A LLM trained only on data from certain time periods to reduce modern bias
Python
Update README.md
Hayk Grigorian6e4d02443ad2
diff --git a/README.md b/README.mdindex 17e2c31..033a5bc 100644--- a/README.md+++ b/README.md@@ -71,7 +71,7 @@ Right now it produces sentences that lack full sentence structure and overall juThis is a nice improvement compared to the last model. The writing style and vocab is Victorian and almost every sentence is grammatically correct with proper punctuation. And again this is trained from scratch so it sticks to 1800's subjects.-+There are a lot of factual hallucinations. A lot (like 100%) of the details (dates, events, historical figures) are made up. Also the sentences don't really have connections to each other, sometimes maybe 2 sentences will relate to each other but beyond that they dont. Another issue is sometimes a stray “Digitized by Google” footer shows up, so the next time I train I really have to make sure the texts are cleaned well. Overall I'm very happy with the results, it's nowhere near an LLM yet but definitely a sentence generator.