Update README.md

README.md

@@ -3,13 +3,16 @@ license: apache-2.0
 library_name: transformers
 ---
 # RadVLM Model Card
-A Multitask Conversational Vision-Language Model for Radiology (paper: https://arxiv.org/abs/2502.03333)
+A Multitask Conversational Vision-Language Model for Radiology (paper: https://arxiv.org/abs/2502.03333). 
+
+Here, we provide the link to access RadVLM github repository and the inference code to use RadVLM once trained following the repo's instructions. 
+Instruction dataset will be shared in the future on the PhysioNet platform. 
+
 
 # Github repo
 The code for data curation, finetuning and evaluation is shared in the following github repo: https://github.com/uzh-dqbm-cmi/RadVLM.git
 
 
-
 ## Model Development
 
 - **Developed by**: KrauthammerLab, University of Zurich, ETH Zurich, Kyoto University of Applied Science, Kobe University, Swiss AI Initiative
@@ -163,6 +166,7 @@ def inference_radvlm(model, processor, image, prompt, chat_history=None, max_new
 
 ## Quick-Start: Multi-turn Demo 
 Below is a demonstration of how to utilize the inference_radvlm function in a multi-turn conversation. 
+For this you need to set the variable `model_id` with the path containing the model weights. 
 
 ```python
 import torch
@@ -172,8 +176,8 @@ import requests
 from io import BytesIO
 import numpy as np
 
- Initialize the model and processor
-model_id = "KrauthammerLab/RadVLM"
+# Initialize the model and processor
+model_id = "your/local/folder/with/RadVLM/weights"
 model = LlavaOnevisionForConditionalGeneration.from_pretrained(
     model_id, 
     torch_dtype=torch.float16,