rjb7731

5 ms·

1.
by rjb7731 2y ago
Isn't this similar to the Deepmind paper on long form factuality posted a few days ago? https://arxiv.org/abs/2403.18802 https://github.com/google-deepmind/long-form-factuality/tree...
2.
by rjb7731 3y ago
I've previously noticed when playing with GPT-4 it can sometimes 'autocomplete' on different sections of the text its feeding back, sometimes what looks like 4 or more different sections. Might be unrelated but is this MoE in
3.
by rjb7731 3y ago
Its interesting this paper & the Orca LLM paper from Microsoft are using GPT3/4 model outputs to train 'powerful' models. Big question is will they allow your average joes/businesses to do the same on their own data?
4.
by rjb7731 3y ago
Thanks Jeremy for putting this course out for everyone, looking forward to completing it.
5.
by rjb7731 4y ago
Seems not too bad, you can tell its been trained on shorter responses. Example #1 Input: explain a neural network in simple terms Alpaca-Output: A neural network is a type of machine learning algorithm that is made up of layers of interconn
6.
by rjb7731 4y ago
With the 7B none-instruction tuned you have to really pre-prompt it to get good answers to things so like: SYSTEM: You are an AI research assistant. You use a tone that is technical and scientific. USER: Hello, who are you? ASSISTANT: Hi! I
7.
by rjb7731 4y ago
The inference on the gradio demo seems pretty slow, about 250 seconds for a request. Maybe I am too used to the 4-bit quant version now ha!
8.
by rjb7731 4y ago
if you look in the dev tools you will see a request to a 'join' file when you click the agree button it adds you to a queue. You can watch where you are up to in the dev tools.
9.
by rjb7731 4y ago
Even with patience, I can see nothing happening in the dev tools when i click agree? Edit: From looking when you click agree you can see a request to a 'join' file with a 101 status, looks like a queuing system and I'm 219th.
10.
by rjb7731 4y ago
interesting, looks like the web demo doesn't work at the moment though. The prompt.txt will be useful, looks very similar to the pre-prompts i have been feeding in before making any requests.
11.
by rjb7731 4y ago
Looks like it might be no bueno on google colab for now, chatbot.py takes prompts via input() too rather then a command line argument.
12.
by rjb7731 4y ago
I wish him the best, I find his streams incredibly useful and they motivate me to be better at solving problems and coding. Hopefully he still does work on AGI and makes tinygrad even cooler.