Nathan fm#79

Open

nathanchenseanwalter wants to merge 89 commits into

foundation_modelfrom

Commits on Mar 17, 2026

Removed the argument "batch_size" from the trainers.

renierts
authored and
renierts
committed
Bugfix in the dataset class. When iterating over movie configurations, the wrong configuration was used to find the correct signal name.

renierts
authored and
renierts
committed
Added base script for video reconstruction. Copied from Aza's branch for debugging purposes.

renierts
authored and
renierts
committed
Added base script for video reconstruction. Copied from Aza's branch for debugging purposes.

renierts
authored and
renierts
committed
Minor changes in the example scripts. More preprocessing options for the dataset class.

renierts
authored and
renierts
committed
Fixed a bug where the dataset class failed when using multiple workers and opening an H5 file prior to distributing the dataset across all workers.

renierts
authored and
renierts
committed
Lots of bugfixes in the dataset, trainer, and models.

renierts
authored and
renierts
committed
Extended checkpointing - the trainer stores now:

renierts
authored and
renierts
committed
Extended checkpointing - the trainer stores now:

renierts
authored and
renierts
committed
Adapted the other reconstruction scripts to match the new API.

renierts
authored and
renierts
committed
Bugfix in the dataset class. When splitting inputs and targets, I forgot to remove unused modalities. This follows the standard getitem function now.

renierts
authored and
renierts
committed
Prepared an option to preprocess movies. This has to be fully integrated!!!

renierts
authored and
renierts
committed
Added a baseline fusion transformer for latent space prediction.

renierts
authored and
renierts
committed
Foundation model (#56)

authored and
renierts
committed
Moved some remaining scripts to the correct subdirectories.

renierts
authored and
renierts
committed
Still working on preparing the dataset. This is not ready to push. Preparation to moving to Stellar.

renierts
authored and
renierts
committed
Updated the data loader. Bugfix for loading the correct slices from H5 files.
renierts
committed
Added scripts for data fetching in Omega.
renierts
committed
Added a documentation for setting up Globus CLI on Omega and start a simple file transfer.
renierts
committed
Updated README.md:
renierts
committed
More PTData to fetch.
renierts
committed
PEP-8 compatible code.
renierts
committed
Generalized make_preprocessing_stats.py and made the function compute_preprocessing_stats more transparent.
renierts
committed
A lot of bugfixes in the dataloader and prepare_data.py
renierts
committed
Many bugfixees in the dataset class and for computing preprocessing stats. This is still not efficient enough and causes memory issues.
renierts
committed
Speed-ups in data_loader.py.
renierts
committed
Speed-ups in the dataloader.
renierts
committed
drawing.py:
renierts
committed
Bugfix in processing methods of the dataloader:
renierts
committed
Added a separate baseline encoder for filterscopes (renamed fast_time_series_baseline.py to filterscope_baseline.py).
renierts
committed
Added a weighted loss to penalize target distributions.
renierts
committed
Modified the default parameters of some profile and time-series signals in data_loader.py
renierts
committed
Added CER related info to the dataset class and to the model factory.
renierts
committed
Merge remote-tracking branch 'origin/foundation_model' into dev-peter
renierts
committed
Added dummy perceiver stuff. Be careful - this is not structured nicely yet. Only work in progress.
renierts
committed

Commits on Mar 31, 2026

Added more RMP point names to the data fetching script.
renierts
committed

Commits on Apr 1, 2026

Updated all scripts according to the increased set of diagnostics and actuators we are using.
renierts
committed

Commits on Apr 2, 2026

Commits on Apr 7, 2026

TS profiles are now slow time series instead of profiles.
renierts
committed

Commits on Apr 13, 2026

Had to update all the profiles and slow time-series. The latent feature space is more compact now.
renierts
committed
Removed the argument "batch_size" from the trainers.

renierts
authored and
renierts
committed
Bugfix in the dataset class. When iterating over movie configurations, the wrong configuration was used to find the correct signal name.

renierts
authored and
renierts
committed
Added base script for video reconstruction. Copied from Aza's branch for debugging purposes.

renierts
authored and
renierts
committed
Minor changes in the example scripts. More preprocessing options for the dataset class.

renierts
authored and
renierts
committed
Fixed a bug where the dataset class failed when using multiple workers and opening an H5 file prior to distributing the dataset across all workers.

renierts
authored and
renierts
committed
Lots of bugfixes in the dataset, trainer, and models.

renierts
authored and
renierts
committed
Adapted the other reconstruction scripts to match the new API.

renierts
authored and
renierts
committed
Foundation model (#56)

authored and
renierts
committed
Moved some remaining scripts to the correct subdirectories.

renierts
authored and
renierts
committed
Updated the data loader. Bugfix for loading the correct slices from H5 files.
renierts
committed
Added scripts for data fetching in Omega.
renierts
committed
Added a documentation for setting up Globus CLI on Omega and start a simple file transfer.
renierts
committed
Updated README.md:
renierts
committed
More PTData to fetch.
renierts
committed
PEP-8 compatible code.
renierts
committed
A lot of bugfixes in the dataloader and prepare_data.py
renierts
committed
Many bugfixees in the dataset class and for computing preprocessing stats. This is still not efficient enough and causes memory issues.
renierts
committed
Speed-ups in data_loader.py.
renierts
committed
Speed-ups in the dataloader.
renierts
committed
Added a separate baseline encoder for filterscopes (renamed fast_time_series_baseline.py to filterscope_baseline.py).
renierts
committed
Updated preprocessing_stats. Here, the statistics are now pre-calculated for both, linear and log10 scale.
renierts
committed
Dev peter (#68) (#69)

committed
TS profiles are now slow time series instead of profiles.
renierts
committed
Had to update all the profiles and slow time-series. The latent feature space is more compact now.
renierts
committed
Merge branch 'dev-peter' of https://github.com/PlasmaControl/FusionAIHub into dev-peter
renierts
committed

Commits on Apr 23, 2026

Big changes. Now, the entire foundation model is trained jointly.
renierts
committed

Commits on Apr 24, 2026

Commits on May 4, 2026

Stage 2 is ready for video support.
renierts
committed

Commits on May 7, 2026

Commits on May 12, 2026

Made the dataset more efficient when it comes to DDP.
Peter Steiner
committed

Commits on May 13, 2026

Bugfixes for the multimodal foundation model. Had to account for missing data in the DDP case. And validation with DDP was too memory-consuming.
Peter Steiner
committed

Commits on May 14, 2026

Commits on May 15, 2026

Stage 2 can be used now.
Peter Steiner
committed
Increased model size to 50M parameters.
Peter Steiner
committed

Commits on May 20, 2026

Merge remote-tracking branch 'origin/dev-peter' into nathan_fm
Nathaniel Chen
committed

Commits on May 23, 2026

Add scaling levers (SDPA attn, gradient checkpoint) + memory probe.

Nathaniel Chen
and
claude
committed
memory improvemtns
Nathaniel Chen
committed