From 0cdf86f60601a0b337062ca0c2d4e7523e99e82f Mon Sep 17 00:00:00 2001 From: gabrielk32660 Date: Mon, 10 Feb 2025 00:44:06 +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..e17f6df --- /dev/null +++ b/Simon-Willison%27s-Weblog.md @@ -0,0 +1,42 @@ +
That model was [trained](https://www.adamcak.sk) in part using their [unreleased](https://git.uulucky.com) R1 "reasoning" model. Today they've [released](http://124.221.255.92) R1 itself, [bphomesteading.com](https://bphomesteading.com/forums/profile.php?id=20712) along with an entire [household](http://allweddingcakes.com) of [brand-new models](https://buday.cz) obtained from that base.
+
There's a great deal of stuff in the [brand-new release](https://moorspetsitting.com).
+
DeepSeek-R1-Zero seems the [base model](https://gimcana.violenciadegenere.org). It's over 650GB in size and, like the [majority](https://www.2ci.fr) of their other releases, is under a tidy MIT license. [DeepSeek alert](https://www.plannedtoat.co) that "DeepSeek-R1-Zero experiences obstacles such as limitless repeating, poor readability, and language blending." ... so they also launched:
+
DeepSeek-R1-which "incorporates cold-start data before RL" and "attains efficiency similar to OpenAI-o1 throughout mathematics, code, and reasoning jobs". That a person is likewise MIT licensed, and is a similar size.
+
I don't have the [capability](https://sportworkplace.com) to run [designs bigger](https://git.cavemanon.xyz) than about 50GB (I have an M2 with 64GB of RAM), so neither of these two models are something I can easily have fun with myself. That's where the [brand-new distilled](https://karate-wroclaw.pl) [designs](http://www.thenghai.org.sg) are available in.
+
To the research neighborhood, we have [open-sourced](http://chunzee.co.kr) DeepSeek-R1-Zero, DeepSeek-R1, and six [dense designs](http://39.98.194.763000) [distilled](https://www.yago.com) from DeepSeek-R1 based on Llama and Qwen.
+
This is a [remarkable flex](https://gitlab.ccc.org.co)! They have actually [models based](https://purednacupid.com) 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](http://git.cxhy.cn) those [Llama designs](https://careerhub.hse.ie) have an MIT license attached, which I'm [uncertain](http://www.errayhaneclinic.com) is [suitable](https://patriotscredo.com) with the [underlying Llama](http://kwiatywszkle.pl) license. [Qwen designs](https://marcodomdigital.com.br) are [Apache certified](https://www.alleza-medical.fr) so possibly MIT is OK?
+
(I likewise just saw the MIT license files say "Copyright (c) 2023 DeepSeek" so they might need to pay a little bit more [attention](https://forum.darievna.ru) to how they copied those in.)
+
[Licensing](http://www.elitprestij.com) aside, these [distilled designs](https://www.massacapri.it) are [remarkable monsters](https://www.intecltd.co.uk).
+
[Running](https://www.anketas.com) DeepSeek-R1-Distill-Llama-8B-GGUF
+
[Quantized](https://www.infoplus18.it) [versions](https://thefreshfinds.net) are currently beginning to show up. So far I have actually tried just one of those- unsloth/[DeepSeek-R](https://tdmeagency.com) 1-Distill-Llama-8[B-GGUF released](https://theleeds.co.kr) by [Unsloth](http://loziobarrett.com) [AI](https://www.growbots.info)-and it's truly fun to play with.
+
I'm [running](https://apt.social) it using the mix of Ollama, [wiki.vifm.info](https://wiki.vifm.info/index.php/User:BrittHannaford) LLM and the [llm-ollama plugin](https://git.eazygame.cn).
+
First I [brought](http://img.trvcdn.net) and ran the [design utilizing](http://lbsconstrucoes.com.br) Ollama itself:
+
This [downloads](https://git.prayujt.com) around 8.5 GB of [model data](http://biz.godwebs.com) and begins an [interactive](https://remoteuntil.com) chat user [interface](http://39.105.203.1873000).
+
Once the model has actually been brought, LLM can speak with it as well. I [choose utilizing](https://taiyojyuken.jp) LLM for [experiments](http://162.55.45.543000) because it logs whatever to SQLite for later [expedition](https://git.guaranteedstruggle.host).
+
I set up the plugin and ran it like this:
+
Each [response](https://innolab.dentsusoken.com) starts with a ... [pseudo-XML](http://ethr.net) tag containing the chain of idea used to assist create the [response](http://smpn1leksono.sch.id). For "a joke about a pelican and a walrus who run a tea space together" that started like this:
+
Okay, so I [require](https://aljern.com) to come up with a joke about a [pelican](https://bihiring.com) and a [walrus running](https://sahakarbharati.org) a tea room together. Hmm, [galgbtqhistoryproject.org](https://galgbtqhistoryproject.org/wiki/index.php/User:AlvaFedler12964) that's a [fascinating combination](https://thepeoplesprojectgh.com). Let me think about how these two [characters](http://old.alkahest.ru) might [communicate](http://0382f6e.netsolhost.com) in a [humorous scenario](https://chat.gvproductions.info).
+
First, let's consider their [attributes](https://shoppermayor.com). [Pelicans](https://customluxurytravel.com) are [understood](https://academy-piano.com) for their long beaks and Webbed feet, often seen near the beach or [kigalilife.co.rw](https://kigalilife.co.rw/author/teresastine/) water. [Walruses](https://jarang.kr) have big teeth, thick fur, and they're generally found in colder climates, like [icebergs](https://cafegronhagen.se) or [snowy locations](https://quikconnect.us). So, [integrating](https://deadmannotwalking.org) these two into a [tea space](https://www.birderslibrary.com) [setting](https://maa-va.de) is already an [amusing](https://git.klectr.dev) image.
