Rebrand AIDO.* -> GB.* (model card)

README.md

@@ -1,21 +1,21 @@
 ---
 license: other
 ---
-# AIDO.Protein2StructureToken-16B
+# GB.Protein2StructureToken-16B
 
 [![License](https://img.shields.io/badge/license-GenBio_AI_Community_License-orange)](https://github.com/genbio-ai/ModelGenerator/blob/main/LICENSE)
 
-**AIDO.Protein2StructureToken-16B** is a fine-tuned version of [AIDO.Protein-16B](https://huggingface.co/genbio-ai/AIDO.Protein-16B), for protein structure prediction. 
-This model uses amino acid sequences as input to predict tokens that can be decoded into 3D structures by [AIDO.StructureDecoder](https://huggingface.co/genbio-ai/AIDO.StructureDecoder).
+**GB.Protein2StructureToken-16B** is a fine-tuned version of [GB.Protein-16B](https://huggingface.co/genbio-ai/GB.Protein-16B), for protein structure prediction. 
+This model uses amino acid sequences as input to predict tokens that can be decoded into 3D structures by [GB.StructureDecoder](https://huggingface.co/genbio-ai/GB.StructureDecoder).
 It surpasses existing state-of-the-art models, such as **ESM3-open**, in structure prediction tasks, demonstrating its robustness and capability in this domain.
 
 ## Model Architecture Details
 
-This model retains the architecture of AIDO.Protein-16B, a transformer encoder-only architecture with dense MLP layers replaced by sparse Mixture of Experts (MoE) layers. 
+This model retains the architecture of GB.Protein-16B, a transformer encoder-only architecture with dense MLP layers replaced by sparse Mixture of Experts (MoE) layers. 
 Each token activates 2 experts using a top-2 routing mechanism. A visual summary of the architecture is provided below:
 
 <center>
-  <img src="https://huggingface.co/genbio-ai/AIDO.Protein-16B/resolve/main/proteinmoe_architecture.png" alt="AIDO.Protein-16B Architecture" style="width:70%; height:auto;" />
+  <img src="https://huggingface.co/genbio-ai/GB.Protein-16B/resolve/main/proteinmoe_architecture.png" alt="GB.Protein-16B Architecture" style="width:70%; height:auto;" />
 </center>
 
 
@@ -54,7 +54,7 @@ The fine-tuning process used **0.4 trillion tokens**, using AlphaFold database w
 The input sequence should be single-chain amino acid sequences.
 
 - **Input Tokenization**: The sequences are tokenized at the amino acid level and terminated with a `[SEP]` token (id=34).
-- **Output Tokenization**: Each input token is converted into a structure token. The output can be decoded into 3D structures in PDB format using [AIDO.StructureDecoder](https://huggingface.co/genbio-ai/AIDO.StructureDecoder).
+- **Output Tokenization**: Each input token is converted into a structure token. The output can be decoded into 3D structures in PDB format using [GB.StructureDecoder](https://huggingface.co/genbio-ai/GB.StructureDecoder).
 
 ## Results
 
@@ -71,7 +71,7 @@ To reproduce the structure prediction results described above, follow these step
 2. Run the prediction command:
 
    ```bash
-   mgen predict --config experiments/AIDO.StructureTokenizer/protein2structoken_16b.yaml
+   mgen predict --config experiments/GB.StructureTokenizer/protein2structoken_16b.yaml
    ```
    This will pull the CASP14, CASP15, and CAMEO dataset from [genbio-ai/casp14-casp15-cameo-test-proteins](https://huggingface.co/datasets/genbio-ai/casp14-casp15-cameo-test-proteins), and predict the structure tokens from the amino acid sequence.
 
