Skip to content

VizFold zarr archive for inference traces (issue #39), ksuleimanova#64

Closed
ksuleimanova wants to merge 56 commits into
AI2Science:mainfrom
ksuleimanova:archive_zarr
Closed

VizFold zarr archive for inference traces (issue #39), ksuleimanova#64
ksuleimanova wants to merge 56 commits into
AI2Science:mainfrom
ksuleimanova:archive_zarr

Conversation

@ksuleimanova

Copy link
Copy Markdown

Methods 8 and 9 mentioned in pr #62 completed in file vizfold_to_zarr.py
Hyperparameter optimization with grid search in file openfold_sweep_to_zarr.py

Elliptic461 and others added 30 commits October 7, 2025 23:37
Add instruction for how to get vizfold working.
…n directory

The cell magic change is so the need for the Bask kernel is removed. Instead, a generic Python kernel can be used instead. The Conda install dir change is so different directories can be used if your home directory is too small
Updated installation instructions for AlphaFold, including new data directory paths and changes to the installation process.
Update install.ipynb and README.md
Clarified instructions for loading Conda and using the OpenFold environment in Jupyter.
Update README with clearer Conda and Jupyter instructions
Update installation instructions for Jupyter kernel
Added verbose flag '-vvv' to various conda and pip commands for detailed output during installation.
Vishram1123 and others added 26 commits February 13, 2026 11:46
Added verbose logging to Conda activation steps.
Updated installation commands to reduce verbosity in output.
Enhance installation script with verbose output
- Implemented a Python pipeline to intercept and patch the pretrained `esmfold_3B_v1.pt` checkpoint.
- Resolved cross architecture weight incompatibilities (Meta ESM Issue #435) by mapping legacy Invariant Point Attention (IPA) tensor keys to align with the modern OpenFold/VizFold state dictionary (e.g., updating `.weight` to `.linear.weight`).
- Added an inference testing script to validate the environment setup and patched weights.
- Verified GPU accelerated forward pass execution using some test sequences which successfully output a properly formatted `.pdb` coordinate file.
… integration

Implements ESMFold inference as a proper vizfold/backends/ package that fully integrates with VizFold's existing attention capture pipeline, replacing the earlier esmfold/ prototype scripts.
Move mamba's command into env and add to readme
- Tested by removing my working conda environment and reinstalling using notebook.
- Covers OpenFold and ESMFold installation
@ksuleimanova

Copy link
Copy Markdown
Author

commits were overwritten from a wrong repository

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants