Pinned
Run GLM 5.2 with only 192 GB ram 🫰
Introducing a REAP50-Q3 quant by Pipe Network.
Reduced 50% in size by keeping half of the most-salient experts per layer.
Download on @huggingface: huggingface.co/pipenetwork/GL…
Pipe Network
1,620 posts
Pipe Network
@pipenetwork
Hyperlocal nodes powering CDN, storage, and accelerated transport. Closer is faster.
Joined April 2022
- Pipe Network reposted
Run GLM 5.2 with only 192 GB ram 🫰 Introducing a REAP50-Q3 quant by Pipe Network. Reduced 50% in size by keeping half of the most-salient experts per layer. Download on @huggingface: huggingface.co/pipenetwork/GL… - Replying to @pipenetworkQuantized to Q3_K_M (~169 GB), you can run it with llama.cpp on: - Mac Studio with 192GB+ unified memory - 2× RTX Pro 6000 - 6× RTX 5090 - 8× RTX 3090
- Fifty one near-lossless models available for download from Pipe. What are you building today?
4/5/6 bit quants are up, 8-bit is uploading 🫡- You take the blue pill, the story ends, you go back to Claude and hope Fable is back soon. You take the red pill, you run Kimi K2.7 MLX quants by Pipe Network, and have private frontier access 👇
- Sovereign intelligence for everyone. Introducing MLX quants for @GoogleAI Gemma 4. Run on Apple Silicon with near-lossless quants from Pipe Network 🧵
00:00Available in 4/5/6/8 bit variants on @huggingface: huggingface.co/pipenetwork/co… Gemma 4 31B collection: huggingface.co/collections/pi… Gemma 4 26B A4B collection: huggingface.co/collections/pi… - By popular request, @MiniMax_AI M3 MLX quants are now live Get the top open-weights coding & agentic model for Apple Silicon 👇
00:00 - Fix bugs like a boss. MLX Quants of @Microsoft FrogBoss and FrogMini from Pipe Network are now live.
00:00 - The future is AI on Solana
- Faster compute is the currency Pipe nodes are 🔒'ed in
