From bcc890fe8e4b9803e9eda1c489fc44171182d2b5 Mon Sep 17 00:00:00 2001
From: Olivv <133010426+olivv-cs@users.noreply.github.com>
Date: Tue, 10 Jun 2025 17:09:43 +0300
Subject: [PATCH] Update README.md

---
 README.md | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/README.md b/README.md
index 924c26fb..0efc1dde 100644
--- a/README.md
+++ b/README.md
@@ -17,6 +17,9 @@ Inputs:
 - pretrained_path - !MODEL WEIGHTS WILL NOT LOAD FROM THERE! Provide a HuggingFace path for config and tokenizer for your model;
 - system_prompt - your system prompt that Instruct model is going to be using.
 
+Technically speaking, it's possible to load just any model as instruct one. It might not even be an instruct model.
+Just make sure whatever you are loading as "llama_instruct_model_name" is matching with config provided in "pretrained_path".
+
 Outputs:
 Just CLIP. Pass it through your nodes like you will do with regular DualCLIPLoader.