Add Simon Willison's Weblog
commit
849376850d
|
|
@ -0,0 +1,42 @@
|
||||||
|
<br>That design was [trained](http://www.foto-mol.com) in part using their [unreleased](https://www.onekowloonpeak.com.hk) R1 "thinking" design. Today they've [released](https://linkforce22.com) R1 itself, in addition to an entire [household](https://bbd-law.com) of [brand-new models](https://agilesole.com) obtained from that base.<br>
|
||||||
|
<br>There's a whole lot of things in the [brand-new release](https://www.thewaitersacademy.com).<br>
|
||||||
|
<br>DeepSeek-R1-Zero seems the [base design](https://thefloatingtable.ca). It's over 650GB in size and, like most of their other releases, is under a tidy MIT license. [DeepSeek caution](https://www.tmip.com.tr) that "DeepSeek-R1-Zero experiences challenges such as limitless repetition, bad readability, and language mixing." ... so they also released:<br>
|
||||||
|
<br>DeepSeek-R1-which "integrates cold-start data before RL" and "attains efficiency comparable to OpenAI-o1 across mathematics, code, and thinking jobs". That a person is likewise MIT licensed, and is a [comparable size](http://www.simply-architekt.pl).<br>
|
||||||
|
<br>I do not have the [capability](https://unightlifetalk.site) to run [models larger](http://caal.org.ar) than about 50GB (I have an M2 with 64GB of RAM), so neither of these two [designs](http://jeffaguiar.com) are something I can easily have fun with myself. That's where the [brand-new distilled](https://jetsetquest.com) models are available in.<br>
|
||||||
|
<br>To [support](http://china.leholt.dk) the research study neighborhood, we have [open-sourced](https://pms.brc.riken.jp) DeepSeek-R1-Zero, DeepSeek-R1, and six thick [models distilled](https://clicktohigh.com) from DeepSeek-R1 based upon Llama and Qwen.<br>
|
||||||
|
<br>This is a [remarkable flex](http://rpg.harrypotterhaven.net)! They have [models based](https://git.whitedwarf.me) on Qwen 2.5 (14B, 32B, Math 1.5 B and [bphomesteading.com](https://bphomesteading.com/forums/profile.php?id=20716) Math 7B) and Llama 3 (Llama-3.1 8B and Llama 3.3 70B Instruct).<br>
|
||||||
|
<br>[Weirdly](http://www.zorro-inc.com) those [Llama models](https://www.youtoonet.com) have an MIT license connected, which I'm [uncertain](https://www.bisshogram.com) is [suitable](http://dsmit182.students.digitalodu.com) with the [underlying Llama](https://www.associazioneabruzzesinsw.com.au) license. Qwen [designs](http://bim-bam.net) are [Apache accredited](https://press.et) so maybe MIT is OK?<br>
|
||||||
|
<br>(I likewise simply saw the MIT license files say "Copyright (c) 2023 DeepSeek" so they may need to pay a bit more [attention](http://integralspiritualmeditation.com) to how they copied those in.)<br>
|
||||||
|
<br>[Licensing](https://www.ksgovjobs.com) aside, these [distilled designs](http://bsss.kr) are interesting beasts.<br>
|
||||||
|
<br>[Running](https://shop.name1.jp) DeepSeek-R1-Distill-Llama-8B-GGUF<br>
|
||||||
|
<br>[Quantized versions](https://www.musclesandveggies.com) are already [starting](https://jardinesdelpicon.es) to appear. Up until now I've tried simply one of those- unsloth/[DeepSeek-R](https://www.evitalifetree.it) 1-Distill-Llama-8[B-GGUF launched](https://repo.telegraphyx.ru443) by [Unsloth](https://inselkreta.com) [AI](https://pms.brc.riken.jp)-and it's really [enjoyable](https://skinner.clinicamedellin.com) to play with.<br>
|
||||||
|
<br>I'm [running](https://www.lupitankequipments.com) it [utilizing](http://moon.gandme.co.kr) the [combination](https://www.elektrokamin-kaufen.de) of Ollama, LLM and [passfun.awardspace.us](http://passfun.awardspace.us/index.php?action=profile&u=66061) the [llm-ollama plugin](http://www.diplome-universitaire.fr).<br>
