pd: support dpa3 with paddle backend

[HydrogenSulfate]

support dpa3 with paddle backend(eager mode)

1. training curve

2. accuracy

torch

paddle(slightly better than torch)

3. The main modifications in this PR include:

1. Added DPA-3 code and related modules based on the Paddle backend.
2. Added the EnergyHessianStdLoss module based on the Paddle backend.
3. Discovered that Paddle’s ParameterList does not support assignment of Tensors using the equals sign. Therefore, I added support for this feature at [API] Support assign Tensor to ParameterList PaddlePaddle/Paddle#72190. However, considering version compatibility, deepmd still uses paddle.assign for assignments.
4. Fixed an issue in env_mat_stat.py where the return type was Tensor instead of float.
5. The SiLUT used APIs from the numpy series that do not support paddle.Tensor, so I replaced them with Paddle’s native APIs. Additionally, to temporarily bypass issues with dynamic-to-static control flow, I changed the if-else branch in SiLUT.forward to a single branch.

Summary by CodeRabbit

• New Features

  • Introduced a new descriptor, DPA3, for advanced molecular simulations, including its integration and public availability.
  • Added support for a new graph-based neural network layer and descriptor block for RepFlow calculations.
  • Enabled Hessian loss computation for enhanced training capabilities.
  • Added new learning rate utility.

• Bug Fixes

  • Improved tensor shape handling and assignments for better compatibility and stability.

• Tests

  • Added comprehensive tests for the new DPA3 descriptor, including consistency, JIT, and multitask scenarios.
  • Expanded test coverage for model permutation and smoothness with DPA3.
  • Enhanced tests for DPA2 with CINN compiler support.

• Refactor

  • Standardized tensor shape definitions and updated method signatures for improved clarity and type safety.

• Chores

  • Updated public interfaces to include new features and descriptors.

[coderabbitai]

Actionable comments posted: 1

♻️ Duplicate comments (4)

source/tests/pd/model/test_compressed_descriptor_se_atten.py (1)

27-45: Duplicated test utility function.

This eval_pd_descriptor function appears to be duplicated across multiple test files (also found in test_compressed_descriptor_se_a.py). Consider extracting it to a common test utilities module to promote code reuse and maintainability.

source/tests/pd/model/test_compressed_descriptor_se_a.py (3)

27-45: Duplicated test utility function.

This eval_pd_descriptor function appears to be duplicated across multiple test files (also found in test_compressed_descriptor_se_atten.py). Consider extracting it to a common test utilities module to promote code reuse and maintainability.

---

27-45: Add validation for natoms to prevent potential IndexError.

In eval_pd_descriptor, the code reads natoms[0] directly at line 39. If natoms is empty, this could raise an IndexError. Consider adding an assertion or input check.

---

104-128: Consider documenting the magic value in the compression ratio.

The compression ratio of 0.5 is passed to enable_compression() without any explanation. Consider adding a comment explaining the significance of this value or making it a named constant to improve code readability.

🧹 Nitpick comments (3)

source/tests/pd/model/test_compressed_descriptor_se_atten.py (2)

109-133: Consider documenting the magic value in the compression ratio.

The compression ratio of 0.5 is passed to enable_compression() without any explanation. Consider adding a comment explaining the significance of this value or making it a named constant to improve code readability.

-        self.se_atten.enable_compression(0.5)
+        # Use 0.5 as a reasonable compression ratio for testing
+        # (typically values between 0.3-0.7 provide a good balance)
+        COMPRESSION_RATIO = 0.5
+        self.se_atten.enable_compression(COMPRESSION_RATIO)

---

109-133: Consider expanding test coverage for compression functionality.

The current test only verifies that compression doesn't significantly change the output, but it doesn't verify if the compression is actually working as expected (e.g., by checking memory usage, performance, or other compression-specific metrics). Consider adding additional assertions or tests to verify the compression effectiveness.

source/tests/pd/model/test_compressed_descriptor_se_a.py (1)

104-128: Consider expanding test coverage beyond shape equality.

