feat(tf2): add training workflow - #5735

Merged

wanghan-iapcm merged 14 commits into

deepmodeling:masterfrom

njzjz:feat/tf2-train-current

Jul 6, 2026

Commits on Jul 4, 2026

Commits on Jul 5, 2026

fix(tf2): keep atomic virial disabled in training
njzjz
committed
6a644aa
perf(tf2): jit lower forward with DP_JIT
njzjz
committed
8defd7c
feat(tf2): support training compile option
njzjz
committed
84657a4
fix(tf2): address training review comments
njzjz
committed
1f1cf09
perf(tf2): compile prepared energy train step
njzjz
committed
f23d383
fix(tf2): enable compiled step for all outputs
njzjz
committed
51679c2
fix(tf2): avoid lower derivative kwarg mismatch
njzjz
committed
2bc8085
fix(tf2): reject empty checkpoint directories
njzjz
committed
d721c65
fix(ci): preload paddle before tf in pytest
njzjz
committed
28a27d3
test(tf2): avoid tensorboard dependency in summary test
njzjz
committed
86f73f9