Update README.md

README.md

@@ -6,13 +6,10 @@ tags: [pathology, vision-language, histopathology, lora, zero-shot]
 
 # Lumen
 
-Cross-modal alignment for a frozen [Virchow2](https://huggingface.co/paige-ai/Virchow2) vision encoder
-and a frozen [BioMedBERT](https://huggingface.co/microsoft/BiomedNLP-BiomedBERT-base-uncased-abstract-fulltext) text encoder.
+Lumen aligns frozen Virchow2 and BioMedBERT encoders using rank-4 LoRA adapters and projection heads into a shared 512-dimensional space.
+
+This repository contains only the trained alignment parameters (2,984,961 parameters; 12.0 MB). The backbone weights are loaded from their original repositories.
 
-This repository holds **only the alignment**: 2,984,961 parameters
-(12.0 MB) comprising rank-4 LoRA adapters on both
-encoders, two projection heads into a shared 512-dimensional space, and a learned
-logit scale.
 
 ## Usage
 
@@ -35,12 +32,13 @@ txt = torch.nn.functional.normalize(out["text_embeds"], dim=-1)
 logits = out["logit_scale"] * img @ txt.T
 ```
 
-The arguments are `images` and `tokens`, not the `pixel_values` and `input_ids`
-that most Hub models take. `encode_image` and `encode_text` are available if you
-want one tower at a time.
+images should have shape (B, 3, 224, 224) and be preprocessed using the Virchow2 transform.
+
+encode_image and encode_text can be used to run either encoder separately.
+
+`timm`, `transformers`, and `peft` are required.
 
 Both backbones are downloaded from their own repositories on the first call.
-`timm`, `transformers` and `peft` are required.
 
 ### The loading warning is expected
 
@@ -79,8 +77,3 @@ and what to do. In short, open its page on the Hub, accept the licence, then
 authenticate with `huggingface-cli login`. Already have the weights locally? Pass
 `vision_path=` or `text_path=` instead.
 
-## What this is
-
-The 740,721,664 backbone parameters withheld here are unmodified by
-training, so this adapter plus the public backbones reconstructs the evaluated
-model exactly.