That model was trained in part using their unreleased R1 "reasoning" design. Today they've launched R1 itself, along with an entire household of new designs obtained from that base.
There's a lot of stuff in the brand-new release.
DeepSeek-R1-Zero appears to be the base design. It's over 650GB in size and, like most of their other releases, is under a clean MIT license. DeepSeek warn that "DeepSeek-R1-Zero experiences difficulties such as limitless repetition, poor readability, and language mixing." ... so they likewise launched:
DeepSeek-R1-which "incorporates cold-start data before RL" and "attains efficiency similar to OpenAI-o1 throughout mathematics, code, and reasoning tasks". That one is likewise MIT licensed, and is a similar size.
I do not have the ability to run designs larger than about 50GB (I have an M2 with 64GB of RAM), so neither of these two designs are something I can quickly have fun with myself. That's where the new distilled models are available in.
To support the research study neighborhood, we have open-sourced DeepSeek-R1-Zero, DeepSeek-R1, photorum.eclat-mauve.fr and 6 thick models distilled from DeepSeek-R1 based on Llama and Qwen.
This is an interesting 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 is compatible with the underlying Llama license. Qwen designs are Apache licensed so possibly MIT is OK?
(I also just observed the MIT license files state "Copyright (c) 2023 DeepSeek" so they might require to pay a bit more attention to how they copied those in.)
aside, these distilled designs are remarkable monsters.
Running DeepSeek-R1-Distill-Llama-8B-GGUF
Quantized variations are already beginning to appear. So far I have actually attempted simply one of those- unsloth/DeepSeek-R 1-Distill-Llama-8B-GGUF released by Unsloth AI-and it's truly fun to play 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 information and begins an interactive chat interface.
Once the model has been fetched, LLM can speak with it also. I prefer utilizing LLM for experiments due to the fact that it logs whatever to SQLite for later exploration.
I installed the plugin and ran it like this:
Each action starts with a ... pseudo-XML tag containing the chain of thought used to help generate 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 believe about how these 2 characters might interact in an amusing scenario.
First, let's consider their characteristics. Pelicans are known for their long beaks and Webbed feet, typically seen near the beach or water. Walruses have big teeth, thick fur, and they're usually discovered in chillier environments, galgbtqhistoryproject.org like icebergs or snowy locations. So, integrating these two into a tea room setting is currently a funny image.
[...]
It then thought for 20 paragraphs before outputting the joke! Here's the full records and here's the output at the end, after the tag:
The Joke:
A pelican and a walrus decide to open a quaint little tea room together. The walrus, with its huge size, oke.zone has a hard time to discover comfy chairs, asteroidsathome.net so it sits on the table by mishap, overturning the teapot. Meanwhile, photorum.eclat-mauve.fr the pelican, attempting to assist, uses its beak to put dishes on the table, causing a couple of spills.
After a series of funny incidents, the walrus searches for and states with a grin, "This isn't so fishy any longer." The pelican smirks and mentions, "Maybe not, however we do have a great deal of krill in our tea!"
Explanation:
- Setup: Pelican and Walrus run an unlikely tea space. Their sizes result in comedic mayhem.
- Punchline: "This isn't so fishy anymore" is a pun on the scenario ending up being less unreasonable. The follow-up joke presents "krill," a play on words that ties into both marine life ("fish") and tea, making it amusing while fitting the style.
So yeah, the joke is awful. But the procedure of arriving was such an intriguing insight into how these new models work.
This was a fairly little 8B design. I'm anticipating 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 preferred GPT-4 class model that I've run on my own machine-I have high expectations.
Update 21st January 2025: I got this quantized variation of that Llama 3.3 70B R1 distilled model working like this-a 34GB download:
Can it draw a pelican?
I tried my classic Generate an SVG of a pelican riding a bike timely too. It did refrain from doing effectively:
It aimed to me like it got the order of the elements incorrect, so I followed up with:
the background wound up covering the remainder of the image
It thought some more and gave me this:
Just like the earlier joke, the chain of thought in the transcript was even more interesting than the end outcome.
Other methods to attempt DeepSeek-R1
If you want to try the model out without setting up anything you can do so utilizing chat.deepseek.com-you'll need to develop an account (sign in with Google, use an email address or provide a Chinese +86 telephone number) and after that pick the "DeepThink" choice below the prompt input box.
DeepSeek use the model by means of their API, using an OpenAI-imitating endpoint. You can access that through LLM by dropping this into your extra-openai-models. yaml configuration file:
Then run llm secrets set deepseek and paste in your API key, then utilize llm -m deepseek-reasoner 'prompt' to run triggers.
This won't reveal you the thinking tokens, sadly. Those are dished out by the API (example here) however LLM does not yet have a way to display them.