A LLM trained only on data from certain time periods to reduce modern bias
Python
Revise README with language options and model updates
Hayk Grigorian015004a1412a
diff --git a/README.md b/README.mdindex 76328b9..88fd6ce 100644--- a/README.md+++ b/README.md@@ -52,7 +52,7 @@ v2 built on llamaforcausallmEarly prompts show the model responding with 1800's language and behavior.Example: Prompt: "Who art Henry?" and it replied "I know that man, I have did not a black, the storm."-+- No mention of modern concepts- Mostly era-accurate vocabulary@@ -65,7 +65,7 @@ A significant improvement over v0.- Still high factual hallucination rate- OCR noise (“Digitized by Google”) still present in outputs-+### **v1**@@ -108,9 +108,9 @@ Charles DarwinECCEMACY. Sir, — The following case is interesting to me : — ITrained using a 90GB (112GB tokenized) datasetTrained to 182K steps--++## Datasets@@ -221,5 +221,6 @@ GPU: A100 SXM rented+