strategic-personas-zh-r8-gemma-4-e4b-merged
Chinese mixed persona and style model based on Gemma 4 E4B, merged from a rank-8 LoRA adapter trained at max_length 1536.
This is a merged Transformers model based on google/gemma-4-E4B-it plus a PEFT LoRA adapter. It is intended for experimental persona and style transfer evaluation.
Intended Use
- Text-generation experiments with Chinese and English prompts.
- Server-side validation with Hugging Face Transformers.
- Conversion experiments for mobile runtimes such as LiteRT-LM.
Source
- Base model:
google/gemma-4-E4B-it
- Adapter source path on training host:
/home/ubuntu/gemma4_e4b_mixed_zh_lora_1536_final_resume_nohup_may11
- Merge format: full Transformers safetensors model with tokenizer/processor files.
Notes
This repository is not published by Google. Validate model quality, safety, and deployment behavior for your own use case.
Gemma is a trademark of Google LLC.