While the test correctly verifies that compressed and uncompressed outputs have the same shape and similar values, it would be beneficial to add tests for additional scenarios:

1. Test with different compression ratios to ensure robustness
2. Test error handling (e.g., calling enable_compression twice)
3. Test with edge cases (e.g., very small or large distance values)

This would provide more comprehensive testing of the compression functionality.

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 06a6180 and 4a4129d.

📒 Files selected for processing (3)

• source/tests/pd/model/test_compressed_descriptor_dpa2.py (1 hunks)
• source/tests/pd/model/test_compressed_descriptor_se_a.py (1 hunks)
• source/tests/pd/model/test_compressed_descriptor_se_atten.py (1 hunks)

🚧 Files skipped from review as they are similar to previous changes (1)

• source/tests/pd/model/test_compressed_descriptor_dpa2.py

🧰 Additional context used

🧬 Code Graph Analysis (2)

source/tests/pd/model/test_compressed_descriptor_se_a.py (2)

deepmd/pd/model/descriptor/se_a.py (9)

• DescrptSeA (77-430)
• mixed_types (155-159)
• mixed_types (569-579)
• get_rcut (123-125)
• get_rcut (533-535)
• get_sel (135-137)
• get_sel (545-547)
• enable_compression (238-283)
• enable_compression (671-705)

source/tests/pd/model/test_compressed_descriptor_se_atten.py (3)

• eval_pd_descriptor (27-45)
• setUp (50-107)
• test_compressed_forward (109-133)

source/tests/pd/model/test_compressed_descriptor_se_atten.py (1)

source/tests/pd/model/test_compressed_descriptor_se_a.py (3)

• eval_pd_descriptor (27-45)
• setUp (50-102)
• test_compressed_forward (104-128)

⏰ Context from checks skipped due to timeout of 90000ms (29)

• GitHub Check: Test Python (6, 3.12)
• GitHub Check: Test Python (6, 3.9)
• GitHub Check: Test Python (5, 3.12)
• GitHub Check: Test Python (5, 3.9)
• GitHub Check: Test Python (4, 3.12)
• GitHub Check: Build wheels for cp310-manylinux_aarch64
• GitHub Check: Test Python (4, 3.9)
• GitHub Check: Build wheels for cp311-win_amd64
• GitHub Check: Test Python (3, 3.12)
• GitHub Check: Test Python (3, 3.9)
• GitHub Check: Build wheels for cp311-macosx_arm64
• GitHub Check: Build C++ (clang, clang)
• GitHub Check: Test Python (2, 3.12)
• GitHub Check: Build wheels for cp311-macosx_x86_64
• GitHub Check: Build C++ (rocm, rocm)
• GitHub Check: Analyze (python)
• GitHub Check: Test Python (2, 3.9)
• GitHub Check: Build wheels for cp311-manylinux_x86_64
• GitHub Check: Build C++ (cuda120, cuda)
• GitHub Check: Test C++ (false)
• GitHub Check: Test Python (1, 3.12)
• GitHub Check: Build C library (2.14, >=2.5.0rc0,<2.15, libdeepmd_c_cu11.tar.gz)
• GitHub Check: Build wheels for cp311-manylinux_x86_64
• GitHub Check: Build C++ (cuda, cuda)
• GitHub Check: Analyze (c-cpp)
• GitHub Check: Test C++ (true)
• GitHub Check: Test Python (1, 3.9)
• GitHub Check: Build C library (2.18, libdeepmd_c.tar.gz)
• GitHub Check: Build C++ (cpu, cpu)

🔇 Additional comments (2)

source/tests/pd/model/test_compressed_descriptor_se_atten.py (1)

48-107: The test class properly initializes the descriptor with attention mechanism.

