From edc0432b3ac8ee8500fb29babdfe595166100e0c Mon Sep 17 00:00:00 2001
From: Adam Belfki <57205266+AdamBelfki3@users.noreply.github.com>
Date: Mon, 19 Jan 2026 21:29:07 -0500
Subject: [PATCH] model loading should use AutoModelForCausalLM

Updated model import for DroPE to use AutoModelForCausalLM.
---
 README.md | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/README.md b/README.md
index d1d877c..31db975 100644
--- a/README.md
+++ b/README.md
@@ -44,10 +44,10 @@ This project uses [Hydra](https://hydra.cc/) for configuration management. Hydra
 
 Our models follow HuggingFace's model loading API. For example, to load a DroPE model, you can use the following code:
 ```bash
-from transformers import AutoModel, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
 tokenizer = AutoTokenizer.from_pretrained('SakanaAI/Llama-2-7b-hf-DroPE', trust_remote_code=True)
-model = AutoModel.from_pretrained('SakanaAI/Llama-2-7b-hf-DroPE', trust_remote_code=True, torch_dtype=torch.bfloat16)
+model = AutoModelForCausalLM.from_pretrained('SakanaAI/Llama-2-7b-hf-DroPE', trust_remote_code=True, torch_dtype=torch.bfloat16)
 ```
 
 Inference is then straightforward: