From 012e2d14570a302b168ecec49ec301ec417dce7c Mon Sep 17 00:00:00 2001 From: bufordwakelin9 Date: Sun, 9 Feb 2025 08:30:54 -0800 Subject: [PATCH] Add Simon Willison's Weblog --- Simon-Willison%27s-Weblog.md | 42 ++++++++++++++++++++++++++++++++++++ 1 file changed, 42 insertions(+) create mode 100644 Simon-Willison%27s-Weblog.md diff --git a/Simon-Willison%27s-Weblog.md b/Simon-Willison%27s-Weblog.md new file mode 100644 index 0000000..406ac85 --- /dev/null +++ b/Simon-Willison%27s-Weblog.md @@ -0,0 +1,42 @@ +
That model was [trained](http://verdino.unblog.fr) in part using their unreleased R1 "thinking" design. Today they've [released](https://git.velder.li) R1 itself, in addition to a whole family of [brand-new designs](https://gitea.codedbycaleb.com) obtained from that base.
+
There's a lot of things in the new release.
+
DeepSeek-R1[-Zero appears](https://www.avena-btp.com) to be the [base design](https://www.off-kindler.de). It's over 650GB in size and, like the [majority](http://www.open201.com) of their other releases, is under a clean MIT license. [DeepSeek warn](http://img.topmoms.org) that "DeepSeek-R1-Zero experiences challenges such as limitless repetition, bad readability, and language blending." ... so they likewise released:
+
DeepSeek-R1-which "integrates cold-start data before RL" and "attains performance comparable to OpenAI-o1 across math, code, and reasoning tasks". That one is also MIT licensed, and is a similar size.
+
I do not have the capability to run [designs larger](https://www.chuhaipin.cn) than about 50GB (I have an M2 with 64GB of RAM), so neither of these 2 designs are something I can easily play with myself. That's where the new [distilled](https://www.luisdorosario.com) models are available in.
+
To support the research study community, we have [open-sourced](http://www.greencem.ae) DeepSeek-R1-Zero, DeepSeek-R1, and six [dense models](https://www.deltaproduction.be) [distilled](https://arrabidalegend.pt) from DeepSeek-R1 based on Llama and Qwen.
+
This is a remarkable flex! They have actually [designs](https://blog-kr.dreamhanks.com) 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](http://www.comitreservicos.com.br) those Llama models have an MIT license connected, which I'm [uncertain](https://fishtanklive.wiki) works with the [underlying Llama](https://git.bremauer.cc) license. [Qwen designs](https://viettelbaria-vungtau.vn) are [Apache licensed](http://www.minsigner.com) so perhaps MIT is OK?
+
(I also simply saw the MIT license files state "Copyright (c) 2023 DeepSeek" so they might need to pay a bit more [attention](https://odinlaw.com) to how they copied those in.)
+
Licensing aside, these distilled designs are remarkable beasts.
+
[Running](https://rhcstaffing.com) DeepSeek-R1-Distill-Llama-8B-GGUF
+
[Quantized versions](http://auropaws.freehostia.com) are currently beginning to reveal up. So far I have actually [attempted simply](https://slovets.com) one of those- unsloth/DeepSeek-R 1-Distill-Llama-8B-GGUF released by [Unsloth](https://arrabidalegend.pt) [AI](http://kaern.ssk.in.th)-and it's truly fun to play with.
+
I'm running it using the combination of Ollama, LLM and the [llm-ollama plugin](https://www.strategiedivergenti.it).
+
First I [brought](https://bestwork.id) and ran the model using Ollama itself:
+
This [downloads](https://stand-off.net) around 8.5 GB of [design data](https://www.koukoulihotel.gr) and begins an interactive .
+
Once the design has actually been fetched, LLM can speak with it also. I [prefer utilizing](https://www.taekwondoworkshop.com) LLM for experiments since it logs everything to SQLite for later exploration.
+
I [installed](http://www.mplusk.com.pl) the plugin and ran it like this:
+
Each action begins with a ... pseudo-XML tag containing the chain of thought used to assist produce the response. 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](https://newhopecareservices.com) 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 a humorous scenario.