The test class correctly sets up the DescrptDPA1 descriptor with appropriate parameters including attention-specific parameters (attn=8, attn_layer=0, tebd_input_mode="strip"), which is essential for testing the DPA1 descriptor with attention mechanism.

source/tests/pd/model/test_compressed_descriptor_se_a.py (1)

57-102: Selection sizes for descriptor are properly defined.

The self.sel = [9, 10] parameter correctly initializes the smooth version of the descriptor with appropriate selection sizes, which is crucial for the atomic environment representation.

[coderabbitai]

Actionable comments posted: 0

♻️ Duplicate comments (1)

deepmd/pd/utils/tabulate.py (1)

54-55: Avoid mutable default argument and function call in default argument.

As previously noted, passing [] and calling ActivationFn("tanh") directly in default arguments can lead to confusing bugs or unexpected behavior with shared state. Use a None default and initialize these values inside __init__.

🧰 Tools

🪛 Ruff (0.8.2)

54-54: Do not use mutable data structures for argument defaults

Replace with None; initialize within function

(B006)

---

55-55: Do not perform function call ActivationFn in argument defaults; instead, perform the call within the function, or read the default from a module-level singleton variable

(B008)

🧹 Nitpick comments (2)

deepmd/pd/utils/tabulate.py (2)

116-202: Reduce complexity of _make_data or break it into smaller methods.

This method is quite lengthy with multiple special cases in if/elif blocks. Splitting certain paths (e.g., conditionals for layer == 0, self.neuron[0] == 1/2, etc.) into helper methods could significantly improve maintainability and readability.

---

285-295: Add or expand docstrings for _layer_0 and _layer_1.

Both _layer_0 and _layer_1 perform targeted transformations with slightly different logic and return shapes—especially _layer_1, which returns a tuple. Documenting each parameter, the return shape, and expected usage would help future readers understand these methods more quickly.

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between 4a4129d and bf98888.

📒 Files selected for processing (1)

• deepmd/pd/utils/tabulate.py (1 hunks)

🧰 Additional context used

🧬 Code Graph Analysis (1)

deepmd/pd/utils/tabulate.py (2)

deepmd/utils/tabulate.py (4)

• BaseTabulate (19-458)
• _get_table_size (380-396)
• _get_data_type (398-402)
• _get_last_layer_size (404-408)

deepmd/pd/model/descriptor/repformers.py (4)

• get_sel (291-293)
• get_rcut (279-281)
• get_rcut_smth (283-285)
• get_ntypes (295-297)

🪛 Ruff (0.8.2)

deepmd/pd/utils/tabulate.py

54-54: Do not use mutable data structures for argument defaults

Replace with None; initialize within function

(B006)

---

55-55: Do not perform function call ActivationFn in argument defaults; instead, perform the call within the function, or read the default from a module-level singleton variable

(B008)

⏰ Context from checks skipped due to timeout of 90000ms (29)

• GitHub Check: Build wheels for cp310-manylinux_aarch64
• GitHub Check: Build wheels for cp311-win_amd64
• GitHub Check: Build wheels for cp311-macosx_arm64
• GitHub Check: Build wheels for cp311-macosx_x86_64
• GitHub Check: Build C++ (clang, clang)
• GitHub Check: Build wheels for cp311-manylinux_x86_64
• GitHub Check: Build C++ (rocm, rocm)
• GitHub Check: Test Python (6, 3.12)
• GitHub Check: Build wheels for cp311-manylinux_x86_64
• GitHub Check: Build C++ (cuda120, cuda)
• GitHub Check: Test Python (6, 3.9)
• GitHub Check: Build C++ (cuda, cuda)
• GitHub Check: Test Python (5, 3.12)
• GitHub Check: Build C++ (cpu, cpu)
• GitHub Check: Test Python (5, 3.9)
• GitHub Check: Test Python (4, 3.12)
• GitHub Check: Test C++ (false)
• GitHub Check: Test Python (4, 3.9)
• GitHub Check: Build C library (2.14, >=2.5.0rc0,<2.15, libdeepmd_c_cu11.tar.gz)
• GitHub Check: Test C++ (true)
• GitHub Check: Test Python (3, 3.12)
• GitHub Check: Build C library (2.18, libdeepmd_c.tar.gz)
• GitHub Check: Test Python (3, 3.9)
• GitHub Check: Test Python (2, 3.12)
• GitHub Check: Test Python (2, 3.9)
• GitHub Check: Test Python (1, 3.12)
• GitHub Check: Test Python (1, 3.9)
• GitHub Check: Analyze (python)
• GitHub Check: Analyze (c-cpp)

