Update README.md

README.md

@@ -20,7 +20,7 @@ This is a **LoRA adapter** trained on the [GSM8K](https://huggingface.co/dataset
 ## Model Details
 
 - **Base model**: [`apple/OpenELM-450M`](https://huggingface.co/apple/OpenELM-450M)
-- **Adapter type**: [LoRA](https://arxiv.org/abs/2106.09685) via [PEFT](https://github.com/huggingface/peft)
+- **Adapter type**: [LoRA](https://arxiv.org/abs/2106.09685) via [PEFT](https://github.com/huggingface/peft) (float32)
 - **Trained on**: GSM8K (math word problems)
 - **Languages**: English
 - **Model size**: ~450M parameters (base); adapter size is small
@@ -32,10 +32,8 @@ This is a **LoRA adapter** trained on the [GSM8K](https://huggingface.co/dataset
 
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
-from peft import PeftModel
 
 base_model = AutoModelForCausalLM.from_pretrained("apple/OpenELM-450M")
 tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neo-125M")
 
-model = PeftModel.from_pretrained(base_model, "your-username/openelm-450m-gsm8k-lora")