Simon Willison's Weblog

That model was trained in part utilizing their unreleased R1 "reasoning" model. Today they have actually launched R1 itself, in addition to an entire household of brand-new models obtained from that base.

There's a lot of things in the brand-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 clean MIT license. DeepSeek warn that "DeepSeek-R1-Zero encounters challenges such as limitless repeating, poor readability, and language mixing." ... so they also launched:

DeepSeek-R1-which "incorporates cold-start data before RL" and "attains performance comparable to OpenAI-o1 across math, code, and thinking tasks". That one is also MIT certified, and is a similar size.

I do not have the capability to run models bigger than about 50GB (I have an M2 with 64GB of RAM), so neither of these 2 models are something I can easily play with myself. That's where the brand-new distilled models are available in.

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

This is an interesting flex! They have actually models based upon 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 have an MIT license attached, wiki.dulovic.tech which I'm uncertain is compatible with the underlying Llama license. Qwen designs are Apache certified so maybe MIT is OK?

Licensing aside, these distilled models are interesting beasts.

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

Quantized variations are currently beginning to reveal up. So far I have actually attempted just one of those- unsloth/DeepSeek-R 1-Distill-Llama-8B-GGUF launched by Unsloth AI-and it's truly fun to have fun with.

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

First I fetched and ran the design utilizing Ollama itself:

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

Once the design has been brought, LLM can talk to it as well. I prefer using LLM for experiments since it logs whatever to SQLite for later exploration.

I installed the plugin and ran it like this:

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

Okay, so I need to come up with a joke about a pelican and a walrus running a tea space together. Hmm, that's an intriguing mix. Let me think about how these 2 characters might interact in an amusing circumstance.

First, let's consider their attributes. Pelicans are understood for their long beaks and Webbed feet, frequently seen near the beach or water. Walruses have huge teeth, thick fur, and they're generally found in cooler climates, like icebergs or snowy locations. So, integrating these 2 into a tea space setting is currently an amusing image.

[...]

It then thought 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 choose to open a quaint little tea room together. The walrus, with its big size, wiki-tb-service.com has a hard time to discover comfy chairs, so it sits on the table by accident, knocking over the teapot. Meanwhile, the pelican, attempting to help, utilizes its beak to put dishes on the table, triggering a couple of spills.

After a series of humorous incidents, the walrus looks up and states with a smile, "This isn't so fishy anymore." The pelican smirks and says, "Maybe not, but we do have a lot of krill in our tea!"

Explanation:

- Setup: Pelican and Walrus run a not likely tea room. Their sizes cause comical chaos.
- 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 connects into both marine life ("fish") and tea, making it amusing while fitting the theme.

So yeah, the joke is terrible. But the procedure of getting there was such a fascinating insight into how these new models work.

This was a fairly small 8B design. I'm eagerly anticipating trying 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 favourite GPT-4 class design that I've 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 tried my traditional Generate an SVG of a pelican riding a bicycle prompt too. It did refrain from doing really well:

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

the background ended up covering the remainder of the image

It believed some more and gave me this:

Similar to the earlier joke, the chain of thought in the records was even more intriguing than the end result.

Other methods to try DeepSeek-R1

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

DeepSeek offer the model via 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 utilize llm -m deepseek-reasoner 'prompt' to run prompts.

This won't show you the thinking tokens, unfortunately. Those are dished out by the API (example here) but LLM doesn't yet have a method to show them.