Testament200156/old-MakeGemma3-abliterated
Testament200156/old-MakeGemma3-abliterated is a 27 billion parameter experimental language model created by Testament200156 through a NuSLERP merge of drwlf/medgemma-27b-it-abliterated and summykai/gemma3-27b-abliterated-dpo. This model, with a 32768 token context length, is an early iteration of the MakeGemma3 series, noted for its experimental nature and potential defects in multilingual functionality. It serves as a foundational merge for further development, with a more refined multilingual version available under Testament200156/MakeGemma3-abliterated.
Loading preview...
Model Overview
Testament200156/old-MakeGemma3-abliterated is a 27 billion parameter experimental language model developed by Testament200156. This model was created using the NuSLERP merge method via mergekit, combining two base models: drwlf/medgemma-27b-it-abliterated and summykai/gemma3-27b-abliterated-dpo (with additional layers). It operates with a context length of 32768 tokens.
Key Characteristics
- Experimental Nature: This model is an early, experimental version of the MakeGemma3 series.
- Merge Method: Utilizes the NuSLERP merging technique for combining model weights.
- Base Models: Merges a medical-domain fine-tuned model (
medgemma) with a DPO-trained Gemma3 variant. - Known Limitations: The model has identified defects, particularly in its multilingual functionality, due to its experimental status.
Usage Considerations
This model is primarily intended for users interested in exploring early merged iterations or for specific use cases where its experimental nature and potential multilingual limitations are understood. For improved multilingual capabilities, users are advised to refer to the more recent and refined version, Testament200156/MakeGemma3-abliterated.