ricursiveAI/Ripple-65B-Preview
AI model · https://huggingface.co/ricursiveAI/Ripple-65B-Preview
First seen by Oriel on .
AI summary
Ripple-65B Preview is a 65B dense Transformer pretrained on 12 trillion tokens, claiming strong agentic coding results and up to 1M context while targeting efficient training and local deployment.
- The model scores 80.52 on TerminalBench 2.1, outperforming several listed US open-weight models with hundreds of billions more parameters, though it trails some competitors on other benchmarks.
- Its architecture has 80 layers and a 5120-wide d_model, with Gemma 3N-like design changes including gated attention, value residuals, and global NoPE.
- Pretraining used 12T predominantly English and code tokens; post-training included midtraining, long-context training, supervised fine-tuning, and mixed-domain reinforcement learning.
- The team reports 40% MFU using simple sharding, no custom kernels, and BF16; generation reached 250 tok/s on 2xB200 at batch size 8 before quantization or custom kernels.
- The BF16 model is provided alongside an NVFP4 checkpoint, and the 65B model is described as fitting on consumer hardware such as NVIDIA DGX Spark.
#large language models #transformer #agentic coding #model training
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Models: 607, latest observation ; +495 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Models: 11,012, latest observation ; +11,012 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | ricursiveAI/Ripple-65B-Preview | https://huggingface.co/ricursiveAI/Ripple-65B-Preview Read discussion |