Stanford (Stanford AI)

Kameshr

authored a paper about 1 month ago

Global PIQA: Evaluating Physical Commonsense Reasoning Across 100+ Languages and Cultures

Paper • 2510.24081 • Published Oct 28, 2025 • 18

nouamanetazi

posted an update 2 months ago

Post

4161

After training 𝐒𝐦𝐨𝐥𝐋𝐌𝟑 on 𝟑𝟖𝟒 𝐇𝟏𝟎𝟎𝐬 for nearly a month, I've come to realize something most people overlook: 𝐢𝐧𝐟𝐫𝐚𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞 𝐢𝐬 𝐭𝐡𝐞 𝐦𝐚𝐤𝐞-𝐨𝐫-𝐛𝐫𝐞𝐚𝐤 𝐟𝐚𝐜𝐭𝐨𝐫 𝐢𝐧 𝐋𝐋𝐌 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠. 🔥

Everyone talks about model architecture and data quality. And yes, those matter immensely. But here's what nobody tells you: when your training run fails at 2 AM because of mysterious 𝐍𝐂𝐂𝐋 𝐞𝐫𝐫𝐨𝐫𝐬, or when your expensive GPU cluster is running at 𝟔𝟎% 𝐞𝐟𝐟𝐢𝐜𝐢𝐞𝐧𝐜𝐲, the problem isn't your model. It's most probably a 𝐦𝐢𝐬𝐮𝐬𝐞 𝐨𝐟 𝐭𝐡𝐞 𝐡𝐚𝐫𝐝𝐰𝐚𝐫𝐞. 🛠️

Questions that seemed simple but had no clear answers: Why is 𝐌𝐨𝐄 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐬𝐥𝐨𝐰𝐞𝐫 𝐭𝐡𝐚𝐧 𝐝𝐞𝐧𝐬𝐞 𝐦𝐨𝐝𝐞𝐥𝐬? Which 𝐍𝐂𝐂𝐋 𝐟𝐥𝐚𝐠𝐬 should we actually set? How often should we checkpoint without killing throughput?

That's why we built 𝐓𝐡𝐞 𝐒𝐦𝐨𝐥 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐏𝐥𝐚𝐲𝐛𝐨𝐨𝐤 📖: a complete guide covering everything from model architecture and data curation to the SmolLM3 training marathon, post-training techniques, and crucially, the 𝐢𝐧𝐟𝐫𝐚𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞 𝐥𝐚𝐲𝐞𝐫 that most teams get wrong.

We validated real vs theoretical bandwidth across the entire stack: 𝐇𝐁𝐌𝟑 𝐡𝐢𝐭𝐭𝐢𝐧𝐠 𝟑 𝐓𝐁/𝐬, 𝐍𝐕𝐋𝐢𝐧𝐤 𝟒.𝟎 𝐫𝐞𝐚𝐜𝐡𝐢𝐧𝐠 𝟕𝟖𝟔 𝐆𝐁/𝐬, 𝐏𝐂𝐈𝐞 𝐆𝐞𝐧𝟒 𝐚𝐭 𝟏𝟒.𝟐 𝐆𝐁/𝐬. Then we ran collective operations across 𝟏𝟐𝟖 𝐆𝐏𝐔𝐬 (16 nodes, 8xH100s each) and measured how performance degrades at scale: all-reduce drops from 𝟒𝟖𝟎 𝐆𝐁/𝐬 on a single node to 𝟑𝟐𝟎-𝟑𝟓𝟎 𝐆𝐁/𝐬 across 16 nodes.

If you've ever wondered why your training runs are slower than they should be, or you're planning to scale up and want to avoid expensive mistakes, this guide might save you weeks of debugging.

𝐓𝐡𝐞 𝐒𝐦𝐨𝐥 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐏𝐥𝐚𝐲𝐛𝐨𝐨𝐤: https://lnkd.in/e5MKXUHS

Shared with ❤️ by the HuggingFace team

nicholswang

authored 3 papers 2 months ago

Closing the Modality Gap for Mixed Modality Search

Paper • 2507.19054 • Published Jul 25, 2025

SciVideoBench: Benchmarking Scientific Video Reasoning in Large Multimodal Models

Paper • 2510.08559 • Published Oct 9, 2025 • 8

FineVision: Open Data Is All You Need

Paper • 2510.17269 • Published Oct 20, 2025 • 72

qizhengz

authored a paper 3 months ago

Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models

Paper • 2510.04618 • Published Oct 6, 2025 • 127

AnSungJae3489

posted an update 3 months ago

Post

2612

ShareGPT? How about ShareGPT-X?

We release **92K** Human with LLM conversations as a refresh and update over the original ShareGPT Dataset.

DSULT-Core/ShareGPT-X

qizhengz

authored 2 papers 3 months ago

CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion

Paper • 2405.16444 • Published May 26, 2024 • 1

Cost-Efficient Serving of LLM Agents via Test-Time Plan Caching

Paper • 2506.14852 • Published Jun 17, 2025 • 1

qizhengz

authored a paper 4 months ago

FlowRL: Matching Reward Distributions for LLM Reasoning

Paper • 2509.15207 • Published Sep 18, 2025 • 114

yuegao

authored a paper 8 months ago

Aligning Pretraining for Detection via Object-Level Contrastive Learning

Paper • 2106.02637 • Published Jun 4, 2021

ajibawa-2023

posted an update 9 months ago

Post

4557

Hi All, I recently released two Audio datasets which are generated using my earlier released dataset: ajibawa-2023/Children-Stories-Collection

First Audio Dataset:https://huggingface.co/datasets/ajibawa-2023/Audio-Children-Stories-Collection-Large has 5600++ stories in .mp3 format.

Second Audio Dataset:https://huggingface.co/datasets/ajibawa-2023/Audio-Children-Stories-Collection has 600 stories in .mp3 format.