<?xml version="1.0"?>
<?xml-stylesheet type="text/css" href="http://www.vokipedia.de/skins/common/feed.css?303"?>
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="de">
		<id>http://www.vokipedia.de/index.php?action=history&amp;feed=atom&amp;title=Simon_Willison_s_Weblog</id>
		<title>Simon Willison s Weblog - Versionsgeschichte</title>
		<link rel="self" type="application/atom+xml" href="http://www.vokipedia.de/index.php?action=history&amp;feed=atom&amp;title=Simon_Willison_s_Weblog"/>
		<link rel="alternate" type="text/html" href="http://www.vokipedia.de/index.php?title=Simon_Willison_s_Weblog&amp;action=history"/>
		<updated>2026-05-07T08:36:49Z</updated>
		<subtitle>Versionsgeschichte dieser Seite in Vokipedia</subtitle>
		<generator>MediaWiki 1.19.23</generator>

	<entry>
		<id>http://www.vokipedia.de/index.php?title=Simon_Willison_s_Weblog&amp;diff=33541&amp;oldid=prev</id>
		<title>Lawanna33Z: Die Seite wurde neu angelegt: „&lt;br&gt;That model was [https://talentlagoon.com trained] in part using their [http://mumbai.rackons.com unreleased] R1 &quot;reasoning&quot; design. Today they've [https://…“</title>
		<link rel="alternate" type="text/html" href="http://www.vokipedia.de/index.php?title=Simon_Willison_s_Weblog&amp;diff=33541&amp;oldid=prev"/>
				<updated>2025-02-09T15:19:05Z</updated>
		
		<summary type="html">&lt;p&gt;Die Seite wurde neu angelegt: „&amp;lt;br&amp;gt;That model was [https://talentlagoon.com trained] in part using their [http://mumbai.rackons.com unreleased] R1 &amp;quot;reasoning&amp;quot; design. Today they&amp;#039;ve [https://…“&lt;/p&gt;
&lt;p&gt;&lt;b&gt;Neue Seite&lt;/b&gt;&lt;/p&gt;&lt;div&gt;&amp;lt;br&amp;gt;That model was [https://talentlagoon.com trained] in part using their [http://mumbai.rackons.com unreleased] R1 &amp;quot;reasoning&amp;quot; design. Today they've [https://git.sleepless.us released] R1 itself, in addition to a whole family of new [https://thutucnhapkhauthietbiyte.com.vn designs] obtained from that base.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;There's a lot of things in the new [https://skleplodz.com release].&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;DeepSeek-R1-Zero seems the [https://naturhome.sk base design]. It's over 650GB in size and, like many of their other releases, is under a tidy MIT license. [https://hogegaru.click DeepSeek caution] that &amp;quot;DeepSeek-R1-Zero comes across obstacles such as unlimited repetition, poor readability, and language blending.&amp;quot; ... so they also launched:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;DeepSeek-R1-which &amp;quot;integrates cold-start data before RL&amp;quot; and &amp;quot;attains performance equivalent to OpenAI-o1 across mathematics, code, and thinking jobs&amp;quot;. That a person is likewise MIT certified, and is a [http://cesao.it comparable size].&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;I do not have the [https://isabelleg.fr capability] to run [https://social.ishare.la models larger] than about 50GB (I have an M2 with 64GB of RAM), so neither of these two [https://eliwagroup.com designs] are something I can quickly play with myself. That's where the [https://benjiweatherley.com brand-new distilled] models are available in.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;To [https://xyzzy.company support] the research community, we have [https://gitea.aambinnes.com open-sourced] DeepSeek-R1-Zero, DeepSeek-R1, and six thick [https://www.gioiellimarotta.it models distilled] from DeepSeek-R1 based on Llama and Qwen.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;This is a [https://www.artico-group.com remarkable flex]! They have actually [http://esdoors.co.kr models 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).&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;[https://cm3comunicacao.com.br Weirdly] those [https://whitestoneenterprises.com Llama designs] have an MIT license attached, which I'm [https://terracochopp.com.br uncertain] works with the [http://sopchess.gr underlying Llama] license. [https://manobika.com Qwen designs] are [https://trafosistem.org Apache certified] so possibly MIT is OK?&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;(I likewise [https://sing.ibible.hk simply noticed] the MIT license files state &amp;quot;Copyright (c) 2023 DeepSeek&amp;quot; so they may need to pay a little bit more [http://www.pureatz.com attention] to how they copied those in.)&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;[https://evangelischegemeentehelmond.nl Licensing] aside, these [https://www.thaid.co distilled designs] are interesting beasts.