Update README.md

README.md

@@ -11,7 +11,12 @@ tags:
 base_model: Salesforce/blip2-opt-2.7b
 ---
 # VLRM
-This repository contains the fine-tuned weights of BLIP-2. You can find the code in the [GitHub Repository](https://github.com/TODO)
+This repository contains the weights of BLIP-2 OPT-2.7B model fine-tuned by reinforcement learning method introduced in the paper [VLRM: Vision-Language Models act as
+Reward Models for Image Captioning](https://arxiv.com).
+
+The RL-tuned model is able to generate longer and more comprehensive descriptions with zero computational overhead compared to the original model. 
+
+You can find other details in the [GitHub Repository](https://github.com/papermsucode).
 # Running the model
 ## Option 1
 <details>
@@ -38,7 +43,7 @@ processor.decode(out[0], skip_special_tokens=True).strip()
 </details>
 
 ## Option 2
-Since the fine-tuned take only small part of the whole model, you could load only neccesary weights.
+Since the fine-tuned layers take small part of the whole model, you can first load the original model, then load the RL-tuned weights.
 <details>
 <summary> Step 1. Load the original model </summary>
 
@@ -64,7 +69,11 @@ processor.decode(out[0], skip_special_tokens=True).strip()
 
 <details>
 <summary> Step 2. Load the RL-tuned weights </summary>
-
+Available checkpoints:
+  
+ - `vlrm-blip2-opt-2.7b.pt` (VLRM in the paper)
+ - `vlrm-rs-blip2-opt-2.7b.pt` (VLRM-RS in the paper)
+  
 ```python
 from huggingface_hub import hf_hub_download
 finetuned_weights_state_dict = torch.load(hf_hub_download(repo_id="sashakunitsyn/vlrm-blip2-opt-2.7b", filename="vlrm-blip2-opt-2.7b.pt"))