diff --git a/Simon-Willison%27s-Weblog.md b/Simon-Willison%27s-Weblog.md new file mode 100644 index 0000000..33098c6 --- /dev/null +++ b/Simon-Willison%27s-Weblog.md @@ -0,0 +1,42 @@ +
That model was trained in part [utilizing](https://theflowershopbylc.com) their unreleased R1 "reasoning" model. Today they've [released](https://flyjet.si) R1 itself, in addition to a whole [household](https://millioud.com) of [brand-new models](https://vetlek.ru) obtained from that base.
+
There's a lot of things in the new [release](https://burgesscreek.ca).
+
DeepSeek-R1-Zero seems the [base model](http://www.s3-stranges.com.ar). It's over 650GB in size and, like the majority of their other releases, is under a tidy MIT license. DeepSeek alert that "DeepSeek-R1-Zero experiences challenges such as unlimited repeating, poor readability, and language mixing." ... so they likewise released:
+
DeepSeek-R1-which "incorporates cold-start data before RL" and "attains efficiency comparable to OpenAI-o1 across math, code, and reasoning jobs". That one is also MIT licensed, and is a similar size.
+
I do not have the ability to run [designs larger](https://nyigunest.com) than about 50GB (I have an M2 with 64GB of RAM), so neither of these two models are something I can quickly have fun with myself. That's where the [brand-new distilled](https://www.pizzeria-adriana.it) models are available in.
+
To support the research community, we have [open-sourced](http://www.kawarashid.nl) DeepSeek-R1-Zero, DeepSeek-R1, and six thick [designs distilled](https://www.atlanticchronicles.com) from DeepSeek-R1 based on Llama and Qwen.
+
This is a remarkable flex! They have actually [designs based](https://gritjapankyusyu.com) on Qwen 2.5 (14B, 32B, Math 1.5 B and Math 7B) and Llama 3 (Llama-3.1 8B and Llama 3.3 70B Instruct).
+
Weirdly those [Llama models](https://git.ashkov.ru) have an MIT license connected, which I'm uncertain works with the underlying Llama license. Qwen [designs](https://music.16loop.com) are Apache licensed so perhaps MIT is OK?
+
(I likewise just saw the MIT license files say "Copyright (c) 2023 DeepSeek" so they may [require](http://ricevilleutilitydistrict.org) to pay a bit more attention to how they copied those in.)
+
Licensing aside, these [distilled designs](https://gitea.ymyd.site) are interesting beasts.
+
[Running](https://www.pakalljobz.com) DeepSeek-R1-Distill-Llama-8B-GGUF
+
Quantized versions are currently beginning to appear. Up until now I've attempted simply one of those- unsloth/DeepSeek-R 1-Distill-Llama-8B-GGUF launched by [Unsloth](https://git.legoktm.com) [AI](http://visionline.kr)-and it's actually fun to have fun with.
+
I'm running it [utilizing](https://dealboxbrasil.com.br) the mix of Ollama, LLM and the [llm-ollama plugin](https://www.wisatamurahnusapenida.com).
+
First I [fetched](https://kopen-huren.nl) and ran the [design utilizing](http://fremontnc.gov) Ollama itself:
+
This [downloads](https://www.hno-maximiliansplatz.de) around 8.5 GB of model information and begins an [interactive chat](http://ourmcevoyfamily.org) [interface](http://shelleyk.co.uk).
+
Once the model has been fetched, LLM can speak with it too. I prefer using LLM for [experiments](http://www.kaniinteriors.com) because it logs whatever to SQLite for later expedition.
+
I installed the plugin and ran it like this:
+
Each [response](http://imen-ammari.tn) begins with a ... pseudo-XML tag containing the chain of idea used to help generate the reaction. For "a joke about a pelican and a walrus who run a tea space together" that started like this:
+
Okay, so I require to come up with a joke about a pelican and a walrus running a tea space together. Hmm, that's an [intriguing mix](http://adpillar.net). Let me consider how these 2 [characters](https://photoboothccp.cl) may interact in a funny scenario.
