Simon Willison's Weblog

That design was trained in part utilizing their unreleased R1 "reasoning" model. Today they've launched R1 itself, in addition to a whole household of brand-new models obtained from that base.

There's a lot of stuff in the new release.

DeepSeek-R1-Zero appears to be the base design. It's over 650GB in size and, like the majority of their other releases, is under a tidy MIT license. DeepSeek warn that "DeepSeek-R1-Zero experiences challenges such as unlimited repetition, poor readability, and language mixing." ... so they also launched:

DeepSeek-R1-which "incorporates cold-start information before RL" and "attains efficiency comparable to OpenAI-o1 across mathematics, code, and thinking tasks". That a person is likewise MIT accredited, and is a comparable size.

I don't have the ability to run designs bigger than about 50GB (I have an M2 with 64GB of RAM), so neither of these two designs are something I can easily have fun with myself. That's where the new distilled designs are available in.

To support the research study neighborhood, we have open-sourced DeepSeek-R1-Zero, DeepSeek-R1, and 6 thick designs distilled from DeepSeek-R1 based on Llama and Qwen.

This is a remarkable flex! They have designs based 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 designs have an MIT license attached, which I'm uncertain works with the underlying Llama license. Qwen designs are Apache licensed so maybe MIT is OK?

Licensing aside, these distilled designs are remarkable beasts.

Running DeepSeek-R1-Distill-Llama-8B-GGUF

Quantized variations are currently starting to show up. Up until now I've tried just among those- unsloth/DeepSeek-R 1-Distill-Llama-8B-GGUF released by Unsloth AI-and it's truly enjoyable to have fun with.

I'm running it using the combination of Ollama, LLM and the llm-ollama plugin.

First I fetched and ran the model utilizing Ollama itself:

This downloads around 8.5 GB of design data and starts an interactive chat user interface.

Once the model has actually been brought, LLM can speak with it also. I prefer utilizing LLM for experiments since it logs whatever to SQLite for later expedition.

I installed the plugin and ran it like this:

Each response starts with a ... pseudo-XML tag containing the chain of idea utilized to assist create the reaction. For "a joke about a pelican and a walrus who run a tea room together" that started like this:

Okay, so I need to come up with a joke about a pelican and a walrus running a tea room together. Hmm, that's a fascinating combination. Let me think about how these two characters may communicate in an amusing scenario.

First, let's consider their attributes. Pelicans are known for their long beaks and Webbed feet, frequently seen near the beach or water. Walruses have huge teeth, thick fur, and they're normally discovered in colder environments, like icebergs or snowy areas. So, combining these 2 into a tea room setting is currently a funny image.

[...]

It then believed for 20 paragraphs before outputting the joke! Here's the full transcript and here's the output at the end, after the tag:

The Joke:

A pelican and a walrus choose to open a quaint little tea space together. The walrus, with its huge size, has a hard time to find comfy chairs, so it sits on the table by accident, knocking over the teapot. Meanwhile, the pelican, trying to help, uses its beak to put dishes on the table, causing a few spills.

After a series of comical incidents, the walrus looks up and states with a smile, "This isn't so fishy any longer." The pelican smirks and mentions, "Maybe not, but we do have a great deal of krill in our tea!"

Explanation:

- Setup: Pelican and Walrus run an unlikely tea room. Their sizes cause comical turmoil.
- Punchline: "This isn't so fishy any longer" is a pun on the scenario ending up being less unreasonable. The follow-up joke introduces "krill," a play on words that ties into both marine life ("fish") and tea, making it funny while fitting the theme.

So yeah, the joke is dreadful. But the process of arriving was such an interesting insight into how these new models work.

This was a fairly small 8B model. I'm anticipating checking out the Llama 70B variation, which isn't yet available in a GGUF I can keep up Ollama. Given the strength of Llama 3.3 70B-currently my preferred GPT-4 class model that I have actually operated on my own machine-I have high expectations.

Update 21st January 2025: I got this quantized version of that Llama 3.3 70B R1 distilled model working like this-a 34GB download:

Can it draw a pelican?

I attempted my traditional Generate an SVG of a pelican riding a bike prompt too. It did refrain from doing extremely well:

It aimed to me like it got the order of the aspects incorrect, so I followed up with:

the background wound up covering the remainder of the image

It believed some more and provided me this:

Similar to the earlier joke, the chain of believed in the records was even more fascinating than the end outcome.

Other methods to try DeepSeek-R1

If you want to attempt the design out without installing anything at all you can do so utilizing chat.deepseek.com-you'll require to develop an account (check in with Google, use an email address or supply a Chinese +86 telephone number) and after that pick the "DeepThink" alternative below the prompt input box.

DeepSeek offer the design via their API, utilizing an OpenAI-imitating endpoint. You can access that through LLM by dropping this into your extra-openai-models. yaml configuration file:

Then run llm keys set deepseek and gdprhub.eu paste in your API secret, then use llm -m deepseek-reasoner 'prompt' to run triggers.

This will not show you the thinking tokens, sadly. Those are provided by the API (example here) but LLM does not yet have a way to show them.