Releases: deepmodeling/dpgen
Releases · deepmodeling/dpgen
v0.10.0
New features:
- Support gromacs in model deviation (#427)
- Support DeePMD-kit v2 (#420)
- Added ABACUS interface for first principle calculations (#433)
- Allow assigning different activation functions to embedding and fitting nets (#461)
Enhancement
- Remove the support for deepmd-kit v0.x (#397 )
- Support for new dpdispatcher (#449)
- Example machine file for 4 GPUs on a local machine (#400 #412 )
- Better error message in too many failed FP jobs (#428 )
- Slurm customized flags (#368 )
- ssh session is not active (#382 )
- Retry slurm commands when failing (#381 )
- Store sftp and reduce the frequency of opening sftp (#387 )
- Taskwise restart for manual cuda device (#383 )
- Add Cu's init_surf examples. (#406)
- Add input files of enhanced sampling method (#446)
- Update CP2K new example and slurm-lsf-mix machine (#456)
Bug fixings:
- PBS error message (#396 )
- Bug
inter_type
in auto_test Lammps support (#423 ) - Bug in auto_test multicomponent surface systems (#378 )
- Too long argument list for
tar
command (#380 ) - force overwrite when using
gzip
(#384 ) - MPI execute error (#405 )
- Support hybrid descriptor (fix #458) (#459)
- Fix parameter error for LSF GPU tasks (#463)
- Fix NSW value in INCAR; fix local_root check (#464)
v0.9.2
v0.9.1
v0.9.0
v0.8.1
v0.8.0
New features:
- support pwmat
- support nopbc
- support init model parameter from old models
- support aniso kspacing for vasp
- support plm_path that defines CV in a separate file
- support checking bad fp confs
- support collect data
- support initializing from an existing model
- support simplify for LabeledSystem and System
- with accurate thresthold, decay number of fp tasks to 0 if it is already accurate enough
- support different activation func for different models
Updates:
- set default multiplicity to auto; add parameter fragment_guesses
- better suppport on deepmd-kit >= v1.1
- tag failures on allow_failure
- better cvasp support
- better ali support
- restart shell by uuid
- restart training if interrupted
- retry slurm commands when failing
Bug fixings:
- error in checking slurm status
- error in ensuring alive
- shell submit script should exit with code 1
- transfer type_map.raw. enforce the the same type_map in the training input.json
- bad surf configurations
- check if the length of batch size list is enough
- auto test user provided INCAR
- getting nframes in make_train
- fix issue #247 add = to flags starting with --
- nve doesn't set the initial temperature
0.7.0
v0.6.0
New features:
- support aws
- support electron temperature, as frame or atom parameter
- data or iteration statatistics by run/report
- build init surface via ase
Bug fixings:
- job_id log in pmap does not work in lazy-local model
- init_bulk cannot use batch system other than slurm