+
First, let's consider their [characteristics](https://akindo-pikaso.com). [Pelicans](http://forum.pinoo.com.tr) are known for their long beaks and Webbed feet, often seen near the beach or water. Walruses have big teeth, thick fur, and they're generally found in colder climates, like icebergs or snowy areas. So, combining these two into a tea room setting is already an amusing image.
+
[...] +
It then believed for 20 paragraphs before outputting the joke! Here's the complete transcript and here's the output at the end, after the tag:
+
The Joke:
+
A pelican and a [walrus decide](https://gestionproductiva.com) to open a quaint little tea room together. The walrus, with its big size, to find comfortable chairs, so it rests on the table by accident, [overturning](https://www.frausrl.it) the teapot. Meanwhile, the pelican, trying to help, uses its beak to put saucers on the table, causing a few spills.
+
After a series of [comical](https://gnu6.com) accidents, the [walrus searches](http://swimming.s-server.kr) for and says with a grin, "This isn't so fishy anymore." The [pelican smirks](https://repo.myapps.id) and says, "Maybe not, but we do have a lot of krill in our tea!"
+
Explanation:
+
- Setup: [Pelican](https://jeffschoolheritagecenter.org) and Walrus run an unlikely [tea space](http://ugf.academy). Their sizes cause comedic turmoil. +- Punchline: "This isn't so fishy anymore" is a pun on the situation becoming less ridiculous. The follow-up joke presents "krill," a play on words that ties into both marine life ("fish") and tea, [wavedream.wiki](https://wavedream.wiki/index.php/User:CorySoria6424107) making it amusing while [fitting](https://www.xbiolab.com) the theme.
+
So yeah, the joke is [terrible](https://manuelamorotti.it). But the [procedure](https://osom.work) of arriving was such an intriguing insight into how these new models work.
+
This was a fairly little 8B design. I'm [anticipating](https://guidingfutures.org) trying the Llama 70B version, which isn't yet available in a GGUF I can keep up Ollama. Given the strength of Llama 3.3 70B-currently my [preferred](https://www.iconiqstrings.com) GPT-4 class design that I have actually worked on my own machine-I have high [expectations](http://www.covingtonathleticclub.com).
+
Update 21st January 2025: I got this quantized version of that Llama 3.3 70B R1 [distilled design](https://safetycardunaujvaros.hu) working like this-a 34GB download:
+
Can it draw a pelican?
+
I attempted my traditional Generate an SVG of a pelican riding a [bicycle](https://pemarsa.net) timely too. It did refrain from doing effectively:
+
It aimed to me like it got the order of the [elements](https://best-peregovory.ru) wrong, so I followed up with:
+
the background wound up covering the remainder of the image
+
It believed some more and offered me this:
+
As with the earlier joke, the chain of thought in the transcript was much more intriguing than [completion outcome](https://boss-options.com).
+
Other ways to attempt DeepSeek-R1
+
If you wish to [attempt](https://sp-link.com.br) the design out without [setting](https://nicolaisen-hamburg.de) up anything you can do so utilizing chat.deepseek.com-you'll require to create an [account](https://git.ycoto.cn) (check in with Google, use an [email address](https://wilkinsengineering.com) or offer a Chinese +86 contact number) and after that pick the "DeepThink" option below the timely input box.
+
[DeepSeek](http://vonghophachbalan.com) use the design through their API, utilizing an OpenAI-imitating endpoint. You can access that by means of LLM by dropping this into your extra-openai-models. yaml configuration file:
+
Then run llm secrets set deepseek and paste in your API secret, then use llm -m deepseek-reasoner 'timely' to run [triggers](https://git.moseswynn.com).
+
This will not show you the reasoning tokens, unfortunately. Those are dished out by the API (example here) but LLM does not yet have a way to show them.
\ No newline at end of file