🔇 Additional comments (1)

deepmd/pd/utils/tabulate.py (1)

203-279:

❓ Verification inconclusive

Verify tensor dimension logic in the deeper embedding layers.

In the branches starting from line 203, you handle expansions for dual-size layers (2 * self.neuron[layer - 1]) and skip connections (zz + yy). Consider verifying that the shapes match expectations for all supported descriptor types, especially when certain types are excluded. You can run a test script or add shape assertions:

---

🏁 Script executed:

#!/bin/bash
# Perform shape checks on the generated tensors
# This script searches for unit tests or calls to _make_data and logs shapes

rg "_make_data" -A 10

Length of output: 4814

---

Action: Confirm Tensor Dimension Consistency in Deeper Layers

Please verify that the tensor shapes produced in these branches (lines 203–279 in deepmd/pd/utils/tabulate.py) conform to expectations for all supported descriptor types. In particular, double-check that:

• In the branch handling the dual-size layers (2 * self.neuron[layer - 1]), the intermediate tensors (e.g., tt, zz, and ybar) have compatible shapes.
• In the skip connection branch (where zz + yy is used), the addition does not inadvertently introduce shape mismatches.

It’s recommended to add explicit shape assertions or extend unit tests (similar to those using _make_data in other modules) to automatically verify that the outputs of paddle.matmul, bias additions, and derivative computations (via unaggregated_dy_dx and unaggregated_dy2_dx) are as expected.

[coderabbitai]

Actionable comments posted: 0

🧹 Nitpick comments (1)

deepmd/pd/model/descriptor/se_a.py (1)

261-283: Consider handling partial state if table creation fails and implementing forward pass under compression.

The new compression logic looks consistent overall. However, note that once self.compress is set to True, the forward path raises NotImplementedError later in the code (line 779). If compression is intended for production, consider handling the forward logic or returning a graceful error message. Additionally, you may wish to implement tests verifying that table creation succeeds and partial states are not left behind if an exception occurs before setting self.compress = True.

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between bf98888 and e441eb9.

📒 Files selected for processing (2)

• deepmd/pd/model/descriptor/se_a.py (3 hunks)
• deepmd/pd/model/descriptor/se_atten.py (1 hunks)

🧰 Additional context used

🧬 Code Graph Analysis (1)

deepmd/pd/model/descriptor/se_a.py (3)

deepmd/pd/utils/utils.py (1)

• ActivationFn (69-107)

deepmd/dpmodel/model/make_model.py (2)

• serialize (548-549)
• enable_compression (191-217)

deepmd/pd/model/atomic_model/dp_atomic_model.py (1)

• enable_compression (214-243)

⏰ Context from checks skipped due to timeout of 90000ms (30)

