🏗️ Building on HF

Boning Cui

Bc-AI

AI & ML interests

He/Him. I like LLM's and VLM's. I work with my other friends to make stuff. We are in year 7 and we are enthusiastic about AI. We are based in Australia 🇦🇺

Recent Activity

new activity about 4 hours ago

Novi-AI/Novi-510x:Hmmmm

reacted to SeaWolf-AI's post with 🤝 about 6 hours ago

💻 Data-center AI, now on a laptop: POCKET-Darwin-180B We're releasing a 4-bit GGUF build of Darwin-180B-RSI, #1 on seven official Hugging Face leaderboards (self-reported), that runs without a GPU. 📦 360 GB → 111 GB (4-bit GGUF, 4 files) 🖥️ No GPU: one server CPU (16 threads) at 18.4–21.0 tokens/s 💻 RTX 5060 laptop (8 GB VRAM) + 32 GB RAM: 4.17 tokens/s 🧊 128 GB mini PC: whole model in memory, no GPU needed 🎯 MMLU-Pro, 2,000 questions, paired: original 87.65% = 4-bit 87.65% How? · Only ~3B of 180B parameters are active per token (10 of 512 experts) · llama.cpp streams just the needed experts from SSD, so 32 GB RAM is enough · Graft quantization: we took the proven Unsloth UD-Q4_K_XL base build and swapped in only the 300 tensors our RSI training changed (300/300 verified) Under the hood is Model-level Recursive Self-Improvement. The model solves verifiable problems, keeps only its own solutions that check out as correct, and trains on them. No human-written solutions or reasoning traces. Built for teams that can't send data to an external cloud (defense, finance, public sector) to run a top-tier model fully offline. 📝 Article: https://huggingface.co/blog/FINAL-Bench/data-center-ai-now-on-a-laptop-pocket-darwin-180b 🤗 Model: https://huggingface.co/FINAL-Bench/POCKET-Darwin-180B-GGUF 🧬 Original: https://huggingface.co/FINAL-Bench/Darwin-180B-RSI #Darwin #RSI #GGUF #llamacpp #OnDevice #MoE

reacted to SeaWolf-AI's post with ➕ about 6 hours ago

View all activity

Organizations

$Refract AI Labs's profile picture$

Hello everyone! 👋

A small SmilyAI Labs update!

G1-MINI has now seen around 8B tokens during its current run, and pretraining is still going strong.

Our E1 (Efficiency-1) prototype has also reached 15B pretraining tokens. E1 has 1B total parameters while activating under 100M parameters per token. It features adaptive activation, meaning easier tokens can use less compute while harder tokens receive more.

We plan to open-source E1 ASAP! 🚀

We’re also excited to announce Project Prism, which will provide limited access to our upcoming Orion Flagship model, powered by our T2 architecture.

Note: T2 here refers to the architecture, not our T2 (Thinker-2) model.

Applications for Project Prism are available through the org page, with more details coming soon!

Finally, welcome @soyL061215 , who joined the Hugging Face org today! 🎉

Thanks to our existing members:
@smilyai-large-team @MUK-IS-GOAT @Keeby-smilyai @Bc-AI

— Bc-AI
SmilyAI Labs

Post

180

Hello Everyone!
I am happy to announce a few things.
1. G1-MINI
G1-MINI is now in pretraining and is training at a steady pace. Our current ETAs state completion and launch in about 15-20 days, somewhere near the end of September.
2. G1-NANO
G1-NANO is also being pretrained as we speak at a pace of over 400K tokens per second processing more than 10B tokens in 12 hours. This allows us to train extremely fast, and we will launch it somewhere around 15th September.
3. We have begun work on FrameShot, a dual image and video generation model at around 4B dense parameters. This is expected to launch around late December with no promised date.
- Bc-AI