d13

10 ms·

61.
by d13 2y ago
This is superb: https://course.fast.ai/
62.
by d13 2y ago
Groq’s rate limits are so bad they’re only good enough for one person using it very rarely with an under 4k context size. It’s just a marketing stunt at the moment.
63.
by d13 2y ago
Try Lm Studio or Ollama. Load up the model, and there you go.
64.
by d13 2y ago
No, it’s bizarrely poorly written, even for an LLM.
65.
by d13 2y ago
Follow your interests.
66.
by d13 2y ago
Exactly. This is how it was always done and it works great.
67.
by d13 2y ago
This is fun and interesting: https://developers.cloudflare.com/workers-ai/tutorials/build...
68.
by d13 2y ago
I also remember that from the same book. The best piece of coding advice I’ve ever received.
69.
by d13 2y ago
Try it! All you need to do is type this into llama 3 8B or 70B 16fp instruct: “You are a text adventure game.” Done.
70.
by d13 2y ago
It’s just the instruct Llama 3 models that are censored. The base (text completion) models aren’t. You can turn the base models into uncensored instruct models very easily by simply providing them a handful of examples of how they should re
71.
Loyal Elfie: Use an LLM as a personal memory assistant(github.com)
2 points by d13 2y ago|0 comments
72.
by d13 2y ago
Abliterated Llama3?
73.
by d13 2y ago
Seconded. GPT4 can do this perfectly.
74.
by d13 2y ago
How does it compare against unequalised Llama 3 8B at 16fp? I’ve been using that locally and it’s almost replaced GPT4 for me. Runs in about 14GB of VRAM.
75.
by d13 2y ago
Why? General anaesthetic is fun! The last time I was under for 6 hours. For a few days afterwards I saw the most incredibly interesting, detailed persistent hallucinations whenever I closed my eyes. They completely vanished when I opened th
76.
by d13 2y ago
It was in the demo, but not in the current tts implementation, where it is very neutral.
77.
by d13 2y ago
I have a theory that the more you use ChatGPT, the worse it becomes due to silent rate limiting - farming the work out to smaller quantized versions if you ask it a lot of questions. I’d like to see if the results of these tests are the sam
78.
by d13 2y ago
Llama 3 8b Q5 is within GPT4 territory. I’ve been using it daily, it’s astonishing.
79.
by d13 2y ago
This was my implemented here. I tested it and it works: https://huggingface.co/hjhj3168/Llama-3-8b-Orthogonalized-ex...
80.
by d13 2y ago
You’re absolutely right, they are incredibly pragmatic.
81.
by d13 2y ago
Try this: https://github.com/mlabonne/llm-course
82.
by d13 2y ago
I’ve also found that Mistral instruct base is just as good, and gives less chatty replies. It all also gave me more consistently correct responses.
83.
by d13 3y ago
More about the Cobweb Palace: https://www.sanfranciscostory.com/the-cobweb-palace/amp/
84.
by d13 3y ago
You live on that platform, you die on that platform.
85.
by d13 3y ago
You may want to take a look at this too: https://github.com/sargam/digital
86.
by d13 3y ago
https://www.wired.com/story/battle-over-books3/
87.
by d13 3y ago
Video calls with screen sharing a couple of hours a day. Just as good as face-to-face
88.
by d13 3y ago
1. Book: Godot 4 Game Development Projects by Chris Bradfield. It’s superb and a great permanent reference. 2. GameDevTv’s Godot 4 course 3. The 4 videos currently on Godotneers Youtube channel. Those will get you there :)
89.
by d13 3y ago
And the biggest is at Longleat, where my wife’s job is to rescue people with a megaphone when they get lost. She also has to go hunting for a surprisingly large number of lost phones.
90.
by d13 3y ago
Wasn’t it Scheme?

More ›