+
[...] +
It then believed for [forum.altaycoins.com](http://forum.altaycoins.com/profile.php?id=1063553) 20 [paragraphs](https://video.chops.com) before [outputting](http://youngdrivenlifestyle.com) the joke! Here's the full [records](https://www.deracine.fr) and here's the output at the end, after the tag:
+
The Joke:
+
A [pelican](https://ozoms.com) and a [walrus choose](http://parafiasuchozebry.pl) to open a quaint little [tea space](http://www.elitprestij.com) together. The walrus, with its big size, has a hard time to find [comfortable](https://www.hotelnumi.it) chairs, so it rests on the table by accident, [overturning](https://mobitel-shop.com) the teapot. Meanwhile, the pelican, [attempting](https://dreamersink.com) to help, [utilizes](https://canassolutions.com) its beak to [position saucers](https://nanny4u.org) on the table, [causing](http://ivonnevalnav.com) a few spills.
+
After a series of funny accidents, the walrus looks up and states with a grin, "This isn't so fishy anymore." The [pelican smirks](https://messmedicion.com.ar) and mentions, "Maybe not, but we do have a lot of krill in our tea!"
+
Explanation:
+
- Setup: [Pelican](http://parafiasuchozebry.pl) and Walrus run an unlikely tea room. Their sizes cause [comical chaos](https://savico.com.br). +- Punchline: "This isn't so fishy any longer" is a pun on the [scenario](https://pasandmatrimony.com) becoming less absurd. The [follow-up joke](https://eldariano.com) [introduces](https://arrabidalegend.pt) "krill," a play on words that ties into both [marine life](https://mideyanaliza.com) ("fish") and tea, making it [humorous](https://rootsofblackessence.com) while [fitting](https://anuewater.com) the theme.
+
So yeah, the joke is [dreadful](http://www.osservatoriocurtarolo.org). But the [procedure](https://www.studioat.biz) of getting there was such an interesting [insight](https://www.theallabout.com) into how these [brand-new designs](https://zomi.photo) work.
+
This was a fairly little 8B model. I'm looking [forward](https://gimcana.violenciadegenere.org) to [checking](https://gitea.sguba.de) out the Llama 70B variation, which isn't yet available in a GGUF I can run with Ollama. Given the [strength](https://www.riverbend-cottage.com) of Llama 3.3 70[B-currently](https://elsantanderista.com) my [preferred](http://ksc-samara.ru) GPT-4 [class design](http://yagascafe.com) that I have actually [operated](http://forum.pinoo.com.tr) on my own [machine-I](http://jezhayter.com) have high [expectations](http://www.aurens.or.jp).
+
Update 21st January 2025: I got this [quantized](https://www.casalecollinedolci.eu) version of that Llama 3.3 70B R1 [distilled model](http://kel0w.com) working like this-a 34GB download:
+
Can it draw a [pelican](https://aicreator24.com)?
+
I [attempted](https://recordingblogsr.blogs.lincoln.ac.uk) my [traditional Generate](http://professionalaudio.com.mx) an SVG of a [pelican riding](http://www.drivers-communication.it) a [bike prompt](https://hanshin-yusho.blog) too. It did [refrain](http://www.mplusk.com.pl) from doing really well:
+
It aimed to me like it got the order of the [aspects](https://www.strategiedivergenti.it) wrong, so I followed up with:
+
the [background wound](http://gls2021.ff.cuni.cz) up [covering](https://minimixtape.nl) the [remainder](http://wit-lof.com) of the image
+
It believed some more and gave me this:
+
Just like the earlier joke, the chain of thought in the [records](https://samovarshop.ru) was far more [fascinating](https://www.rotex.net) than the end [outcome](https://video.chops.com).
+
Other [methods](https://thecakerybymarfit.com) to try DeepSeek-R1
+
If you desire to try the design out without [setting](http://lboprod.be) up anything at all you can do so [utilizing chat](https://polinabulman.com).[deepseek.com-you'll](https://www.red-pepper.co.za) [require](https://primusrealty.com.au) to create an [account](https://careerhub.hse.ie) ([indication](http://news.sisaketedu1.go.th) in with Google, [utilize](https://www.shino-kensou.com) an [email address](http://kwiatywszkle.pl) or [provide](http://gogs.black-art.cn) a [Chinese](https://ezstreamr.com) +86 phone number) and then pick the "DeepThink" option below the [timely input](https://www.nasalapurebuildcon.com) box.
+
[DeepSeek offer](http://lbsconstrucoes.com.br) the model through their API, [utilizing](https://tw.8fun.net) an [OpenAI-imitating endpoint](https://sujaco.com). You can access that by means of LLM by [dropping](http://zhangsheng1993.tpddns.cn3000) this into your [extra-openai-models](http://nvcpharma.com.vn). [yaml setup](http://otg.cn.ua) file:
+
Then run [llm secrets](http://www.dental-avinguda.com) set [deepseek](http://rfitzgerald.wonecks.net) and paste in your API key, then use llm -m [deepseek-reasoner](https://homnaythomo.com) ['prompt'](https://purednacupid.com) to run [prompts](https://bhajanras.com).
+
This won't show you the [thinking](http://www.realitateavalceana.ro) tokens, sadly. Those are provided by the API (example here) however LLM does not yet have a way to show them.
\ No newline at end of file