Nathan Frey

PyTorch normalizing flows

1

## 🚀 Feature A PyTorch implementation of normalizing flows for deep generative modeling and distribution learning. ## Motivation The current implementation for normalizing flows in DeepChem is written in TensorFlow...

Error: Fasta entry is invalid in createdb

11

## Expected Behavior I'm running `mmseqs createdb` on a large fasta file. I have checked that every entry is "valid" (only valid AA characters, can be read by biopython, there...

Add model performance profiling

8

- Add memory usage tracking - Add inference time metrics - Add embedding quality metrics - Add batch processing efficiency metrics

- Add embedding visualization - Add attention pattern visualization - Add sequence similarity visualization - Add model architecture visualization

Add efficient fine-tuning

2

- Add [QLoRA](https://huggingface.co/blog/AmelieSchreiber/esm2-ptm) with demos for [predicting binding sites and PTMs](https://www.biorxiv.org/content/10.1101/2023.11.13.566930v1)

Include PTM datasets

1

- Add [PTM-mamba](https://www.nature.com/articles/s41592-025-02656-9) datasets

Expand model test coverage

1

Add more tests for `lobster.model._ume.Ume` - Add tests for embed_sequences() method - Add tests for freeze() and unfreeze() methods - Add tests for different modality handling - Add tests for...

Strange Boolean Ternary

2

**Feedback:** `use_flash_attn: bool | None = None` is a tri-state Boolean. - [ ] Replace the tri-state Boolean with a clearer alternative: - Either make it strictly a `bool` with...

**Feedback:** Forward signature is neither clearly static nor dynamic. - [ ] Refine `forward()` method signature: - Decide between static: `forward(self, input_ids: Tensor, attention_mask: Tensor)` - Or dynamic: `forward(self, inputs:...