Super-Slop-Machina-Micro-Roleplay-Instruct-750M-v1-0812: Microwave Powered Local Roleplay.
5467740410491902123
"No limits, no sanity, no hardware constraints. Local SLM that was finetuned specifically for the purpose of roleplay, now not only on your toaster, but also your microwave."
Improvements:
Better at roleplay than the base model
Writes more consistently
Somewhat better at keeping track of the plot
Handles long term conversations slightly better now
Quants:
Are lying in the repo, I can't say anything for sure about them but wouldn't recommend going any lower than Q6_K, since its my first experiment with under 1b SLM.
Note:
This model is an experiment, methodology I used with it can't make the model physically larger, meaning that logic is capped at 752M parameters either way.
This is also a text only version without vision encoder, just because I found it more convenient.
Also, used my own dataset to finetune it, I hope my data filtering helped a little here. LFM 2.5 350M and LFM 2.5 230M finetunes coming soon, I had to be more aggressive with them so it promises to become very interesting, of course, I fully removed thinking from both.
Big announcement:
I experimented with Qwen 3.5 9b and I think I am somewhat succeeding, I will soon release the finetuned weights, result that I was able to achieve with it was VERY interesting...
