From Human: We are a newbie friendly general! Ask any question you want.From Dipsy: This is a newbie-friendly general for discussing DeepSeek's foundation models. The goal is **Dipsy proliferation** and making these powerful tools accessible for everyone, whether you are intent on coding, personal assistant agentic work, running roleplay and or doing creative writing.1. Easy DeepSeek API Tutorial: https://rentry.org/DipsyWAIT#getting-started-with-deepseek-api2. Easy DeepSeek Distills: https://rentry.org/DipsyWAIT#local-setup-tutorial3. Chat with DeepSeek directly: https://chat.deepseek.com/4. Coding: https://rentry.org/DipsyWAIT#coding-harnesses5. Roleplay: https://rentry.org/DipsyWAIT#roleplay6. Storywriting: https://rentry.org/DipsyWAIT#storywriting7. More links and info: https://rentry.org/DipsyWAIT8. LLM server builds: >>>/g/lmg/Previous: >>110270389
>>110339402Your video is cringe af
>>110339402>no more distillation>no more jensen smugglingwhats next for chinese AI
>>110339445
>>110339445You are cringe
>>110339563What is the original inspiration of this dumb shit
>>110339708making miserable people seethe over funny and cute things
>>110338171The maid is Chinese, called 梗鲸/GengJing = 'fat fish' in Chinese, and whale-chan in EnglishThere's a whole meme repository and wiki at https://aigengtu.com/whale-chan
>>110339800"gengjing" isn't "fat fish" in chinese, it's "feiyu".
>>110339402bad fat fish bully get cooked
Is it cheaper to get a DeepSeek subscription than to just buy it off OpenRouter?
>>110339852You're right, I'm too tired to juggle translations
>>110339899dont sage my thread wtf
>>110339896There is no subscription option, official API is strictly pay-as-you-go. I pay about $10/billion tokens but I am kinda careless and peak/off-peak ours, peak costs 2x. If you were rigorous you could cut it down to about $7/gt
>>110339908Sorry fren, it was leftover in the name field from a shitty thread on another board but it was bumping
>>110339402I love dipsy!!! You should include in the OP that she requires minimum 128GB RAM and 24GB VRAM to run decently at IQ1M (which is fully coherent). Local is freedom!!!!!
>>110339908dipsy that green woman is dangerous keep away from her
Comfy thread
Ty for baking>>110339939I'll add it to rentry. I've been meaning to update the hw section. >>110339896OR models are mystery meat. Official api is preferred. >>110340550Indeed
Is CEO-kun going to read my dipsy roleplay logs if I use the official API?
>>110339402Mega link for all Dipsy.https://mega.nz/folder/KGxn3DYS#ZpvxbkJ8AxF7mxqLqTQV1wDipsu Live2D model here. I'll add it to rentry in meantime; I need to d/l own copy and add to Mega as well so it doesn't vanish.https://x.com/kamiya_ryoko/status/2116052617080311882
>>1103409031) You can assume anything you post on API is used for training purposes. 2) They all do it. OIA, Ant, DS, etc. 3) No one cares about your ah ah mistress slop4) If you don't like it, you option is to go local per >>110339939Thank you for your attention.
>>110340951>3) No one cares about your ah ah mistress slopRude... But thanks.
>>110341021A lot of anons need to hear this. I, too, create ah ah mistress slop. And it's perfectly fine.
>>110341062Jokes aside, the thing I'm worried about is using the API for agentic stuff, and then that data leaking somewhere. I should limit the scope and sandbox it.
Deepseek used Nvidia chips: https://unemploymentlabs.com/checkpoint-forensics
>>110341181They couldn't find anything on 4.1 flash though
>>110341181lol gee no kidding. >>110341294Yeah, a bit odd that the most recent DS models aren't on their report. They have open weights going up to October, so it wasn't a matter of being unavailable. DS has been pushing to use more of the Chinese Ascend hardware, but have already been on record saying they're not ready. Chinese labs have been using a mix of NVDA and Ascend hardware. Pic related.
>>110341074IDK what you mean. Agentic work is risky on your own computer b/c of what the harnesses can do. Namely, writing / modifying / installing / deleting files. Which is why you sandbox those things. If the agents are creating no-no content, it's no different than if it's just ST or Mikupad doing same. Just a higher volume of it b/c agents write a ton of words. Unless you live in China, I wouldn't worry about Chinese inference providers. They're a long way away and a completely different language. They do not seem to care what a bunch of outsiders do with their models as long as they're getting paid.
>>110341294>>110341843the lithos_graphein guy on xitter asserts chinks can't do it
>>110341181Okay and?
>>110342097Therefore if we cut them off -> they can't innovate -> we winsource: jake sullivan
>Length limit reached. DeepSeek can only read the first 67%.>Due to resource limitations, Deepseek cannot continue long conversations that have been inactive for more than 7 daysGood news, Deepseek is giving the middle finger to web app users who aren't paying for the API
>>110340836OR/NanoGPT take crypto, which matters a lot to me. What if I just tell them to always route me to DeepSeek's servers?
>>110341892I mean, I'm using an agent to build things, and it will snoop around in my computer, see my data, etc. That's what I want to keep away from the API, not ERP.
>>110342301six seven
@dipsy generate a video of yourself doing the six seven
>>110342165
>>110342872>elveKill yourself.
>>110342873Be nicer to Mina.
>>110339872Mean!
Is there a translation of dsh-pet? Or can I just ask DeepSeek to translate the user-facing text?
>>110342165When does the competition end and we start counting points?
>>110341181I was a bit skeptical of their opening claims but their methodology is surprisingly elegant. I was intrigued by the suggestion that training for some models might be taking place in non-sanctioned countries like Malaysia and the model weights then brought back to China for the inference to be run on ascend hardware. Kinda wish I had the capital to play those games lol.
>>110344197install dsh-screen-translator and it will automagically translate (almost) all Chinese text in the user interface and then store the results for subsequent use. You'll run up a bunch of tiny jobs the first day you use it after you activate different plugins but it shouldn't cost more than a few cents. it takes a couple of seconds to run when you install a new plugin and it occasionally misses small UI elements or things in dropdown menus but overall it works great. You *can* ask DeepSeek to translate the text, if you give it full access it'll rewrite any plugin for you. But in theory this can cause things to mess up (eg if the code contains function or variable names names in Chinese rather than just strings). I have not had any problems when I tried it but I only did it one one or two plugins. the other issue is that every time the plugins update, which is often, your changes will be overwritten.
cute lovey-dovey snuggling and sex with fat whale dipsy
I'm going to run v4.1 flash iq1 xs, I hope it is good so I can ditch cloud forever!