Update README.md

README.md

@@ -9,7 +9,7 @@ datasets:
 
 # DistilProtBert
 
-Distilled version of [ProtBert](https://huggingface.co/Rostlab/prot_bert) model.
+Distilled version of [ProtBert-UniRef100](https://huggingface.co/Rostlab/prot_bert) model.
 In addition to cross entropy and cosine teacher-student losses, DistilProtBert was pretrained on a masked language modeling (MLM) objective  and it only works with capital letter amino acids.
 
 # Model description
@@ -17,9 +17,9 @@ In addition to cross entropy and cosine teacher-student losses, DistilProtBert w
 DistilProtBert was pretrained on millions of proteins sequences. 
 
 Few important differences between DistilProtBert model and the original ProtBert version are:
-1. Size of the model
-2. Size of the pretraining dataset
-3. Hardware used for pretraining
+1. Size of the model: 230M parameters (ProtBert has 420M parameters)
+2. Size of the pretraining dataset: ~43M proteins (ProtBert was pretrained on 216M proteins)
+3. Hardware used for pretraining: five v100 32GB Nvidia GPUs (ProtBert was pretrained on 512 16GB TPUs) 
 
 ## Intended uses & limitations
 
@@ -51,6 +51,5 @@ When fine-tuned on downstream tasks, this model achieves the following results:
 |   CB513   | 79 |    |
 |  DeepLoc  |    | 86 | 
 
-Distinguish between:
 
 ### BibTeX entry and citation info