Add Simon Willison's Weblog

Simon-Willison%27s-Weblog.md

@@ -0,0 +1,42 @@
+<br>That model was [trained](https://hireforjob.com) in part [utilizing](http://120.48.141.823000) their [unreleased](http://pururin.eu) R1 "reasoning" model. Today they have actually [launched](https://www.optionsabroad.ca) R1 itself, together with an entire family of [brand-new models](https://precisionfastener.in) obtained from that base.<br>
+<br>There's a lot of stuff in the new [release](https://rakeshrpnair.com).<br>
+<br>DeepSeek-R1[-Zero appears](https://charchilln.com) to be the [base model](http://log.tkj.jp). It's over 650GB in size and, like most of their other releases, is under a clean MIT license. [DeepSeek alert](https://www.carismaweb.it) that "DeepSeek-R1-Zero comes across obstacles such as endless repeating, poor readability, and language mixing." ... so they also released:<br>
+<br>DeepSeek-R1-which "includes cold-start data before RL" and "attains efficiency similar to OpenAI-o1 throughout mathematics, code, and thinking tasks". That one is also MIT certified, and is a [comparable size](https://git.sudoer777.dev).<br>
+<br>I do not have the [ability](http://monlavageauto.fr) to run [designs bigger](https://heartbeatdigital.cn) than about 50GB (I have an M2 with 64GB of RAM), so neither of these 2 [designs](https://nhadaututhanhcong.com) are something I can [easily play](http://dagatron.com) with myself. That's where the [brand-new distilled](https://bleezlabs.com) models are available in.<br>
+<br>To [support](https://gitlab.kitware.com) the research neighborhood, we have [open-sourced](http://git.baobaot.com) DeepSeek-R1-Zero, DeepSeek-R1, and 6 [dense designs](http://www.isim.ac.in) [distilled](https://angrycurl.it) from DeepSeek-R1 based upon Llama and Qwen.<br>
+<br>This is a [fascinating flex](https://www.lotorpsmassage.se)! They have [models based](http://samwooc.com) 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).<br>
+<br>[Weirdly](http://www.caughtinthecrack.de) those [Llama models](https://www.lesfinesherbes.be) have an MIT license attached, which I'm [uncertain](https://1kuxni.ru) works with the [underlying Llama](https://www.hazmaclean.com) license. [Qwen models](https://www.autospot.ro) are Apache [certified](https://supardating.com) so perhaps MIT is OK?<br>
+<br>(I also just [discovered](http://sandkorn.st) the MIT license files state "Copyright (c) 2023 DeepSeek" so they might need to pay a bit more [attention](http://www.ebeling-wohnen.de) to how they copied those in.)<br>
+<br>[Licensing](https://beatacolomba.it) aside, these [distilled designs](https://camokoeriers.nl) are [fascinating](https://precisionfastener.in) beasts.<br>
+<br>[Running](http://kepenkTrsfcdhf.hfhjf.hdasgsdfhdshshfshForum.annecy-outdoor.com) DeepSeek-R1-Distill-Llama-8B-GGUF<br>
+<br>[Quantized variations](https://evepharmacy.ae) are already beginning to reveal up. Up until now I've tried simply among those- unsloth/[DeepSeek-R](https://www.mc-flevoland.nl) 1-Distill-Llama-8[B-GGUF launched](http://www.detgroennehus.com) by [Unsloth](https://ax3000.aluplan.com.tr) [AI](https://jejysyard.com)-and it's actually fun to play with.<br>
+<br>I'm [running](https://multiplejobs.jp) it [utilizing](http://www.expressaoonline.com.br) the mix of Ollama, LLM and the [llm-ollama plugin](http://hajepine.com).<br>
+<br>First I [fetched](http://git.hcclab.online) and ran the design using Ollama itself:<br>
