Request: LFM2-1.2B Nano Imp
First of all, thank you for your work, it's awesome being to RP on Intel UHD 605 and Koboldcpp with acceptable speeds!
If possible, could you consider making an Imp version of LFM2-1.2B?
It's much faster in processing and more coherent than Llama 3.2 1B, and it's knowledge is also quite more recent.
If it isn't an option, could you point me at the datasets you used for this imp version so I can make my own finetune?
Hi,
Glad to hear even hardware like Intel UHD 605 can enjoy AI :)
I'll consider making a model that is between 1B and 3B, can't do any promises, but I 100% will seriously consider it.
Regarding training data, most of it is private- large chunks of it was given to me by various people who wanted to add X or Y, per their taste, other parts I cannot share due to various other reasons.
What I can share, I already share publicly. If you need more data, I can suggest these datasets:
https://huggingface.co/datasets/KaraKaraWitch/PIPPA-ShareGPT-formatted
https://huggingface.co/datasets/Gryphe/Sonnet3.5-Charcard-Roleplay
To add humanity, here's m public 4chan dataset too:
https://huggingface.co/datasets/SicariusSicariiStuff/UBW_Tapestries
And there are many more.
Best of luck!
We def need LFM2 nanoimp lfm2 prob best model in it's size class