My attempt at moving out of the Gemma 3 family!
Based on Qwen3 4B Instruct 2507, fine tuned on my AlteredDataset. It seems a lot weeker than my Gemma finetunes, but I'm sure I'll get there eventually.
Update: Maybe it won't get better. I'm pretty sure Qwen3's refusals are baked in way longer than Gemma's. I'll try finding more models around the same parameter count to try out soon. In the meantime, this repository has been renamed to Oryza-Spontanea-4B, on account of it looking and acting like weedy rice. It acts like my Gemma finetunes on the surface but actually has terrible traits. 10 hours of A100 Colab credits down the drain.
Uploaded finetuned model
- Developed by: DrRiceIO7
- License: apache-2.0
- Finetuned from model : unsloth/qwen3-4b-instruct-2507-unsloth-bnb-4bit
This qwen3 model was trained 2x faster with Unsloth and Huggingface's TRL library.
- Downloads last month
- 2