A LLM trained only on data from certain time periods to reduce modern bias
Python
Update README.md
Hayk Grigorian96979a887c7f
diff --git a/README.md b/README.mdindex b9689dc..d9f3c6d 100644--- a/README.md+++ b/README.md@@ -178,16 +178,25 @@ For this project I'm trying to create a language model that is unclouded from moI'm using books, legal documents, newspapers, and other writings from 1800–1850 London. The list I linked has like 200 but for the first training I just used 50 files about ~187 MB. You can view a list of the documents:https://github.com/haykgrigo3/TimeCapsuleLLM/blob/main/Copy%20of%20London%20Documents%20for%20Time%20Capsule%20LLM.txt-## How large is the Version 0 model ?+## How large are the models ?-This model is very small right now, I'm just doing this for fun and following a strict training rule of no modern sources. It has almost 16 million parameters but I'm gonna start gathering more old texts to begin another model training. Will give updates as I go.+V0: 16M Parameters-## Training Specs ?+V0.5 123M Parameters+V1: 700M Parameters++# Training Specs ?++#V0/V0.5GPU: Geforce rtx 4060CPU: i5-13400FRam: 16GB DDR5.+#V1+GPU: A100 rented++