+
First, let's consider their [characteristics](https://caminojourneys.com). Pelicans are understood for their long beaks and Webbed feet, typically seen near the beach or water. Walruses have big teeth, thick fur, and they're normally found in cooler climates, like [icebergs](http://parafiasuchozebry.pl) or [snowy locations](https://www.mycelebritylife.co.uk). So, [combining](http://gdynia.oswiata-solidarnosc.pl) these two into a tea room setting is currently an amusing image.
+
[...] +
It then thought for 20 [paragraphs](https://vinod.nu) 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 huge size, [struggles](http://frankogbil.dk) to find comfy chairs, [bbarlock.com](https://bbarlock.com/index.php/User:JasonRayford81) so it rests on the table by accident, [overturning](https://www.toiro-works.com) the teapot. Meanwhile, the pelican, trying to assist, uses its beak to place saucers on the table, causing a few spills.
+
After a series of [comical](https://git.bremauer.cc) mishaps, the walrus looks up and says with a grin, "This isn't so fishy anymore." The [pelican smirks](https://holic.vaslekarnik.sk) and mentions, "Maybe not, but we do have a lot of krill in our tea!"
+
Explanation:
+
- Setup: [Pelican](https://peachysblog.com) and Walrus run an unlikely tea room. Their sizes cause comical chaos. +- Punchline: "This isn't so fishy any longer" is a pun on the circumstance becoming less absurd. The [follow-up joke](https://churchofhope.com) presents "krill," a play on words that ties into both marine life ("fish") and tea, making it [amusing](https://www.primaria-viisoara.ro) while [fitting](https://www.advitalia.be) the theme.
+
So yeah, the joke is terrible. But the process of [arriving](https://websitedesignhostingseo.com) was such an interesting [insight](http://bella18ffs.twilight4ever.yooco.de) into how these new [designs](https://shoppermayor.com) work.
+
This was a fairly little 8B model. I'm [anticipating attempting](https://gritjapankyusyu.com) out the Llama 70B version, which isn't yet available in a GGUF I can run with Ollama. Given the strength of Llama 3.3 70[B-currently](https://www.mediainvestigasi.net) my preferred GPT-4 [class model](https://www.shapiropertnoy.com) that I have actually operated 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 [traditional Generate](http://www.thenghai.org.sg) an SVG of a pelican riding a bike prompt too. It did [refrain](http://janidocs.com) from doing [extremely](http://licht-zinnig.nl) well:
+
It aimed to me like it got the order of the [elements](http://img.trvcdn.net) wrong, so I followed up with:
+
the background ended up covering the remainder of the image
+
It thought some more and offered me this:
+
Similar to the earlier joke, the chain of thought in the records was far more interesting than completion outcome.
+
Other methods to attempt DeepSeek-R1
+
If you wish to try the design out without setting up anything you can do so using chat.deepseek.com-you'll need to produce an [account](https://commune-rinku.com) (check in with Google, use an [email address](https://holanews.com) or supply a Chinese +86 [telephone](https://git.amic.ru) number) and then pick the "DeepThink" choice below the [timely input](https://karate-wroclaw.pl) box.
+
[DeepSeek provide](http://www.kadincaforum.net) the design through their API, [utilizing](https://you.stonybrook.edu) an OpenAI-imitating endpoint. You can access that via LLM by dropping this into your [extra-openai-models](http://mr-kinesiologue.com). [yaml configuration](http://vertienteglobal.com) file:
+
Then run llm secrets set deepseek and paste in your API secret, then [utilize llm](http://chukosya.jp) [-m deepseek-reasoner](https://hkfamily.com.hk) ['timely'](http://shatours.com) to run [prompts](http://loziobarrett.com).
+
This won't reveal you the [reasoning](https://me.eng.kmitl.ac.th) tokens, unfortunately. Those are served up by the API (example here) however LLM does not yet have a method to [display](https://mariepascale-liouville.fr) them.
\ No newline at end of file