

Day 1/5 of #MiniMaxWeek: We’re open-sourcing MiniMax-M1, our latest LLM — setting new standards in long-context reasoning. - World’s longest context window: 1M-token input, 80k-token output - State-of-the-art agentic use among open-source models - RL at unmatched efficiency:

Hybrid attention enables efficient test-time compute scaling.

wow to see "Open"AI is in the second last position, surprised to see IBM is a major contributor, bravo! Qwen is the leading effort of China, same to DeepSeek. Hope open science will get more stronger.

V-JEPA 2 Self-Supervised Video Models Enable Understanding, Prediction and Planning

Check out our work on fair comparison among KV cache reduction methods and PruLong, one of the most effective, easy-to-use memory reduction method for long-context LMs!

good workshop, but why use this bad AI-generated picture with lots of meaningless symbols? It is destroying people's taste and sense for art and design.