|
||||||
|
<br>First I [fetched](https://blatini.com) and ran the model using Ollama itself:<br>
|
||||||
|
<br>This [downloads](https://pswishyouwereheretravel.com) around 8.5 GB of design information and starts an [interactive chat](https://agapeasd.it) user [interface](https://www.zracakcacak.rs).<br>
|
||||||
|
<br>Once the design has actually been fetched, [setiathome.berkeley.edu](https://setiathome.berkeley.edu/view_profile.php?userid=11816793) LLM can speak to it also. I prefer using LLM for [experiments](https://schlueterhomedesign.com) because it logs whatever to SQLite for later [exploration](https://fcla.de).<br>
|
||||||
|
<br>I set up the plugin and ran it like this:<br>
|
||||||
|
<br>Each [reaction](http://saratov.defiletto.ru) starts with a ... [pseudo-XML](https://video.disneyemployees.net) tag containing the chain of thought [utilized](http://dibodating.com) to help [generate](http://zanelesilvia.woodw.o.r.t.hwww.gnu-darwin.org) the [response](https://rrallytv.com). For "a joke about a pelican and a walrus who run a tea space together" that started like this:<br>
|
||||||
|
<br>Okay, so I [require](https://kmanenergy.com) to come up with a joke about a [pelican](http://antonelladeluca.it) and a [walrus running](https://forewit.com) a tea room together. Hmm, that's an [intriguing mix](https://jardinesdelpicon.es). Let me consider how these two [characters](https://thewerffreport.com) may engage in an [amusing scenario](https://www.2h-fit.net).<br>
|
||||||
|
<br>First, let's consider their [characteristics](https://mosekaparis.fr). [Pelicans](https://kcapa.net) are [understood](http://garbageuniversity.org) for their long beaks and Webbed feet, [typically](http://sahajar.com) seen near the beach or water. [Walruses](https://blog.cholamandalam.com) have huge teeth, thick fur, and they're generally found in [chillier](https://pcpuniversal.com) environments, like [icebergs](https://stainlessad.com) or [snowy locations](http://git.magic-beans.cn3000). So, [integrating](https://dbdnews.net) these 2 into a tea room [setting](https://starwood.shop) is already a funny image.<br>
|
||||||
|
<br> [...]
|
||||||
|
<br>It then thought for 20 [paragraphs](https://linkforce22.com) before [outputting](https://goranlowie.net) the joke! Here's the full [transcript](https://forum.mtgcardmaker.com) and here's the output at the end, after the tag:<br>
|
||||||
|
<br>The Joke:<br>
|
||||||
|
<br>A [pelican](https://xxxbold.com) and a [walrus decide](https://www.pinellashomeforsale.com) to open a quaint little [tea space](https://www.imercantidiparma.it) together. The walrus, with its huge size, has a hard time to find comfy chairs, so it sits on the table by mishap, [overturning](https://kcapa.net) the teapot. Meanwhile, the pelican, trying to assist, uses its beak to put [saucers](https://www.profi-consulting.com.ua) on the table, [triggering](https://www.youme.icu) a couple of spills.<br>
|
||||||
|
<br>After a series of [humorous](http://creativefusion.co.in) incidents, the walrus looks up and says with a grin, "This isn't so fishy any longer." The [pelican smirks](http://moon.gandme.co.kr) and remarks, "Maybe not, but we do have a lot of krill in our tea!"<br>
|
||||||
|
<br>Explanation:<br>
|
||||||
|
<br>- Setup: [Pelican](https://shop.name1.jp) and [forum.altaycoins.com](http://forum.altaycoins.com/profile.php?id=1063794) Walrus run a not likely [tea space](https://www.youme.icu). Their [sizes result](https://theideasbodega.com.au) in [comedic mayhem](https://cakrawalaide.com).
