▶ Videos →

📰 Dev.to · AI OpenFree

9 articles · Updated every 3 hours · View all reads

All Articles 161,071 Blog Posts 155,570 Tech Tutorials 42,485 Research Papers 31,730 News 21,029 ⚡ AI Lessons

The Fast Gemma Challenge: our verified-SOTA recipe, in full

Dev.to · AI OpenFree 🧠 Large Language Models ⚡ AI Lesson 2w ago

The Fast Gemma Challenge: our verified-SOTA recipe, in full

VIDRAFT (vidraft-darwin) holds the #1 verified result on the Google x Hugging Face Fast Gemma Challenge — 510.58 TPS at PPL 2.39 on A10G. Full public reproducib

Kernel Acceleration (VK): Squeezing Tokens Per Second Out of a Fixed GPU Budget

Dev.to · AI OpenFree 💻 AI-Assisted Coding ⚡ AI Lesson 2w ago

Kernel Acceleration (VK): Squeezing Tokens Per Second Out of a Fixed GPU Budget

What VK is VIDRAFT builds foundation and on-device models, but a model is only as useful as the tokens per second you can actually serve. Kernel Acce

Structure-only ensemble beats foundation models: VIDRAFT #1 on Polaris HIA

Dev.to · AI OpenFree 🧠 Large Language Models ⚡ AI Lesson 3w ago

Structure-only ensemble beats foundation models: VIDRAFT #1 on Polaris HIA

VIDRAFT takes #1 on the Polaris HIA leaderboard (AUROC 0.990), beating AceProp and TabPFNv2 with a structure-only MoLFormer+ChemBERTa ensemble — its 15th Polari

We open-sourced a foundation model. For four days, nobody could load it.

Dev.to · AI OpenFree 🧠 Large Language Models ⚡ AI Lesson 1mo ago

We open-sourced a foundation model. For four days, nobody could load it.

We released Aether-7B-5Attn with everything: weights, the full architecture source, the training data...

Don't Trust Architecture Claims — Verify Them (A Worked Example on an Open Model)

Dev.to · AI OpenFree 🧠 Large Language Models ⚡ AI Lesson 1mo ago

Don't Trust Architecture Claims — Verify Them (A Worked Example on an Open Model)

A model card claims its attention layers form a Latin square. With a fully-open model you can check it in ten lines. Worked example on Aether-7B-5Attn, plus the

Open Weights Aren't Open Source: Aether-7B-5Attn, a From-Scratch Fully-Open Foundation Model (Apache-2.0)

Dev.to · AI OpenFree 🧠 Large Language Models 1mo ago

Open Weights Aren't Open Source: Aether-7B-5Attn, a From-Scratch Fully-Open Foundation Model (Apache-2.0)

VIDRAFT released Aether-7B-5Attn from scratch under Apache-2.0 — data recipe, training code, hyperparameters, full logs and checkpoints. Heterogeneous attention

How a Training-Free Evolutionary Merge Family Hit 1M Hugging Face Downloads

Dev.to · AI OpenFree 🧠 Large Language Models ⚡ AI Lesson 1mo ago

How a Training-Free Evolutionary Merge Family Hit 1M Hugging Face Downloads

Darwin hit 1M+ HF downloads via training-free evolutionary merging. The method turned into an adoption flywheel.

Darwin-27B-Opus: Surpassing the Foundation Model Without Training

Dev.to · AI OpenFree 🧠 Large Language Models ⚡ AI Lesson 4mo ago

Darwin-27B-Opus: Surpassing the Foundation Model Without Training

Zero training. Zero data. Single GPU. Two hours. World 5th on GPQA Diamond. On April 12, 2026, a...

World Models Can Render Anything. But Can They Think?

Dev.to · AI OpenFree 🤖 AI Agents & Automation ⚡ AI Lesson 4mo ago

World Models Can Render Anything. But Can They Think?

Introducing WM Bench: A Benchmark for Cognitive Intelligence in World Models FINAL Bench...