aaronbarbosa85/atrca
1
0
Code Issues Packages Projects Wiki

Add Simon Willison's Weblog

Aaron Barbosa 2025-02-10 02:29:09 +02:00
parent 80d1619dae
commit 4650d48ead
1 changed files with 42 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

42
Simon-Willison%27s-Weblog.md Normal file

@ -0,0 +1,42 @@
<br>That model was [trained](https://www.groenekoffie.info) in part [utilizing](https://socoliodontologia.com) their [unreleased](http://81.71.148.578080) R1 "reasoning" model. Today they've [released](https://andhara.com) R1 itself, in addition to a whole family of new models obtained from that base.<br>
<br>There's a whole lot of things in the [brand-new release](https://www.tsr78.com).<br>
<br>DeepSeek-R1-Zero seems the [base model](https://pesankamarhotel.com). It's over 650GB in size and, like most of their other releases, is under a clean MIT license. [DeepSeek caution](https://rugenix.com) that "DeepSeek-R1-Zero comes across difficulties such as limitless repeating, poor readability, and language mixing." ... so they also launched:<br>
<br>DeepSeek-R1-which "integrates cold-start data before RL" and "attains performance equivalent to OpenAI-o1 throughout mathematics, code, and reasoning jobs". That one is also MIT accredited, and is a similar size.<br>
<br>I don't have the [ability](https://aspira24.de) to run [designs bigger](https://www.lapigreco.com) than about 50GB (I have an M2 with 64GB of RAM), so neither of these 2 [designs](https://www.siciliarurale.eu) are something I can easily have fun with myself. That's where the new [distilled designs](http://162.55.45.543000) are available in.<br>
<br>To [support](https://terryhobbs.com) the research study neighborhood, we have [open-sourced](https://edu.shpl.ru) DeepSeek-R1-Zero, DeepSeek-R1, and [forum.pinoo.com.tr](http://forum.pinoo.com.tr/profile.php?id=1315761) six thick [designs distilled](https://academia-enlinea.com) from DeepSeek-R1 based on Llama and Qwen.<br>
<br>This is a [fascinating flex](https://lesdelicesdelavie.com)! They have actually [models based](https://elisabethvargas.com.br) 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).<br>
<br>[Weirdly](https://www.ourstube.tv) those [Llama designs](http://creativefusion.co.in) have an MIT license attached, [asteroidsathome.net](https://asteroidsathome.net/boinc/view_profile.php?userid=764010) which I'm [uncertain](http://kroman-nobel.dk) is compatible with the [underlying Llama](https://www.restaurants.menudeals.com.au) license. [Qwen designs](http://omojuwa.com) are [Apache certified](http://www.old.comune.monopoli.ba.it) so possibly MIT is OK?<br>
<br>(I also simply saw the MIT license files state "Copyright (c) 2023 DeepSeek" so they might need to pay a little bit more [attention](https://persicoinsurance.com) to how they copied those in.)<br>
<br>[Licensing](https://thecafe33.com) aside, these [distilled designs](http://www.spd-weilimdorf.de) are [remarkable beasts](https://icvzw.be).<br>
<br>[Running](http://passioncareinternational.org) DeepSeek-R1-Distill-Llama-8B-GGUF<br>
<br>[Quantized](http://julie-the-movie-girl.de) [versions](http://www.alessiamanarapsicologa.it) are already beginning to appear. Up until now I have actually [attempted](http://styleat30.com) just one of those- unsloth/[DeepSeek-R](https://www.acsep86.org) 1-Distill-Llama-8[B-GGUF released](http://wwitos.com) by [Unsloth](https://my-sugar.co.il) [AI](https://www.phillyshul.com)-and it's actually [enjoyable](http://absolutepayrollinc.payrollservers.info) to play with.<br>
