Google just dropped a new light-weight 270M model! Gemma-3-270M is a lightweight, open-weight LLM that's perfect for task-specific fine-tuning with strong instruction-following. This notebook explains how to build Gemma-3-270M from scratch using PyTorch, step by step.
LLMS
-
XQuant: KV Cache Rematerialization Breaks LLM Memory Limits
By
–
XQuant Breaking the Memory Wall for LLM Inference with KV Cache Rematerialization
-
Altman Announces GPT-5 Personality Customization Feature
By
–
Altman annonce un changement majeur pour GPT-5. La personnalisation de GPT-5. Chaque personne pourra donner la personnalité qu'il souhaite à son chatGPT Super !
-
Comparing Prompt Responses Across Different AI Models
By
–
All 6 took the same prompt, but delivered in different ways. Which AI builder do you think nailed it?
-
Video-RAG: Training-Free Retrieval for Long-Video LVLMs
By
–
📢☑️Video-RAG: Training-Free Retrieval for Long-Video LVLMs
— Satya Mallick (@LearnOpenCV) 18 août 2025
In this week’s deep dive, we implement Video-RAG as a training-free, single-pass pipeline and integrate it with LLaVA-Video-7B (Qwen2, 32K context), without APE – to keep things reproducible on today’s stacks. We enable… pic.twitter.com/ZVNMVXr0WQ-RAG: Training-Free Retrieval for Long-Video LVLMs In this week’s deep dive, we implement Video-RAG as a training-free, single-pass pipeline and integrate it with LLaVA-Video-7B (Qwen2, 32K context), without APE – to keep things reproducible on today’s stacks. We enable
-
LFM2-VL: New Open-Source Vision Language Models Released
By
–
Demo and code made by Tarek Dakhran. Credit to @ngxson for his incredible help with llama.cpp, and to Anna Banaszak for leading the VLM project! LFM2-VL-1.6B-GGUF: https://
huggingface.co/LiquidAI/LFM2-
VL-1.6B-GGUF
… LFM2-VL-450M-GGUF: -
LFM2-VL Support Added for GGUF and llama.cpp
By
–
LFM2-VL support with GGUF and llama.cpp 🥳
— Maxime Labonne @ ICLR (@maximelabonne) 18 août 2025
You can now run these tiny, hyper-efficient VLMs on your watch!
We released quantized checkpoints for LFM2-VL-450M and LFM2-VL-1.6B on @huggingface pic.twitter.com/DfK3p1MA08LFM2-VL support with GGUF and llama.cpp You can now run these tiny, hyper-efficient VLMs on your watch! We released quantized checkpoints for LFM2-VL-450M and LFM2-VL-1.6B on @huggingface
-
Knoll’s Law: Media and LLM Information Reliability
By
–
I think this is partly due to Knoll’s law, which we could restate as “everything you read in the newspapers (or from LLMs) is absolutely true, except for the rare story (or chat) of which you happen to have firsthand knowledge (or deep expertise)”.
-
Personal AI Tutor: Making Learning Easier for Everyone
By
–
A personal AI tutor could make learning so much easier.
-
Databricks Updates: SharePoint, LLM Data Exploration, Data Lineage
By
–
New episode of Over Architected has dropped!
— Databricks (@databricks) 17 août 2025
Nick Karpov and Holly Smith explore the latest feature updates to Databricks, and try to fit them all into a single architecture:
– Microsoft SharePoint connector
– Explore table data using an LLM
– Bring your own data lineage
–… pic.twitter.com/ihloZSTUcGNew episode of Over Architected has dropped! Nick Karpov and Holly Smith explore the latest feature updates to Databricks, and try to fit them all into a single architecture:
– Microsoft SharePoint connector – Explore table data using an LLM – Bring your own data lineage
–