@@ -79,13 +79,13 @@ To reproduce the structure prediction results described above, follow these step
 
    ```bash
    # convert the predicted structures in tsv into one pt file
-   python experiments/AIDO.StructureTokenizer/struct_token_format_conversion.py logs/protein2structoken_16b/predict_predictions.tsv logs/protein2structoken_16b/predict_predictions.pt
+   python experiments/GB.StructureTokenizer/struct_token_format_conversion.py logs/protein2structoken_16b/predict_predictions.tsv logs/protein2structoken_16b/predict_predictions.pt
    # extract the codebook of the structure tokenizer
-   python experiments/AIDO.StructureTokenizer/extract_structure_tokenizer_codebook.py --output_path logs/protein2structoken_16b/codebook.pt
+   python experiments/GB.StructureTokenizer/extract_structure_tokenizer_codebook.py --output_path logs/protein2structoken_16b/codebook.pt
    ```
 5. Run the decoding command to get 3D structures in PDB format (currently this script only supports single GPU inference):
    ```bash
-   CUDA_VISIBLE_DEVICES=0 mgen predict --config experiments/AIDO.StructureTokenizer/decode.yaml \
+   CUDA_VISIBLE_DEVICES=0 mgen predict --config experiments/GB.StructureTokenizer/decode.yaml \
      --data.init_args.config.struct_tokens_datasets_configs.name=protein2structoken_16b \
      --data.init_args.config.struct_tokens_datasets_configs.struct_tokens_path=logs/protein2structoken_16b/predict_predictions.pt \
      --data.init_args.config.struct_tokens_datasets_configs.codebook_path=logs/protein2structoken_16b/codebook.pt
@@ -93,15 +93,15 @@ To reproduce the structure prediction results described above, follow these step
    The outputs are in `logs/protstruct_decode/protein2structoken_16b_pdb_files/`
 6. You can compare the predicted structures with the ground truth PDBs in [genbio-ai/casp14-casp15-cameo-test-proteins](https://huggingface.co/datasets/genbio-ai/casp14-casp15-cameo-test-proteins/tree/main).
 
-Alternatively, you can provide your own input amino acid sequence in a CSV file. Here is one example csv at `experiments/AIDO.StructureTokenizer/protein2structoken_example_input.csv` in `ModelGenerator`:
+Alternatively, you can provide your own input amino acid sequence in a CSV file. Here is one example csv at `experiments/GB.StructureTokenizer/protein2structoken_example_input.csv` in `ModelGenerator`:
 ```
 idx,aa_seq
 example,KEFWNLDKNLQLRLGIVFLG
 ```
 Here, `idx` is a unique name, and `aa_seq` is the amino acid sequence. To use this customized CSV file, replace the second step with
 ```bash
-mgen predict --config experiments/AIDO.StructureTokenizer/protein2structoken_16b.yaml \
- --data.init_args.path=experiments/AIDO.StructureTokenizer/ \
+mgen predict --config experiments/GB.StructureTokenizer/protein2structoken_16b.yaml \
+ --data.init_args.path=experiments/GB.StructureTokenizer/ \
  --data.init_args.test_split_files=[protein2structoken_example_input.csv]
 ```
 
@@ -111,7 +111,7 @@ For more information, visit: [Model Generator](https://github.com/genbio-ai/mode
 mgen fit --model SequenceClassification --model.backbone aido_protein_16b --data SequenceClassificationDataModule --data.path <hf_or_local_path_to_your_dataset>
 mgen test --model SequenceClassification --model.backbone aido_protein_16b --data SequenceClassificationDataModule --data.path <hf_or_local_path_to_your_dataset>
 ```
-The usage of this model is the same as [AIDO.Protein-16B](https://huggingface.co/genbio-ai/AIDO.Protein-16B). 
+The usage of this model is the same as [GB.Protein-16B](https://huggingface.co/genbio-ai/GB.Protein-16B). 
 You only need to change the `model.backbone` to `aido_protein2structoken`.
 
 
@@ -155,7 +155,7 @@ print(logits)
 ```
 
 ## Citation
-Please cite AIDO.Protein and AIDO.StructureTokenizer using the following BibTex codes:
+Please cite GB.Protein and GB.StructureTokenizer using the following BibTex codes:
 ```
 @inproceedings{zhang_balancing_2024,
 	title = {Balancing Locality and Reconstruction in Protein Structure Tokenizer},