Simon Willison's Weblog
maricruzgoldsm このページを編集 7 ヶ月 前


That design was trained in part utilizing their unreleased R1 "thinking" design. Today they have actually launched R1 itself, in addition to a whole family of brand-new designs obtained from that base.

There's a great deal of things in the new release.

DeepSeek-R1-Zero appears to be the base model. 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 encounters challenges such as limitless repeating, poor readability, and language blending." ... so they likewise launched:

DeepSeek-R1-which "includes cold-start information before RL" and "attains performance similar to OpenAI-o1 throughout math, code, and reasoning tasks". That one is also MIT accredited, and is a comparable size.

I do not have the capability to run models larger 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 new distilled designs are available in.

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

This is a fascinating 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, which I'm uncertain works with the underlying Llama license. Qwen models are Apache accredited so possibly MIT is OK?

(I likewise simply observed the MIT license files say "Copyright (c) 2023 DeepSeek" so they might need to pay a little bit more attention to how they copied those in.)

Licensing aside, these distilled designs are interesting beasts.

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

Quantized variations are currently starting to appear. So far I have actually tried simply among those- unsloth/DeepSeek-R 1-Distill-Llama-8B-GGUF launched by Unsloth AI-and it's really enjoyable to have fun with.

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

First I fetched and social.concienciacasanare.com ran the model utilizing Ollama itself:

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

Once the design has been brought, LLM can talk to it too. I prefer using LLM for experiments due to the fact that it logs everything to SQLite for sitiosecuador.com later expedition.

I set up the plugin and ran it like this:

Each action begins with a ... pseudo-XML tag containing the chain of idea used to assist generate the response. 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. Let me think of how these 2 characters may connect in a humorous scenario.

First, larsaluarna.se let's consider their qualities. Pelicans are known for their long beaks and canadasimple.com Webbed feet, often seen near the beach or water. Walruses have big teeth, thick fur, and they're usually discovered in colder climates, like icebergs or snowy locations. So, integrating these two into a tea space setting is currently a funny image.

[...]
It then believed for 20 paragraphs before outputting the joke! Here's the complete records 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 discover comfortable chairs, asteroidsathome.net so it sits on the table by mishap, knocking over the teapot. Meanwhile, the pelican, to help, utilizes its beak to position saucers on the table, causing a few spills.

After a series of funny mishaps, the walrus searches for and states with a smile, "This isn't so fishy anymore." The pelican smirks and mentions, "Maybe not, but we do have a lot of krill in our tea!"

Explanation:

- Setup: Pelican and Walrus run a not likely tea space. Their sizes result in comical turmoil.

  • Punchline: "This isn't so fishy any longer" is a pun on the scenario becoming less ridiculous. The follow-up joke introduces "krill," a play on words that ties into both marine life ("fish") and tea, archmageriseswiki.com making it amusing while fitting the style.

    So yeah, the joke is horrible. But the process of arriving was such an interesting insight into how these brand-new designs work.

    This was a fairly little 8B model. I'm looking forward to experimenting with 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 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 classic Generate an SVG of a pelican riding a bike timely too. It did refrain from doing extremely well:

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

    the background ended up covering the remainder of the image

    It thought some more and gave me this:

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

    Other ways to try DeepSeek-R1

    If you wish to try the design out without installing anything at all you can do so utilizing chat.deepseek.com-you'll need to develop an account (indication in with Google, utilize an email address or utahsyardsale.com provide a Chinese +86 telephone number) and then select the "DeepThink" alternative listed below the timely input box.

    DeepSeek provide the design through 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 paste in your API secret, then use llm -m deepseek-reasoner 'prompt' to run triggers.

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