--- a
+++ b/documentation/inference.md
@@ -0,0 +1,63 @@
+# Inference Usage Instruction
+
+Merlin can be run by instantiating the model in PyTorch. Merlin weights are also publicly available on [HuggingFace](https://huggingface.co/stanfordmimi/Merlin).
+- Image/Text contrastive embeddings
+- Image-only embeddings (provide similar functionality to Google CT Foundation)
+
+For a better understanding of the phenotypes and their associated PheWAS attributes, please refer to the [phenotypes](phenotypes.csv) file.
+
+**Please see the [demo](demo.py) for programmatic examples.**
+
+#### Image/Text contrastive embeddings
+
+To get the image/text constrastive embeddings for inference, the breakdown is as follows:
+
+```python
+import torch
+from merlin import Merlin
+
+model = Merlin()
+model.eval()
+model.cuda()
+
+for batch in dataloader:
+    outputs = model(
+        batch["image"].to(device), 
+        batch["text"]
+        )
+```
+
+where `outputs` is a tuple:
+- `outputs[0]` : returns the constrative image embeddings (shape: [1, 512])
+- `outputs[1]` : returns the phenotype prediction (shape: [1, 1692])
+- `outputs[2]` : returns the constrative text embeddings (shape: [1, 512])
+
+#### Image-only embeddings
+
+```python
+import torch
+from merlin import Merlin
+
+model = Merlin(ImageEmbedding=True)
+model.eval()
+model.cuda()
+
+for batch in dataloader:
+    outputs = model(
+        batch["image"].to(device), 
+        )
+```
+
+where `outputs` is a tuple:
+- `outputs[0]` : returns the image embeddings (shape: [1, 2048])
+
+
+## 👨‍💻 Merlin Finetuning
+
+Since both Merlin’s model architecture and pretrained weights are provided, Merlin allows for straightforward finetuning in PyTorch VLM and vision-only pipelines. Additionally, Merlin was trained on a single NVIDIA A6000 GPU (with a Vision-Language batch size of 18), meaning finetuning can be performed even in compute-constrained environments.
+
+Merlin supports both Image/Text and Image-only finetuning. To perform finetuning, simply remove the following lines of code and train on your data:
+~~`model.eval()`~~  
+~~`model.cuda()`~~  
+
+For compute-efficient finetuning, we recommend using mixed-precision training and gradient accumulation.
\ No newline at end of file