feat(pt_expt): full model and refact the module output names of dpmodel backend#5243

Draft

wanghan-iapcm wants to merge 77 commits intodeepmodeling:masterfrom

wanghan-iapcm:feat-full-model

Commits on Feb 5, 2026

implement pytorch-exportable for se_e2_a descriptor
Han Wang
committed
better type for xp.zeros
Han Wang
committed

Commits on Feb 6, 2026

implement env, base_descriptor and exclude_mask, remove the dependency on pt backend.
Han Wang
committed
mv to_torch_tensor to common
Han Wang
committed
simplify __init__ of the NaiveLayer
Han Wang
committed
fix bug
Han Wang
committed
fix bug
Han Wang
committed
simplify init method of se_e2_a descriptor. fig bug in consistent UT
Han Wang
committed
restructure the test folders. add test_common.
Han Wang
committed
add test_exclusion_mask.py
Han Wang
committed
fix poitential import issue in test.
Han Wang
committed
correct __call__(). fix bug
Han Wang
committed
fix registration issue
Han Wang
committed
fix pt-expt file extension
Han Wang
committed
fix(pt): expansion of get_default_nthreads()
Han Wang
committed
fix bug of intra-inter
Han Wang
committed
fix bug of default dp inter value
Han Wang
committed
fix cicd
Han Wang
committed
feat: add support for se_r
Han Wang
committed
fix device of xp array
Han Wang
committed
fix device of xp array
Han Wang
committed
revert extend_coord_with_ghosts
Han Wang
committed
raise error for non-implemented methods
Han Wang
committed
restore import torch
Han Wang
committed
fix(pt,pt-expt): guard thread setters
Han Wang
committed
make exclusion mask modules
Han Wang
committed
fix(pt-expt): clear params on None
Han Wang
committed

Commits on Feb 7, 2026

fix bug
Han Wang
committed

Commits on Feb 8, 2026

utility to handel dpmodel -> pt_expt conversion
Han Wang
committed
fix to_numpy_array device
Han Wang
committed
chore(dpmodel,pt_expt): refactorize the implementation of embedding net
Han Wang
committed
feat: se_t and se_t_tebd descriptors for the pytroch exportable backend.
Han Wang
committed
fix bug
Han Wang
committed
refact: fitting net
Han Wang
committed
fix bug
Han Wang
committed
merge master
Han Wang
committed
Merge branch 'refact-auto-setattr' into refact-fitting-net
Han Wang
committed

Commits on Feb 9, 2026

feat(pt_expt): add fitting
Han Wang
committed

Commits on Feb 10, 2026

merge master
Han Wang
committed

Commits on Feb 11, 2026

merge with master
Han Wang
committed
fix the API consistency issue in descriptors
Han Wang
committed
feat: add stat for dpmodel's atomic model. implement atomic model for pt_expt
Han Wang
committed

Commits on Feb 12, 2026

merge master
Han Wang
committed
feat: full energy model (but not exportable)
Han Wang
committed
add missing file
Han Wang
committed
add missing file
Han Wang
committed
merge master
Han Wang
committed
fix test
Han Wang
committed
fix test
Han Wang
committed
fix test
Han Wang
committed
Merge branch 'feat-fitting' into feat-atomic-model
Han Wang
committed
merge with updated feat-atomic-model
Han Wang
committed
use torch_module to simplify the def of modules.
Han Wang
committed
simplify three autograd to one by vmap, which was made inpossible by jit
Han Wang
committed
export forward_lower, but not successful
Han Wang
committed

Commits on Feb 13, 2026

Commits on Feb 14, 2026

simplify the code
Han Wang
committed
Merge remote-tracking branch 'origin/feat-atomic-model' into feat-atomic-model
Han Wang
committed
fix bug
Han Wang
committed
fix issues
Han Wang
committed
Merge branch 'feat-atomic-model' into feat-full-model
Han Wang
committed
more careful check on the compute_or_load_stat
Han Wang
committed
merge master
Han Wang
committed

Commits on Feb 15, 2026

add guard for eval_descriptor and eval_fitting_last_layer
Han Wang
committed
fix issues
Han Wang
committed
remove eval_ hooks
Han Wang
committed
rm eval_return_middle_output
Han Wang
committed
make output of energy model compatible among backends
Han Wang
committed

Commits on Feb 16, 2026

fix bugs
Han Wang
committed
fix bugs
Han Wang
committed
implement rename for all models
Han Wang
committed
fixes
Han Wang
committed
fix bugs
Han Wang
committed