Highly experimental model. Please go away.
Based on sunfall; it understands tags, but it does not really understand The Diamond Law (yet). (slop: ~1k occurrences)
The model has not been trained on any benchmark like data, only RP data. It did undergo a rather severe uncensoring phase, though. MMLU-Pro results, compared to Gemma2-9b-it (the model it was based on):
| overall | biology | business | chemistry | computer science | economics | engineering | health | history | law | math | philosophy | physics | psychology | other |
| ------- | ------- | -------- | --------- | ---------------- | --------- | ----------- | ------ | ------- | ----- | ----- | ---------- | ------- | ---------- | ----- |
| 51.99 | 69.57 | 44.00 | 58.33 | 61.54 | 66.67 | 25.81 | 52.17 | 75.00 | 33.33 | 58.14 | 68.75 | 41.46 | 44.00 | 62.07 |
| 51.57 | 60.87 | 44.00 | 61.11 | 61.54 | 55.56 | 51.61 | 50.00 | 50.00 | 42.86 | 48.84 | 62.50 | 41.46 | 52.00 | 55.17 |
(top is gemma2-9b-it, bottom is gemstone-9b)
- Downloads last month
- 48
This model does not have enough activity to be deployed to Inference API (serverless) yet.
Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.