v3.0.2

What's Changed

This patch version only contains minor features, bug fixes, enhancements, and documentation improvements.

New features

• feat(tf): support tensor fitting with hybrid descriptor by @njzjz in #4542

Enhancement

• Perf: replace unnecessary torch.split with indexing by @caic99 in #4505
• Perf: use F.linear for MLP by @caic99 in #4513
• chore: improve neighbor stat log by @njzjz in #4561
• chore: bump pytorch to 2.6.0 by @njzjz in #4575

Bugfix

• Fix: Modify docs of DPA models by @QuantumMisaka in #4510
• fix(pt): fix clearing the list in set_eval_descriptor_hook by @njzjz in #4534
• [fix bug] load atomic_*.npy for tf tensor model by @ChiahsinChu in #4538
• fix: lower num_workers to 4 by @caic99 in #4535
• fix: fix YAML conversion by @njzjz in #4565
• fix(cc): remove C++ 17 usage by @njzjz in #4570
• Fix version in DeePMDConfigVersion.cmake by @RMeli in #4577
• fix(pt): detach computed descriptor tensor to prevent OOM by @njzjz in #4547
• fix(pt): throw errors for GPU tensors and the CPU OP library by @njzjz in #4582
• use variable to store the bias of atomic polarizability by @Yi-FanLi in #4581
• Fix: pt tensor loss label name by @anyangml in #4587
• CI: pin jax to 0.5.0 by @njzjz in #4613
• fix(array-api): fix xp.where errors by @njzjz in #4624

Documentation

• docs: fix the header of the scaling test table by @njzjz in #4507
• docs: add sphinx.configuration to .readthedocs.yml by @njzjz in #4553
• docs: add v3 paper citations by @njzjz in #4619
• docs: add PyTorch Profiler support details to TensorBoard documentation by @caic99 in #4615

CI/CD

• CI: switch linux_aarch64 to GitHub hosted runners by @njzjz in #4557

New Contributors

• @QuantumMisaka made their first contribution in #4510
• @RMeli made their first contribution in #4577

Full Changelog: v3.0.1...v3.0.2