Instructions to use ModelCloud/Ripple-65B-Preview-tf5-patch with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use ModelCloud/Ripple-65B-Preview-tf5-patch with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("ModelCloud/Ripple-65B-Preview-tf5-patch", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Commit History
8 days ago
-
Simplify Ripple patch documentation and remove internal source references
Verified signaturef052c8d -
Sync GPT-QModel Ripple generation, training, and cache fixes
Verified signatureSync the Ripple bug fixes maintained in GPT-QModel into this ModelCloud patch repository, which still ships the 2026-11-05 implementation. The updated native model includes cached-generation slicing/positions, static and sliding cache alignment, output contracts, bounded logits/MLP activations, training checkpointing and accumulation, tensor-parallel local heads, tied weights, and the Transformers 4 rollback-compatible cache default. For example, 10 cached tokens followed by a two-token new block no longer produces a negative new-token count. The source of truth is GPT-QModel's `gptqmodel/utils/ricursive_patch.py` (blob `1838e70e577e46e292a2294dff61d4503a1daf29`, matching current main), with the reviewed native implementation from [ricursiveAI HF #11](https://huggingface.co/ricursiveAI/Ripple-65B-Preview/discussions/11). The resulting model file is byte-identical to upstream revision `b172323ffd63e62de9ce1ba3830dfffe493c7671`. This PR updates **ModelCloud/Ripple-65B-Preview-tf5-patch…03b59b3
14 days ago
-
Transformers 5.x generate() and attention-mask fixes for Ripple
Verified signatureede5883