Update logging messages in predict_structure function to reflect changes in ProteinTTT visualization. Adjusted labels for clarity in user interface regarding TTT steps.

Refactor create_3dmol_html function to streamline pLDDT visualization and update logging in predict_structure function. Removed deprecated code and improved clarity in progress messages for ESMFold and ProteinTTT predictions.

try to fix GPU task aborted

Update DEMO_SEQUENCE

Enhance create_3dmol_html function with improved pLDDT coloring and a detailed legend. Implement base64 encoding for HTML output to ensure compatibility in sandboxed environments. Update styling for better visual presentation.

back to the last version

Add progress bar

plddt alpha fold colors

Implement background logging for test-time training in predict_structure function. Added a threading mechanism to capture logs in real-time, enhancing user feedback during model training.

Add HTML progress bar for prediction steps in predict_structure function. Updated progress tracking to enhance user feedback during model inference and visualization preparation.

Refactor create_3dmol_html function to simplify HTML generation and improve pLDDT coloring visualization. Removed unnecessary imports and enhanced the structure for better readability.

Add base64 encoding for HTML visualization in create_3dmol_html function and enhance error handling during visualization creation.

fix vizualisation issue

40 sec

log with plddt per step

change hparams

change font

Reorganize advanced settings layout in app.py to improve user experience. Moved TTT Steps slider into the advanced settings section and adjusted example sequence parameters for better clarity.

Change demo seq

add settings with different hparam

log proteinTTT

change font

AlphaFold-like pLDDT coloring

remove unnecessary comments and streamline the lora-diffusion installation process

Fix UnboundLocalError in predict_structure by enabling eval_each_step in TTT configuration. This change addresses an upstream bug in ProteinTTT that could lead to errors during execution.

Refactor lora-diffusion installation in app.py to patch __init__.py before the first import, ensuring only the necessary .lora submodule is loaded. This change improves dependency management by preventing unnecessary imports of additional libraries.

Refactor lora-diffusion installation in app.py to patch __init__.py for minimal dependencies, ensuring only necessary components are loaded. Enhance predict_structure function to save baseline PDB files for download and create 3D visualizations. Remove torchvision from requirements.txt to streamline dependencies.

Refactor lora-diffusion installation in app.py to check for existing installation before attempting to install. Enhance verification step to ensure successful installation, improving error handling and user feedback.

add torchvision

Refactor app.py to streamline the installation process for openfold and lora-diffusion, incorporating a verification step post-installation. Update requirements.txt to clarify the need for --no-build-isolation during runtime installations, enhancing overall dependency management.

Refactor lora-diffusion installation in app.py to include verification after installation and update requirements.txt to clarify the need for --no-build-isolation during installation, enhancing dependency management.

Add cache invalidation for lora-diffusion installation in app.py to ensure immediate visibility of the new package. Update requirements.txt to reflect the direct installation of lora-diffusion.

Update requirements.txt to pin the ProteinTTT dependency to a specific commit hash, ensuring consistency and stability in the package version used.

Update app.py to modify the lora-diffusion installation command by adding the --no-build-isolation flag for improved compatibility. Adjust requirements.txt to reflect that lora-diffusion is now installed at runtime, streamlining dependency management.

Update requirements.txt to pin specific package versions for numpy, scipy, pandas, biopython, tqdm, pyyaml, requests, and click, ensuring consistency with the local working environment. Remove duplicate entries for pandas and biopython to streamline dependencies.

Refactor app.py to streamline package imports by removing numpy and outdated comments, and adjust demo sequence for clarity. Update requirements.txt to include lora-diffusion directly, enhancing dependency management.

Update lora-diffusion installation command in app.py to include --no-build-isolation flag, enhancing compatibility during package installation.

Update requirements.txt to add setuptools and wheel for improved package management, and remove lora-diffusion dependency to streamline the environment

Update app.py to remove outdated comments regarding package imports and add numpy as a runtime dependency in requirements.txt, ensuring compatibility and clarity in the codebase

Refactor GPU handling in app.py to support HF Spaces and local environments. Introduce a no-op fallback for spaces import when not in HF Spaces. Adjust prediction GPU duration from 120 to 60 seconds and remove theme specification from Blocks context manager for improved flexibility.

Reduce GPU duration in prediction logic from 300 to 120 seconds

Enhance app.py by adding runtime checks for openfold and lora-diffusion installations, ensuring necessary packages are available before import. Refactor comments for clarity and improve model loading efficiency. Remove deprecated dependencies from requirements.txt to streamline the environment.

Add installation check for lora-diffusion in app.py to ensure availability during runtime. This prevents silent failures during requirements processing and enhances user experience.

Refactor ESMFold model loading in app.py to pre-load on CPU at import time, improving performance by reducing GPU memory peak. Update prediction logic to utilize the pre-loaded model and adjust GPU handling for efficiency.

Upgrade C++ standard in app.py to C++17 for PyTorch 2.x compatibility, replacing hardcoded C++14 in CUDA flags.

Refactor setup.py patching in app.py to replace deprecated compute capabilities with modern architectures, ensuring compatibility with CUDA 12.x. Comment out GPU detection import for builds without GPU access.

Update requirements.txt to specify versions for omegaconf, biotite, einops, pytorch-lightning, transformers, diffusers, and huggingface-hub, ensuring compatibility with the current environment and dependencies.

Enhance openfold installation process in app.py by adding steps to clone, patch, and install the library, ensuring compatibility with CUDA 12.x and removing deprecated architecture flags.

Update app.py to set CUDA architectures for openfold installation, ensuring compatibility with CUDA 12.x and ZeroGPU hardware.

Remove Dockerfile, implement runtime installation of openfold in app.py, update README.md to reflect SDK and app file changes, and adjust requirements.txt to indicate openfold installation method.