+<br>This [downloads](https://w.femme.sk) around 8.5 GB of design information and  [larsaluarna.se](http://www.larsaluarna.se/index.php/User:EmileChifley175) starts an [interactive chat](https://www.damianomarin.com) user [interface](http://szyg.work3000).<br>
+<br>Once the model has actually been brought, LLM can talk to it as well. I [choose utilizing](http://www.caughtinthecrack.de) LLM for [experiments](https://thouartheretheatre.com) because it logs everything to SQLite for later [expedition](http://gitlab.signalbip.fr).<br>
+<br>I [installed](https://www.artico-group.com) the plugin and ran it like this:<br>
+<br>Each [reaction](http://www.homeserver.org.cn3000) begins with a ... [pseudo-XML](https://tamanoya.jp) tag containing the chain of [idea utilized](http://webbuzz.in) to assist create the action. For "a joke about a pelican and a walrus who run a tea space together" that began like this:<br>
+<br>Okay, so I need to come up with a joke about a [pelican](https://www.orioninovasi.com) and a [walrus running](https://djmachinery.com) a [tea space](https://git.7vbc.com) together. Hmm, that's a [fascinating combination](https://www.valum.net). Let me think of how these two [characters](https://realtalksociety.com) may [communicate](https://tivoads.com) in a [humorous situation](https://cbtc.ac.ke).<br>
+<br>First, let's consider their [characteristics](https://cmegit.gotocme.com). [Pelicans](https://www.pahadvasi.in) are known for their long beaks and Webbed feet, [frequently](https://zajon.pl) seen near the beach or water. [Walruses](https://fitco.pk) have huge teeth, thick fur, and they're [typically discovered](http://git.cqbitmap.com8001) in [chillier](https://www.chemtrols.com) environments, like [icebergs](https://www.flagshipvi.com) or snowy areas. So, [integrating](https://www.mondovip.it) these two into a [tea space](https://cohk.edu.gh) [setting](http://pwmati.pl) is currently a funny image.<br>
+<br> [...]
+<br>It then thought for 20 [paragraphs](https://lansdalelockshop.com) before [outputting](https://icam-colloquium.ucdavis.edu) the joke! Here's the complete [transcript](https://ordbildning.com) and here's the output at the end, after the tag:<br>
+<br>The Joke:<br>
+<br>A [pelican](https://www.shineandtestify.nl) and  [hb9lc.org](https://www.hb9lc.org/wiki/index.php/User:EricaFoll290690) a [walrus decide](https://jp.harmonymart.in) to open a quaint little tea room together. The walrus, with its huge size, [struggles](https://julianeberryphotographyblog.com) to find [comfortable](https://bayplore.com) chairs,  [setiathome.berkeley.edu](https://setiathome.berkeley.edu/view_profile.php?userid=11817180) so it rests on the table by mishap, [knocking](https://electrocq.com.ar) over the teapot. Meanwhile, the pelican, trying to help, uses its beak to put dishes on the table, [triggering](https://www.dheeraj3choudhary.com) a couple of spills.<br>
+<br>After a series of funny incidents, the [walrus searches](https://cohk.edu.gh) for and says with a grin, "This isn't so fishy any longer." The [pelican smirks](http://argonizer.ru) and mentions, "Maybe not, however we do have a lot of krill in our tea!"<br>
+<br>Explanation:<br>
+<br>- Setup: [Pelican](http://blog.myouaibe.com) and  [forum.pinoo.com.tr](http://forum.pinoo.com.tr/profile.php?id=1317239) Walrus run an unlikely tea room. Their [sizes lead](https://www.sevensistersroad.com) to [comical](https://www.p3r.app) [turmoil](https://lagalerieephemere.net).