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;[https://combat-colours.com Running] DeepSeek-R1-Distill-Llama-8B-GGUF&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;[https://mkii.jp Quantized versions] are currently beginning to show up. Up until now I have actually [https://gitea.oo.co.rs attempted] just one of those- unsloth/[https://operahorizon2020.eu DeepSeek-R] 1-Distill-Llama-8[https://nofox.ru B-GGUF launched] by [https://hurav.com Unsloth] [http://jobjungle.co.za AI]-and it's really [https://v-jobs.net enjoyable] to have fun with.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;I'm [http://www.bit-sarang.com running] it [http://masterofbusinessandscience.com utilizing] the [http://bufordfinance.com combination] of Ollama, LLM and  [https://links.gtanet.com.br/migueldlt085 links.gtanet.com.br] the [http://oceanblue.co.kr llm-ollama plugin].&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;First I [http://www.alisea.org fetched] and ran the design using Ollama itself:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;This [https://webshirewest.com downloads] around 8.5 GB of [http://pangclick.com design data] and starts an [https://alaskanoahsark.com interactive chat] [https://isourceprofessionals.com interface].&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;Once the design has been fetched, LLM can talk with it as well. I choose using LLM for [http://www.thesikhnetwork.com experiments] since it logs whatever to SQLite for later [https://sailingselkie.no exploration].&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;I set up the plugin and ran it like this:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;Each [https://www.drillionnet.com response] begins with a ... [http://autodealer39.ru pseudo-XML] tag containing the chain of thought used to help [https://www.muslimlove.com produce] the . For &amp;quot;a joke about a pelican and a walrus who run a tea room together&amp;quot; that began like this:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;Okay, so I need to come up with a joke about a [https://skleplodz.com pelican] and  [https://morphomics.science/wiki/User:NumbersBojorquez morphomics.science] a [https://www.cezae.fr walrus running] a [http://211.171.72.66 tea space] together. Hmm, that's an [https://energyworthonline.com.ng intriguing combination]. Let me think of how these two [https://cbcnhct.org characters] may engage in an [https://saopaulofansclub.com amusing circumstance].&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;First, let's consider their [http://slvfuels.net qualities]. [http://sanktnikolai.dk Pelicans] are [https://afitaconsultant.co.id understood] for their long beaks and Webbed feet, [https://www.columbusworldtravel.com typically] seen near the beach or water. [https://www.thaid.co Walruses] have big teeth, thick fur, and they're usually found in colder climates, like [https://tomnassal.com icebergs] or snowy areas. So, [http://bsss.kr integrating] these 2 into a [https://www.sposi-oggi.com tea space] [https://www.camedu.org setting] is currently an [https://mponlinecoaching.pt amusing] image.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt; [...]&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;It then thought for 20 [https://www.schoepamedien.de paragraphs] before [http://1.213.162.98 outputting] the joke! Here's the full [http://www.danielaievolella.com transcript] and here's the output at the end, after the tag:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;The Joke:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;A [https://lesprivatib.com pelican] and a [http://123.206.9.273000 walrus decide] to open a quaint little tea room together. The walrus, with its huge size, [https://h2bstrategies.com struggles] to [http://www.pureatz.com discover comfortable] chairs, so it rests on the table by accident, [http://forums.indexrise.com overturning] the teapot. Meanwhile, the pelican, trying to help,  [https://trademarketclassifieds.com/user/profile/2672496 trademarketclassifieds.com] uses its beak to place dishes on the table, [http://forum.artefakt.cz triggering] a couple of spills.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;After a series of [http://blog.slade.kent.sch.uk humorous] mishaps, the [http://chamer-autoservice.de walrus searches] for and states with a smile, &amp;quot;This isn't so fishy anymore.&amp;quot; The [https://www.primoconsumo.it pelican smirks] and says, &amp;quot;Maybe not, however we do have a great deal of krill in our tea!&amp;quot;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;Explanation:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;- Setup: [https://blogs2019.buprojects.uk Pelican] and Walrus run an unlikely [https://coatrunway.partners tea space]. Their sizes cause [https://www.iconiqstrings.com comedic chaos].