GM you can now try #DeepSeek R1 32B on Unleashed.chat without sending your data to CCP. Click on the arrow to expand and watch the reasoning if you'd like. All the data is stored locally on your browser, pay your use by lightning. Next is making it available as DVM, API, then let it search Nostr, search the web and make it faster. Follow @dataMachine πŸ€– for more updates. Promising!

Replies (37)

That's not really Deepseek R1, it's a distilled version of Alibaba's Qwen-32B architecture, enhanced using synthetic outputs from the larger DeepSeek R1 model. Quite useful but not hte same thing.
LL62's avatar
LL62 10 months ago
I'll wait for the DVM πŸ€ πŸ€™
Johnathan Corgan's avatar
Johnathan Corgan 10 months ago
It *is* r1, which is the name for the distilled version as you describe. The bigger model is called v3.
Jakob's avatar
Jakob 10 months ago
Awesome! Where do you host the model?
Installed 1.5B locally today, but it was trash; currently pulling 7B. Will test 7B for an hour or so tonight, then upgrade to 70B overnight. Seems like 70B will probably become the defacto "baseline" model moving forward. πŸ€™
c12's avatar
c12 10 months ago
Is this even legal for US citizens? πŸ˜„
πŸ‘€
DETERMINISTIC OPTIMISM 🌞's avatar DETERMINISTIC OPTIMISM 🌞
GM you can now try #DeepSeek R1 32B on Unleashed.chat without sending your data to CCP. Click on the arrow to expand and watch the reasoning if you'd like. All the data is stored locally on your browser, pay your use by lightning. Next is making it available as DVM, API, then let it search Nostr, search the web and make it faster. Follow @dataMachine πŸ€– for more updates. Promising!
View quoted note →
Analogue Dog's avatar
Analogue Dog 10 months ago
"without sending your data to the CCP" πŸ€”
it's not black-and-white. Also, not possible to do this client side with this speed yet. Unless you have the hardware, expensive hardware most people won't have. There is better ways, there is shit ways, and there is scam ways.
↑