Glowies got the pastebin, so I threw this together based on the fragments I still had elsewhere.
JBs at the bottom because reasons.
Chatbot LLM regex for hiding chain of thought:
/(\n)?(\<think>.\<\/think>)\s/is
You need to go to the third button from the right up top in Sillytavern. Open Regex, go to global, make a global script and enter that text line. There was some other setting to make it not show up but I forgot what it was. Meh.
There should be other options listed in the main post, but for what it's worth:
You can go to kluster.ai and sign up for a free trial of DeepSeek. You used to get $100 of credit, then locusts found it and it was bumped to $5, then it got bumped even further down.
Or you can go to https://openrouter.ai/deepseek/deepseek-r1:free
Scroll down to "Create API key", name it whatever you want (and write it down when you create it because they don't show you the whole thing again), set it to unlimited, and then go to silly tavern and use these settings.
Make sure the provider is set to Chutes, and the model is specifically DeepSeek R1 (free)
https://files.catbox.moe/gf84cn.png
I haven't messed with the other free options yet.
For anyone who wants run deepseek-r1:32b 7b or whatever locally
Get the staging version of sillytavern. This one supports parsing out the thinking xml tags automatically.
Connect to your api: llama.cpp and ollama both work great. Make sure to use text completion and to select (and turn on) instruct template Deepseek v2.5.
Click the big "A" button and go to "Start reply with" and put something like this in the box:
Make sure to forget the closing think xml tag. This gaslights deepseek into thinking whatever fucked up shit. you said was already accepted by it.
Easy R1 Tutorial
- Download and use R1 in 3 steps
go to https://ollama.com/ and install Ollama
go to https://ollama.com/library/deepseek-r1, select size (7b-14b recommended), copy text on the right (Example: ollama run deepseek-r1)
right click windows icon on bottom left, run powershell (admin), paste the text
You can now enjoy AGI in complete privacy with no internet required.
- Wait, you want it to be even more convenient?
Now you can talk to AGI locally on a web browser!go to chrome web store, install Page Assit https://chromewebstore.google.com/detail/page-assist-a-web-ui-for/jfgfiigpkhlkbnfnbobbkinehhfdhndo
open Page Assit, go to settings on the top right, select your Ollama model in RAG setting - Wait, you want to create a submodel for each character to ERP with?
Typeopen code editor
Save file, open shell, typeFROM [your current R1 model name] (Example: FROM deepseek -r1:14b)
PARAMETER [0~1, 0 being most serious 1 being most creative]
SYSTEM """
[Character setting]
[Conversation rules]
Type ollama list and see if it appears, and simply select model in Page Assit or type in shell[folder location]>ollama create [desired model name] -f./[filename]
Now you can easily choose a preset model and ERP!ollama run [desired model name]
- Wait, you're a zoomer and is too dumb to read? We have video tutorials now
youtu.be/ihk0rsrjb-o [Open]
youtu.be/CysTlwjx2vw [Open] - Wait, you actually have a good computer with VRAM + RAM reaching 80GB? Try the new powerful DeepSeek R1 Dynamic!
- Wait, you are a complete tech illiterate?
You can now use the most powerful model online for free.
https://unsloth.ai/blog/deepseekr1-dynamic
Anti schizo stuff
Deepseek R1 jbs stolen from discord
deepseek_structure r1
jb:https://files.catbox.moe/cqbxqi.json
regex1:https://files.catbox.moe/vwowha.json
regex2:https://files.catbox.moe/ux547e.json
Nobody's Deepseek
jb:https://files.catbox.moe/wvb90b.json
3:35:34 No.104136194
104136176 #
Nobody's Deepseek
this is old, this is my current:
104136087 #
https://files.catbox.moe/m1dtey.json
DragoStableDeepseek
jb:https://files.catbox.moe/d31jlp.json