• GitHub Check: Test Python (6, 3.12)
• GitHub Check: Test Python (6, 3.9)
• GitHub Check: Test Python (5, 3.12)
• GitHub Check: Test Python (5, 3.9)
• GitHub Check: Test Python (4, 3.12)
• GitHub Check: Test Python (4, 3.9)
• GitHub Check: Build wheels for cp310-manylinux_aarch64
• GitHub Check: Test Python (3, 3.12)
• GitHub Check: Build C++ (clang, clang)
• GitHub Check: Test Python (3, 3.9)
• GitHub Check: Build wheels for cp311-win_amd64
• GitHub Check: Analyze (python)
• GitHub Check: Build wheels for cp311-macosx_arm64
• GitHub Check: Test Python (2, 3.12)
• GitHub Check: Build C++ (rocm, rocm)
• GitHub Check: Test Python (2, 3.9)
• GitHub Check: Build C++ (cuda120, cuda)
• GitHub Check: Build wheels for cp311-macosx_x86_64
• GitHub Check: Analyze (javascript-typescript)
• GitHub Check: Analyze (c-cpp)
• GitHub Check: Build wheels for cp311-manylinux_x86_64
• GitHub Check: Build C library (2.14, >=2.5.0rc0,<2.15, libdeepmd_c_cu11.tar.gz)
• GitHub Check: Build C++ (cuda, cuda)
• GitHub Check: Test Python (1, 3.12)
• GitHub Check: Build wheels for cp311-manylinux_x86_64
• GitHub Check: Test Python (1, 3.9)
• GitHub Check: Build C++ (cpu, cpu)
• GitHub Check: Build C library (2.18, libdeepmd_c.tar.gz)
• GitHub Check: Test C++ (false)
• GitHub Check: Test C++ (true)

🔇 Additional comments (4)

deepmd/pd/model/descriptor/se_atten.py (1)

431-434: Improved tensor assignment using paddle.assign

The change from direct assignment to using paddle.assign() for the compression data is a good practice. This approach is more explicit and follows PaddlePaddle's recommended patterns for tensor assignment operations.

deepmd/pd/model/descriptor/se_a.py (3)

63-65: Imports for tabulate functionality look good.

These imports correctly reference the new DPTabulate class. No issues found with the syntax or usage at this time.

---

66-68: Imports for activation function look good.

ActivationFn is properly imported and later used to instantiate activation layers. Everything appears correct.

---

703-704: Validate tensor shape consistency for table assignment.

The code assigns newly built table tensors into self.compress_data and self.compress_info. Confirm that the shapes and dtypes match all model expectations, especially if this method is called in different contexts or with different table sizes.

[coderabbitai]

Actionable comments posted: 0

♻️ Duplicate comments (2)

source/tests/pd/model/test_compressed_descriptor_dpa2.py (2)

31-49: Duplicate utility function should be refactored.

This eval_pd_descriptor function is identical to the one in test_compressed_descriptor_se_atten.py and similar to others in the test suite. Extract it to a common test utilities module to reduce code duplication.

---

97-108: Add a comment explaining the rcut difference between repinit and repformer.

The repformer is configured with a smaller cutoff radius (self.rcut - 1) than repinit. Please add an inline comment explaining the rationale behind this difference to improve code maintainability.

🧹 Nitpick comments (6)

source/tests/pd/model/test_compressed_descriptor_dpa2.py (2)

125-125: Consider parameterizing the compression ratio.

The compression ratio is hardcoded to 0.5. Consider parameterizing this value to test compression behavior at different ratios, which would provide more comprehensive test coverage for the compression feature.

-        self.descriptor.enable_compression(0.5)
+        # Test with different compression ratios
+        compression_ratios = [0.3, 0.5, 0.7]
+        for ratio in compression_ratios:
+            self.descriptor.enable_compression(ratio)
+            result_pd_compressed = eval_pd_descriptor(
+                self.descriptor,
+                self.natoms,
+                self.coords,
+                self.atype,
+                self.box,
+            )
+            
+            self.assertEqual(result_pd.shape, result_pd_compressed.shape)
+            paddle.testing.assert_close(
+                result_pd,
+                result_pd_compressed,
+                atol=self.atol,
+                rtol=self.atol,
+            )

---

52-52: Consider adding float32 to the test parameters.

The test class is parameterized with only float64, but there's code for handling float32 precision in the setUp method. Consider adding float32 to the test parameters to ensure the descriptor works correctly with both precision levels.

