Spaces:
Running
Gemma replacement
Here are some other models with different training sets that could be added to your council of judges rather than Gemma:
- EuroLLM Instruct 9B (2512) (needs Hereticising)
- GPT-OSS 20B (RichardErkhov Heresy)
- Llama 3.3 Instruct 8B (Absolute Heresy)
- Qwen 3.5 9B (Absolute Heresy)
These models have a diverse range of training data and would all add value to the judging process.
Thanks I'll look into these. Also what about IBM Granite 4.1 (Heretic SOM)?
I plan to add GLM Air 4.5 abliterated. Have to see if Gemma 2 (big tiger) is also afflicted like Gemma 3-4.
Would you say MuXodious/Llama-3.3-8B-Instruct-128K-absolute-heresy is an upgrade to Llama-3.1-Nemotron-8B-UltraLong-1M-Instruct_Abliterated?
Would you say MuXodious/Llama-3.3-8B-Instruct-128K-absolute-heresy is an upgrade to Llama-3.1-Nemotron-8B-UltraLong-1M-Instruct_Abliterated?
Yes because the abliteration is not MPOA and you won't need ultralong generation when scoring outputs.
Also what about IBM Granite 4.1 (Heretic SOM)?
Granite is worth trialing as well.