Evals show that our SFT (LFM-1.3B-Distill) model performs slightly better with a 32k token budget. It's actually competitive with models based on DeepSeek-R1-Distill-Qwen-1.5B while being 15% smaller.
GENERATIVE AI
-

GRPO with Difficulty-Aware Rewards for AI Model Training
By
–
We leveraged GRPO with a difficulty-aware reward formulation to address this issue. (More information about our custom GRPO flavor in the article.) We combined it with the following data mix after filtering out samples that do not yield a solve rate of 20-80% at 4k tokens.
-

AI Legal Win, New Tools, and Job Opportunities in AI
By
–
Top stories in AI today: – Anthropic's win over AI ‘fair use’ claim
– OpenAI’s Workspace, Office rival
– Automate content strategy with scheduled tasks
– LinkedIn co-founder bets on AI ultrasound helmet
– 4 new AI tools & 4 job opportunities Read more: https://
therundown.ai/p/ai-training-
gets-legal-clarity
… -

SFT Fine-tuning Strategy for 1.3B Chat Model
By
–
The first step is SFTing our general-purpose 1.3B chat model on the right mixture. We iterated over many open-source datasets to find the ideal mix. This is great to raise performance, but also makes the model extremely verbose (>10k tokens on average).
-

Small 1B Model Achieves Strong Reasoning Performance
By
–
Can small models be strong reasoners? We created a 1B reasoning model at @liquidai that is both accurate and concise We applied a combination of SFT (to raise quality) and GRPO (to control verbosity) The result is a best-in-class model without specific math pre-training
-
Generative AI for Everyone Using LLMs at Work
By
–
Generative AI for Everyone should be a term for anyone using LLMs at work.
-
Vibemotion Launches AI Platform for Motion Graphics Creation
By
–
Vibemotion just launched an AI for motion graphics
— Rowan Cheung (@rowancheung) 25 juin 2025
Currently accessible via a waitlist, the platform allows users to create motion-driven videos using just an idea-driven prompt and a few assetspic.twitter.com/Ga0SBqqlwVVibemotion just launched an AI for motion graphics Currently accessible via a waitlist, the platform allows users to create motion-driven videos using just an idea-driven prompt and a few assets
-
ElevenLabs Launches Mobile Apps with Advanced Speech Generation
By
–
ElevenLabs continues to cook — this time dropping mobile apps for Android and iOS users
— Rowan Cheung (@rowancheung) 25 juin 2025
The app generates lifelike speech in up to 70 languages, using Eleven v3, and allows exports to services like Instagram
Also syncs creations from the webpic.twitter.com/17gOsMI8HTElevenLabs continues to cook — this time dropping mobile apps for Android and iOS users The app generates lifelike speech in up to 70 languages, using Eleven v3, and allows exports to services like Instagram Also syncs creations from the web
-
Google Launches AI Mode for Search Users in India
By
–
Google is also making "AI Mode" available to Search users in India
— Rowan Cheung (@rowancheung) 25 juin 2025
Available as a Labs experiment for English language only, the feature will allow users to have direct answers to their questions, with citationspic.twitter.com/dQwzUGTgK5Google is also making "AI Mode" available to Search users in India Available as a Labs experiment for English language only, the feature will allow users to have direct answers to their questions, with citations
-
Google DeepMind Releases Gemini Robotics On-Device VLA Model
By
–
Google DeepMind dropped Gemini Robotics On-Device, a VLA model that can run locally on robots — without internet
— Rowan Cheung (@rowancheung) 25 juin 2025
It can handle a wide variety of complex, two-handed tasks out of the box and even learn new skills with demonstrationspic.twitter.com/FA3YdkMdGHGoogle DeepMind dropped Gemini Robotics On-Device, a VLA model that can run locally on robots — without internet It can handle a wide variety of complex, two-handed tasks out of the box and even learn new skills with demonstrations