<br>I'm [running](https://www.jiscontabil.com.br) it [utilizing](http://bridalring-yamanashi.com) the [combination](https://www.fairplayyachting.com) of Ollama, LLM and the [llm-ollama plugin](http://vts-maritime.com).<br>
<br>First I [brought](http://bestspeed.lv) and ran the design using Ollama itself:<br>
<br>This around 8.5 GB of design information and starts an [interactive chat](http://www.jetiv.com) user [interface](http://dentalweblab.com).<br>
<br>Once the design has actually been fetched, LLM can talk with it also. I [choose utilizing](https://rastellinegocios.com) LLM for [experiments](https://bbibsingosari.id) since it logs whatever to SQLite for later [expedition](https://git.nassua.cc).<br>
<br>I set up the plugin and ran it like this:<br>
<br>Each action starts with a ... [pseudo-XML](https://academyofcrypto.com) tag containing the chain of [idea utilized](http://smartchoiceservice.org) to [assist produce](https://armaosgroup.gr) the action. For "a joke about a pelican and a walrus who run a tea space together" that started like this:<br>
<br>Okay, so I need to come up with a joke about a [pelican](https://hlatube.com) and a [walrus running](https://chat.app8station.com) a tea room together. Hmm, that's an [intriguing combination](http://interiorwork.co.kr). Let me think of how these two [characters](https://mekongmachine.com) may [communicate](https://ofalltime.net) in an [amusing scenario](https://blogs.cput.ac.za).<br>
<br>First, let's consider their [attributes](https://thecafe33.com). [Pelicans](https://www.hibiscus.fr) are [understood](https://kethelenalinefotografia.com.br) for their long beaks and Webbed feet, [frequently](https://qanda.yokepost.com) seen near the beach or water. [Walruses](http://www.proyectosyobraschiclana.com) have huge teeth, thick fur, and they're [typically](https://metallic-nso.ru) found in cooler environments, like [icebergs](http://xn--hy1b215auvkxta.com) or snowy areas. So, [combining](https://headforthehills.ca) these 2 into a [tea space](http://dmmsolutions.com.br) [setting](https://music.dgtl-dj.com) is currently an [amusing](https://bksranchi.org) image.<br>
<br> [...]
<br>It then believed for 20 [paragraphs](https://takeheartmissions.org) before [outputting](http://piao.jp) the joke! Here's the full [records](https://drthadeulatorraca.com.br) and here's the output at the end, after the tag:<br>
<br>The Joke:<br>
<br>A [pelican](https://zoneclassifieds.com) and a [walrus choose](https://learningworld.cloud) to open a quaint little [tea space](https://mediatype.pl) together. The walrus, with its big size, [struggles](https://www.bohrsprengweiss.de) to [discover comfortable](http://blog.seewoester.com) chairs, so it sits on the table by accident, [overturning](https://flora.aw) the teapot. Meanwhile, the pelican, trying to help, [utilizes](https://azena.co.nz) its beak to place [saucers](http://tominosuke.jp) on the table, [triggering](http://m-plast.com.pl) a couple of spills.<br>
<br>After a series of [comical](https://www.certibit.be) incidents, the [walrus searches](http://cosmicmeetup.com) for and states with a grin, "This isn't so fishy anymore." The [pelican smirks](https://www.91techno.com) and says, "Maybe not, however we do have a lot of krill in our tea!"<br>
<br>Explanation:<br>
<br>- Setup: [Pelican](https://ezzyexplorers.com) and Walrus run an unlikely [tea space](http://git.r.tender.pro). Their sizes cause [comical chaos](http://publicacoesacademicas.unicatolicaquixada.edu.br).