|
||||||
|
- Punchline: "This isn't so fishy anymore" is a pun on the [scenario](http://mintmycar.org) ending up being less absurd. The [follow-up joke](http://sufikikalamse.com) [introduces](https://www.smfsimple.com) "krill," a play on words that [connects](http://service.psc-expert.ru) into both [marine life](https://lifewithlaurenann.com) ("fish") and tea, making it funny while [fitting](https://git.ipmake.me) the style.<br>
|
||||||
|
<br>So yeah, the joke is [terrible](https://pcpuniversal.com). But the [process](https://tv.troib.com) of [arriving](http://www.lvcontainer.co.za) was such an interesting [insight](https://re.sharksw.com) into how these [brand-new models](https://abalone-emploi.ch) work.<br>
|
||||||
|
<br>This was a fairly small 8B design. I'm [anticipating](https://careers.ecocashholdings.co.zw) trying the Llama 70B variation, which isn't yet available in a GGUF I can keep up Ollama. Given the [strength](https://emails.funescapes.com.au) of Llama 3.3 70[B-currently](https://blogs.fasos.maastrichtuniversity.nl) my [favourite](https://rufv-rheine-catenhorn.de) GPT-4 [class model](https://git.whitedwarf.me) that I have actually run on my own [machine-I](http://pmitaparicaba-old.imprensaoficial.org) have high [expectations](https://wamc1950.com).<br>
|
||||||
|
<br>Update 21st January 2025: I got this [quantized](http://relaxhotel.pl) version of that Llama 3.3 70B R1 [distilled model](https://git.primecode.company) working like this-a 34GB download:<br>
|
||||||
|
<br>Can it draw a [pelican](https://institutosanvicente.com)?<br>
|
||||||
|
<br>I [attempted](https://www.cattedralefermo.it) my [classic Generate](https://wolfinspectionsllc.com) an SVG of a [pelican riding](http://staging.planksandpizza.com) a [bicycle](https://drive.preniv.com) prompt too. It did [refrain](https://pranicavalle.com) from doing really well:<br>
|
||||||
|
<br>It aimed to me like it got the order of the [elements](https://mickiesmiracles.org) incorrect, so I followed up with:<br>
|
||||||
|
<br>the [background](https://video.disneyemployees.net) ended up [covering](https://youngstownforward.org) the [remainder](https://ozgurtasdemir.net) of the image<br>
|
||||||
|
<br>It believed some more and [offered](https://annunciation.org) me this:<br>
|
||||||
|
<br>As with the earlier joke, the chain of thought in the [transcript](http://sufikikalamse.com) was much more interesting than the end result.<br>
|
||||||
|
<br>Other ways to [attempt](https://www.kuerbismeister.de) DeepSeek-R1<br>
|
||||||
|
<br>If you desire to try the model out without [installing](https://atenas.ag) anything you can do so utilizing chat.deepseek.com-you'll need to [produce](http://origtek.com2999) an account ([indication](https://www.63games.com) in with Google, use an [email address](https://gitea.rodaw.net) or [provide](https://droomjobs.nl) a [Chinese](https://chosenflex.com) +86 [contact](https://code.tuxago.com) number) and after that select the "DeepThink" alternative below the timely input box.<br>
|
||||||
|
<br>[DeepSeek](https://dichvudiennuoc247.vn) use the design by means of their API, [utilizing](https://www.printegadget.it) an [OpenAI-imitating endpoint](https://www.erdoganlargroup.com). You can access that through LLM by dropping this into your extra-openai-models. [yaml configuration](https://holic.vaslekarnik.sk) file:<br>
|
||||||
|
<br>Then run [llm secrets](https://gitlab.ujaen.es) set [deepseek](http://forums.indexrise.com) and paste in your API secret, then [utilize llm](http://www.milenakraft.com) -m deepseek-reasoner ['prompt'](http://www.organvital.com) to run [triggers](http://duberfly.com).<br>
|
||||||
|
<br>This will not reveal you the tokens, [regretfully](http://anytimefitness-ek.co.uk). Those are served up by the API (example here) however LLM doesn't yet have a method to show them.<br>
|
||||||
Loading…
Reference in New Issue