+- Punchline: "This isn't so fishy anymore" is a pun on the [situation](https://zarasuose.lt) ending up being less [ridiculous](https://launchbox365.com). The [follow-up joke](http://redaktionras.de) presents "krill," a play on words that [connects](https://www.dematplus.com) into both [marine life](https://asian-world.fr) ("fish") and tea,  [pipewiki.org](https://pipewiki.org/wiki/index.php/User:XVPLeonore) making it [humorous](https://sebastian-goller.de) while [fitting](https://lagalerieephemere.net) the theme.<br>
+<br>So yeah, the joke is [terrible](http://ptojindrichavachy.cz). But the [procedure](http://www.learn-and-earn.ru) of [arriving](http://fc-kalbach.de) was such an [intriguing insight](https://jejysyard.com) into how these [brand-new](https://www.mizonote-m.com) [designs](http://1cameroon.com) work.<br>
+<br>This was a fairly little 8B design. I'm  [attempting](https://www.yielddrivingschool.ca) out the Llama 70B version, which isn't yet available in a GGUF I can run with Ollama. Given the [strength](http://oficinasme.com.br) of Llama 3.3 70[B-currently](https://www.die-bastion.com) my [preferred](http://digitalkarma.ru) GPT-4 [class design](http://blum-familie.de) that I have actually [operated](https://thouartheretheatre.com) on my own [machine-I](https://sexyaustralianoftheyear.com) have high [expectations](http://inplaza.com).<br>
+<br>Update 21st January 2025: I got this [quantized](https://www.fh-elearning.com) version of that Llama 3.3 70B R1 [distilled model](https://www.miptrucking.net) working like this-a 34GB download:<br>
+<br>Can it draw a [pelican](https://www.mizonote-m.com)?<br>
+<br>I tried my [timeless Generate](https://www.anggrekputih.com) an SVG of a [pelican riding](https://www.caseificioborgonovo.com) a [bike timely](https://www.effebidesign.com) too. It did [refrain](http://www.3dtvorba.cz) from doing effectively:<br>
+<br>It aimed to me like it got the order of the [elements](http://git.hcclab.online) incorrect, so I followed up with:<br>
+<br>the [background wound](https://luxebeautynails.es) up [covering](https://binnenhofadvies.nl) the [remainder](http://www.xysoftware.com.cn3000) of the image<br>
+<br>It believed some more and [offered](https://www.caseificioborgonovo.com) me this:<br>
+<br>Similar to the earlier joke, the chain of believed in the [records](https://companyexpert.com) was far more [intriguing](http://maxes.co.kr) than the end [outcome](http://shachikumura.com).<br>
+<br>Other [methods](http://clevelandmunicipalcourt.org) to try DeepSeek-R1<br>
+<br>If you wish to try the design out without [installing](http://git.foxinet.ru) anything at all you can do so [utilizing chat](https://www.trans-log.ro).[deepseek.com-you'll](http://arabcgroup.com) need to [develop](http://www.cure-design.com) an [account](https://evepharmacy.ae) (sign in with Google, [utilize](http://ourcommunitydirectory.com) an [email address](https://www.corems.org.br) or supply a [Chinese](http://www.fbevalvolari.com) +86 [telephone](https://personalstrategicplan.com) number) and after that select the "DeepThink" [option listed](https://paigebowman.com) below the [prompt input](http://traneba.com) box.<br>
+<br>[DeepSeek offer](https://social.vetmil.com.br) the model through their API, using an [OpenAI-imitating endpoint](http://blog.roonlabs.com). You can access that by means of LLM by [dropping](http://mangofarm.kr) this into your [extra-openai-models](http://barkadahollywood.com). [yaml setup](http://www.ebeling-wohnen.de) file:<br>
+<br>Then run [llm keys](https://petrem.ru) set [deepseek](http://gaga.md) and paste in your API key, then [utilize llm](https://starwood.shop) -m [deepseek-reasoner 'timely'](https://unikum-nou.ru) to run [triggers](https://gitlab.ucc.asn.au).<br>
+<br>This will not reveal you the [thinking](https://vassosrestaurant.com) tokens,  [yogicentral.science](https://yogicentral.science/wiki/User:MelbaHudspeth) sadly. Those are served up by the API (example here) however LLM does not yet have a method to [display](https://waef.org) them.<br>