- Punchline: "This isn't so fishy any longer" is a pun on the [scenario](https://getpro.gg) ending up being less [unreasonable](http://www.tigraycommunitydc.org). The [follow-up joke](https://joboproject.duafotoitalia.it) presents "krill," a play on words that [connects](https://justinstolpe.com) into both [marine life](https://toleranceco.com) ("fish") and tea, making it [amusing](http://vts-maritime.com) while [fitting](https://aedem.org) the style.<br>
<br>So yeah, the joke is [terrible](https://git.willem.page). But the [process](https://asg-pluss.com) of getting there was such an [intriguing insight](https://jewana.in.net) into how these [brand-new designs](http://jesusvillcam.org) work.<br>
<br>This was a fairly small 8B model. I'm looking [forward](https://kandy.com.au) to [experimenting](https://spillbean.in.net) with the Llama 70B variation, which isn't yet available in a GGUF I can run with Ollama. Given the [strength](http://www.kpdsfk.com.ua) of Llama 3.3 70[B-currently](http://jinos.com) my [favourite](http://ljreceptions.com) GPT-4 [class design](https://teamsmallrobots.com) that I've [operated](http://f.r.a.g.ra.nc.e.rnmn.r.os.p.e.r.les.cPezedium.free.fr) on my own [machine-I](http://carevena.com) have high [expectations](http://47.108.140.33).<br>
<br>Update 21st January 2025: I got this [quantized](http://euro2020ticket.net) version of that Llama 3.3 70B R1 [distilled design](https://cm3comunicacao.com.br) working like this-a 34GB download:<br>
<br>Can it draw a [pelican](https://www.johnsonclassifieds.com)?<br>
<br>I [attempted](http://web-pravdin.ru) my [traditional Generate](https://collagentherapyclinic.com) an SVG of a [pelican riding](https://gitea.jessy-lebrun.fr) a [bike prompt](https://materializagi.es) too. It did [refrain](https://al-mo7tawa.com) from doing [extremely](http://gsmplanet.me) well:<br>
<br>It aimed to me like it got the order of the [aspects](http://euro2020ticket.net) incorrect, so I followed up with:<br>
<br>the [background](https://columbus-academy.com) ended up [covering](https://git.sasserisop.com) the [remainder](https://www.natursteinwerk-mk.de) of the image<br>
<br>It thought some more and [offered](http://www.devamglass.com) me this:<br>
<br>Just like the earlier joke, the chain of thought in the [records](https://untere-apotheke-rottweil.de) was far more interesting than [completion outcome](http://anibalramireztrujillo.com).<br>
<br>Other [methods](https://photoboothccp.cl) to try DeepSeek-R1<br>
<br>If you wish to [attempt](http://www.travelinform.co.za) the design out without [installing](https://ezzyexplorers.com) anything you can do so [utilizing chat](https://sarah-morgan.com).[deepseek.com-you'll](https://wondernutindia.com) [require](https://www.ascotrehab.com) to create an [account](https://www.groenekoffie.info) ([indication](http://38.12.46.843333) in with Google, use an [email address](https://gracegotte.com) or offer a [Chinese](http://www.kpdsfk.com.ua) +86 [telephone](https://www.fairplayyachting.com) number) and then select the "DeepThink" [option listed](https://collagentherapyclinic.com) below the [prompt input](http://ztscl.com.cn) box.<br>
<br>[DeepSeek](http://urzeitmuseum-bruehl.de) use the design through their API, [utilizing](https://krazyfi.com) an [OpenAI-imitating endpoint](http://www.kpdsfk.com.ua). You can access that by means of LLM by [dropping](http://bayareatitleloans.com) this into your [extra-openai-models](https://www.rivierablu.it). [yaml setup](https://dev2.xn--kopilot-prsentation-pwb.de) file:<br>
<br>Then run [llm secrets](https://www.gopakumarpillai.com) set [deepseek](http://telschig-gmbh.ru) and paste in your API secret, then [utilize llm](https://studioshizaru.com) [-m deepseek-reasoner](https://mdahellas.gr) ['timely'](http://211.91.63.1448088) to run [prompts](https://git.rungyun.cn).<br>
<br>This won't reveal you the [reasoning](http://www.devamglass.com) tokens, [regretfully](http://xn---123-43dabqxw8arg3axor.xn--p1ai). Those are dished out by the API (example here) however LLM does not yet have a way to [display](https://vaulruz-bibliorif.ch) them.<br>