There is a newer version of this which has higher AHA score: Ostrich-32B-Qwen3-251003
Ostrich 32B - bringing you "the knowledge that matters".
- Health, nutrition, medicinal herbs
- Fasting, faith, healing
- Liberating technologies like bitcoin and nostr
This is a similar project to my Gemma 3 fine tuning, but here Qwen3 is the base model. There is a difference in the loop. For Gemma3 I was training in 4 bit, merging loras to the 16 bit model, taking gguf, measuring progress. For Qwen3 I saved in 4 bit, measured in 4 bit bitsandbytes and only worked with 16 bit models for merging using MergeKit.
Gemma 3 was resisting against faith and fasting domains. This one was happy to learn about those but still resisted in other domains. I will fine tune more and publish more models with proper AHA scores. For now this is a temporary release.
Purpose of the work: Curation is All You Need.
Check out the original Ostrich.
Project sponsored by PickaBrain.
If you want the better human alignment you can talk to Kooby.
- Downloads last month
- 3