lopuhin

9 ms·

61.
by lopuhin 6y ago
Check out https://en.wikipedia.org/wiki/Immanuel_Velikovsky#Ideas
62.
by lopuhin 7y ago
Good point, assuming some extent of collapse is crucial, and the question is if different perplexities due to tokenization can happen in principle. You are right that in "Alice" vs. "A|lice" example we get the same perpl
63.
by lopuhin 7y ago
The paper rises a great point on tokenization affecting perplexity, that we can't compare perplexities of different tokenizers even re-normalizing taking token counts into account, say BPE vs word tokenization. This example nails it:
64.
by lopuhin 7y ago
Small corrections: > I’m focusing on chips designed for training TPU 1 is designed for inference AFAIK. > TPU v2: 45 TFLOPs I think it would be great to clarify that what is commonly referred to as "TPU v2" (e.g. on GCP pric
65.
by lopuhin 7y ago
From the release announcement: > Development will focus on tf.keras going forward. We will keep maintaining multi-backend Keras over the next 6 months, but we will only be merging bug fixes. So it's the end of Keras
66.
by lopuhin 8y ago
Sorry, just to clarify, is that BigQuery integration into Sheets your are talking about? Can't find it in the web UI.
67.
by lopuhin 8y ago
That's the quote from the article. Do you have any additional info on when that would become available? The link from another reply is completely unrelated, as of writing this.
68.
by lopuhin 8y ago
This is a product our team was working on for the last 1.5 y, excited for it to go live! Gives you articles and products from any web page in a uniform JSON format, combining data extracted from the web page (using both text, html structure
69.
by lopuhin 11y ago
I think you just sign in and click on a big green button "Go to Course" on the right - at least it worked for me with neuralnets course by Hinton, which is really great.