Jesus to Political [email protected] • 3 months agoWhat could possibly go wronglemmy.worldimagemessage-square68fedilinkarrow-up1547arrow-down149
arrow-up1498arrow-down1imageWhat could possibly go wronglemmy.worldJesus to Political [email protected] • 3 months agomessage-square68fedilink
minus-square@[email protected]linkfedilinkEnglish56•3 months agoIt’s open source. Apparently folks have already made mods of it that add CCP-sensitive info back in. Disclaimer: I have yet to see this for myself.
minus-square@[email protected]linkfedilinkEnglish52•3 months agoThe answer I got out of DeepSeek-R1-Distill-Llama-8B-abliterate.i1-Q4_K_S
minus-square@[email protected]linkfedilink28•3 months agoSo a real answer, basically. Too bad your average person isn’t going to bother with that. Still nice it’s open source.
minus-square@[email protected]linkfedilink11•3 months agoSeems like the model you mentioned is more like a fine tuned Llama? Specifically, these are fine-tuned versions of Qwen and Llama, on a dataset of 800k samples generated by DeepSeek R1. https://github.com/Emericen/deepseek-r1-distilled
minus-square@[email protected]linkfedilinkEnglish8•edit-23 months agoYeah, it’s distilled from deepseek and abliterated. The non-abliterated ones give you the same responses as Deepseek R1.
minus-square@[email protected]linkfedilink3•3 months agojust running it locally, apparently - the output of this model is being filtered by another AI, but only on the public hosted copy.
minus-squarestebolinkfedilink2•3 months agoif it’s open source, can we also see what words/topics are being blocked?
It’s open source. Apparently folks have already made mods of it that add CCP-sensitive info back in. Disclaimer: I have yet to see this for myself.
The answer I got out of DeepSeek-R1-Distill-Llama-8B-abliterate.i1-Q4_K_S
So a real answer, basically. Too bad your average person isn’t going to bother with that. Still nice it’s open source.
Seems like the model you mentioned is more like a fine tuned Llama?
https://github.com/Emericen/deepseek-r1-distilled
Yeah, it’s distilled from deepseek and abliterated. The non-abliterated ones give you the same responses as Deepseek R1.
deleted by creator
just running it locally, apparently - the output of this model is being filtered by another AI, but only on the public hosted copy.
if it’s open source, can we also see what words/topics are being blocked?