mradermacher

/

model_requests

like 100

It's queued! :D
Qwen 3 based fully uncensored roleplay and storytelling model ignoring any kind of moral or ethical boundaries allowing the user full freedom to do whatever he desires no matter how messed up it is. What's really interesting is that they started finetuning from Qwen3 8B Base so this model truly is something special. The training for this one must have been really time consuming and expensive. @mradermacher Maybe worth to add on your list of models you want to try - it will likely to be quite a good at story writing and likely differ quite a lot from most similar themed models due to being based on Qwen3 8B Base and trained which such an extensive dataset.

You can check for progress at http://hf.tst.eu/status.html or regularly check the model
summary page at https://hf.tst.eu/model#Bald-Beaver-8B-GGUF for quants to appear.

The compute for the fine-tune was graciously donated by my friends at Anthracite Org, the full fine-tune took just over 2 hours spread across 8x H100. It actually took longer to perform the KTO using 2x H200 at nearly 5 hours for a single epoch. There is nothing inherently special about this model, it is a standard training formula just on an arch that famously refuses everything. I am happy to report that in my limited testing, this model responds as expected when prompted in an intentional way.

@nicoboss an 8b??? well, if you say so, i will give it a try. should be fast at least :)