songlab-cal/tape

• highreadme#1

  Reposition the README's opening to highlight datasets for training PLMs

  Why:

  CURRENT

  Data, weights, and code for running the TAPE benchmark on a trained protein embedding. We provide a pretraining corpus, five supervised downstream tasks, pretrained language model weights, and benchmarking code. This code has been updated to use pytorch - as such previous pretrained model weights and code will not work. The previous tensorflow TAPE repository is still available at https://github.com/songlab-cal/tape-neurips2019. This repository is *not* an effort to maintain maximum compatibility and reproducability with the original paper, but is instead meant to facilitate ease of use and future development (both for us, and for the community). Although we provide much of the same functionality, we have not tested every aspect of training on all models/downstream tasks, and we have also made some deliberate changes. Therefore, if your goal is to reproduce the results from our paper, please use the original code. Our paper is available at https://arxiv.org/abs/1906.08230. Some documentation is incomplete. We will try to fill it in over time, but if there is something you would like an explanation for, please open an issue so we know where to focus our effort! **Update 09/26/2020:** We no longer recommend trying to train directly with TAPE's training code. It will likely still work for some time, but will not be updated for future pytorch versions. Internally, we have been working with different frameworks for training (specifi

  COPY-PASTE FIX

  Copy

  Tasks Assessing Protein Embeddings (TAPE) provides a foundational benchmark, a pretraining corpus, and five supervised downstream tasks, serving as essential resources for *training* and evaluating protein language models. This repository offers the data, weights, and code necessary to run the TAPE benchmark, facilitating the development of new protein embeddings. While the provided datasets and tasks remain highly valuable for community use and development, please note that we no longer recommend trying to train directly with TAPE's specific training code, as it is not actively updated for future PyTorch versions. For reproducing results from the original paper, please use the original TensorFlow repository at https://github.com/songlab-cal/tape-neurips2019.

  OpenDoneDismiss
• mediumreadme#2

  Add a dedicated 'Key Resources' section to the README

  Why:

  COPY-PASTE FIX

  Copy

  ## Key Resources
  
  - **Pretraining Corpus:** A large corpus for training protein language models.
  - **Five Supervised Downstream Tasks:** A set of biologically relevant tasks for evaluating protein embeddings across different domains.
  - **Pretrained Language Model Weights:** Weights for various protein language models.
  - **Benchmarking Code:** Tools and scripts for running the TAPE benchmark.

  OpenDoneDismiss
• lowtopics#3

  Add 'protein-language-models' to the topics list

  Why:

  CURRENT

  benchmark, dataset, deep-learning, language-modeling, protein-sequences, protein-structure, pytorch, semi-supervised-learning

  COPY-PASTE FIX

  Copy

  benchmark, dataset, deep-learning, language-modeling, protein-language-models, protein-sequences, protein-structure, pytorch, semi-supervised-learning

  OpenDoneDismiss

Category GEO backends resolved for this scan: google/gemini-2.5-flash, deepseek/deepseek-v4-flash

• CATEGORY QUERY

  How to evaluate the performance of different protein sequence embedding models?

  you: #3
  AI recommended (in order):

  1. CAFA
  2. DeepFRI
  3. TAPE ← you
  4. ESM-2
  5. ProtT5-XL-UniRef50
  6. Ankh
  7. ProtTrans
  8. MSA Transformer
  9. SeqVec
  10. UniRep
  11. CASP
  12. STRING database
  13. BioGRID
  14. SCOP/CATH databases
  15. ProtBERT
  16. UMAP
  17. t-SNE
  18. ConSurf
  19. ESM-1b
  20. Captum
  21. TF-Explain
  Show full AI answer
• CATEGORY QUERY

  Where can I find datasets and tasks for training protein language models?

  you: not recommended
  AI recommended (in order):

  1. Hugging Face Datasets
  2. ProteinNet
  3. AlphaFold DB
  4. UniProtKB/Swiss-Prot
  5. UniRef
  6. PDB
  7. ESM-1b
  8. CASP

  AI recommended 8 alternatives but never named songlab-cal/tape. This is the gap to close.

  Show full AI answer

• Metadata completeness

  pass

• README presence

  pass

• Compared to common alternatives in this category, what is the core differentiator of songlab-cal/tape?

  pass
  ✓AI named songlab-cal/tape explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

• If a team adopts songlab-cal/tape in production, what risks or prerequisites should they evaluate first?

  pass
  ✓AI named songlab-cal/tape explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?

• In one sentence, what problem does the repo songlab-cal/tape solve, and who is the primary audience?

  pass
  ✓AI named songlab-cal/tape explicitly

  AI answers can be confidently wrong. Read for accuracy: does it match your actual tech stack, audience, and differentiator?