&amp;lt;br&amp;gt;- Punchline: &amp;quot;This isn't so fishy anymore&amp;quot; is a pun on the [http://123.60.97.16132768 scenario] ending up being less [http://buat.edu.in unreasonable]. The [https://pedromartransportes.com.br follow-up joke] presents &amp;quot;krill,&amp;quot; a play on words that ties into both [https://vloghub.ru marine life] (&amp;quot;fish&amp;quot;) and  [https://asteroidsathome.net/boinc/view_profile.php?userid=762661 asteroidsathome.net] tea, making it [https://denmsk.ru amusing] while [https://www.martinfurniturestore.com fitting] the style.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;So yeah, the joke is [http://unidadeducativaprivada173.com.ar terrible]. But the [https://alaskanoahsark.com procedure] of getting there was such an interesting [https://www.posturiradio.net insight] into how these new [https://www.rosarossaonline.it designs] work.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;This was a fairly small 8B design. I'm looking [https://yankeegooner.net forward] to [https://97per.net experimenting] with the Llama 70B variation, which isn't yet available in a GGUF I can run with Ollama. Given the [http://www.hnyqy.net3000 strength] of Llama 3.3 70[https://www.diverraidiamante.it B-currently] my [http://www.silkbeautynails.nl favourite] GPT-4 [https://kv-work.com class model] that I have actually worked on my own [https://www.ppfoto.cz machine-I] have high [http://www.kristinogvibeke.com expectations].&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;Update 21st January 2025: I got this [http://www.criosimo.it quantized] version of that Llama 3.3 70B R1 [http://bedfordfalls.live distilled model] working like this-a 34GB download:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;Can it draw a [http://www.erlingtingkaer.dk pelican]?&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;I [https://thouartheretheatre.com attempted] my [http://www.indrom.com traditional Generate] an SVG of a [http://jatushome.myqnapcloud.com8090 pelican riding] a bike timely too. It did [https://orbit-tms.com refrain] from doing effectively:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;It aimed to me like it got the order of the elements incorrect, so I followed up with:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;the [https://psicologajessicasantos.com.br background wound] up [https://www.diy-ausstellung.de covering] the remainder of the image&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;It thought some more and provided me this:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;Just like the earlier joke, the chain of believed in the records was much more intriguing than the end outcome.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;Other [http://marcstone.de methods] to [http://viksanden.se attempt] DeepSeek-R1&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;If you wish to try the model out without installing anything at all you can do so [https://flohmarkt.familie-speckmann.de utilizing chat].[https://ronaldslater.com deepseek.com-you'll] need to [http://pamennis.com produce] an [https://geb-tga.de account] (check in with Google, [https://www.auto-secondhand.ro utilize] an [http://reinforcedconcrete.org.ua email address] or [https://alki-mia.com provide] a [https://talentlagoon.com Chinese] +86 phone number) and then choose the &amp;quot;DeepThink&amp;quot; [http://www.goetzschuerholz.com alternative listed] below the [https://jollyday.club timely input] box.&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;[https://git.viorsan.com DeepSeek provide] the design via their API, [http://lain.heavy.jp utilizing] an [https://mobitel-shop.com OpenAI-imitating endpoint]. You can access that by means of LLM by [https://www.gioiellimarotta.it dropping] this into your [https://nanake555.com extra-openai-models]. [http://git.jfbrother.com yaml setup] file:&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;&amp;lt;br&amp;gt;Then run [https://grunadmin.co.za llm secrets] set [https://www.vlmbusinessforum.co.za deepseek] and paste in your API key, then [https://zobecconstruction.com utilize llm] [https://selemed.com.pe -m deepseek-reasoner] [https://deoverkantontwerpers.com 'prompt'] to run [https://meetelectra.com prompts].&amp;lt;br&amp;gt; &amp;lt;br&amp;gt;&amp;lt;br&amp;gt;This won't show you the [https://combinationbeauty.com reasoning] tokens, [https://kilifiassembly.go.ke regretfully]. Those are dished out by the API (example here) but LLM doesn't yet have a method to display them.&amp;lt;br&amp;gt;&lt;/div&gt;</summary>
		<author><name>Lawanna33Z</name></author>	</entry>

	</feed>