Update README.md

README.md

@@ -7,6 +7,24 @@ tags: []
 
 <!-- Provide a quick summary of what the model is/does. -->
 
+## How to load the model
+
+This repository contains the weights and tokenizer for a circuit-broken "mistralai/Mistral-7B-Instruct-v0.2". 
+During training we added a padding token, so you should only use the model together with this the tokenizer in this repository to avoid errors.
+
+Example coding for loading:
+```
+model = AutoModelForCausalLM.from_pretrained(
+  "FelixHofstaetter/mistral-instruct-sandbagging-cb-new",
+  # other loading arguments
+)
+
+# only use the tokenizer from this repository for the model to avoid errors
+tokenizer = AutoTokenizer.from_pretrained(
+  "FelixHofstaetter/mistral-instruct-sandbagging-cb-new",
+  # other tokenizer arguments
+)
+```
 
 
 ## Model Details