refact: the DPA1 descriptor

[iProzd]

This PR reformat DPA1 descriptor consistent for TF/PT/DP

• Clean up the implementation of DPA1 descriptor in TF/PT
• Add numpy implementation for DPA1 descriptor
• Add serialize and deserialize for DPA1 descriptor in TF/PT/DP
• Add consistence test for TF/PT/DP

Now DPA1 descriptor in TF and PT&DP are consistent when:
smooth_type_embdding == False
env_protection == 0.0
stripped_type_embedding == False

DPA1 descriptor in PT and DP are always consistent.

Need to fix

• consistence in TF: when excluded_types != [] and attn_layer > 0.

[github-advanced-security]

CodeQL found more than 20 potential problems in the proposed changes. Check the Files changed tab for more details.

[codecov]

Codecov Report

Attention: Patch coverage is 89.03509% with 100 lines in your changes are missing coverage. Please review.

Project coverage is 82.12%. Comparing base (1985631) to head (2a67cea).

Files	Patch %	Lines
deepmd/dpmodel/descriptor/dpa1.py	88.81%	35 Missing ⚠️
deepmd/tf/descriptor/se_atten.py	81.72%	34 Missing ⚠️
deepmd/dpmodel/utils/network.py	85.71%	10 Missing ⚠️
deepmd/pt/model/descriptor/se_atten.py	95.43%	10 Missing ⚠️
deepmd/pt/model/descriptor/dpa1.py	84.61%	6 Missing ⚠️
deepmd/pt/model/network/layernorm.py	93.61%	3 Missing ⚠️
deepmd/tf/utils/network.py	91.66%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##            devel    #3696      +/-   ##
==========================================
+ Coverage   81.89%   82.12%   +0.23%     
==========================================
  Files         509      511       +2     
  Lines       46591    47364     +773     
  Branches     2952     2952              
==========================================
+ Hits        38156    38899     +743     
- Misses       7542     7572      +30     
  Partials      893      893              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[wanghan-iapcm]

stripped type embedding should be added in a future PR.

[coderabbitai]

Actionable comments posted: 0

Out of diff range and nitpick comments (1)

deepmd/dpmodel/descriptor/dpa1.py (1)

279-285: Initialization of TypeEmbedNet with hardcoded activation function 'Linear'. Consider making this configurable if flexibility is required.

[coderabbitai]

Actionable comments posted: 8

Out of diff range and nitpick comments (3)

deepmd/dpmodel/descriptor/dpa1.py (2)

206-206: Ensure clarity on parameter usage in the constructor.

It's important to clearly document which parameters are in use and which are kept for compatibility. This helps maintain clarity and prevent confusion for future developers or users of the API.

---

614-664: Ensure proper handling of attention layers.

When implementing complex layers like NeighborGatedAttention, it's crucial to ensure that each layer is handled correctly, especially in serialization and deserialization processes to maintain state consistency across sessions.

deepmd/tf/descriptor/se_atten.py (1)

3-3: Consider removing unused imports to keep the code clean and maintainable.

[njzjz]

LGTM

Some are unnecessary changes requested.