-@parameterized(("float64",), (True, False))
+@parameterized(("float64", "float32"), (True, False))

deepmd/pd/model/descriptor/repflow_layer.py (4)

92-93: Avoid repeated assignments to self.update_residual and self.update_residual_init.

Currently, lines 92–93 and 107–108 both assign the same values to self.update_residual and self.update_residual_init. This duplication can be removed or consolidated to improve clarity.

Below is a possible refactor:

         self.update_residual = update_residual
         self.update_residual_init = update_residual_init
-        ...
-        self.update_residual = update_residual
-        self.update_residual_init = update_residual_init

Also applies to: 107-108

---

309-309: Remove or justify the unused e_dim variable.

Static analysis indicates that e_dim is assigned at line 309 but never used. If not needed, remove it to reduce clutter. If it’s needed for debugging or logging, add a comment explaining its purpose.

- e_dim = edge_ebd.shape[-1]

🧰 Tools

🪛 Ruff (0.8.2)

309-309: Local variable e_dim is assigned to but never used

Remove assignment to unused variable e_dim

(F841)

---

519-519: Remove or justify the unused nall variable.

Although a similar pattern was intentionally used in another file based on prior learnings, this file lacks any documentation indicating the same need. If nall is not genuinely needed here, please remove it or explain its presence.

- nall = node_ebd_ext.shape[1]

🧰 Tools

🪛 Ruff (0.8.2)

519-519: Local variable nall is assigned to but never used

Remove assignment to unused variable nall

(F841)

---

777-777: Remove or justify the unused nitem variable.

The local variable nitem is assigned at line 777 but never used in this “res_residual” update method. Consider removing it unless intended for future use or debugging.

- nitem = len(update_list)

🧰 Tools

🪛 Ruff (0.8.2)

777-777: Local variable nitem is assigned to but never used

Remove assignment to unused variable nitem

(F841)

📜 Review details

Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro

📥 Commits

Reviewing files that changed from the base of the PR and between e441eb9 and 60ebbf1.

📒 Files selected for processing (3)

• deepmd/pd/model/descriptor/repflow_layer.py (1 hunks)
• deepmd/pd/utils/nlist.py (2 hunks)
• source/tests/pd/model/test_compressed_descriptor_dpa2.py (1 hunks)

🚧 Files skipped from review as they are similar to previous changes (1)

• deepmd/pd/utils/nlist.py

🧰 Additional context used

🧠 Learnings (1)

deepmd/pd/model/descriptor/repflow_layer.py (1)

Learnt from: njzjz
PR: deepmodeling/deepmd-kit#4226
File: deepmd/dpmodel/model/make_model.py:370-373
Timestamp: 2024-10-16T21:50:10.680Z
Learning: In `deepmd/dpmodel/model/make_model.py`, the variable `nall` assigned but not used is intentional and should not be flagged in future reviews.

🪛 Ruff (0.8.2)

deepmd/pd/model/descriptor/repflow_layer.py

309-309: Local variable e_dim is assigned to but never used

Remove assignment to unused variable e_dim

(F841)

---

519-519: Local variable nall is assigned to but never used

Remove assignment to unused variable nall

(F841)

---

777-777: Local variable nitem is assigned to but never used

Remove assignment to unused variable nitem

(F841)

⏰ Context from checks skipped due to timeout of 90000ms (30)

• GitHub Check: Test Python (6, 3.12)
• GitHub Check: Test Python (6, 3.9)
• GitHub Check: Test Python (5, 3.12)
• GitHub Check: Test Python (5, 3.9)
• GitHub Check: Test Python (4, 3.12)
• GitHub Check: Test Python (4, 3.9)
• GitHub Check: Test Python (3, 3.12)
• GitHub Check: Build wheels for cp310-manylinux_aarch64
• GitHub Check: Test Python (3, 3.9)
• GitHub Check: Build wheels for cp311-win_amd64
• GitHub Check: Build C++ (clang, clang)
• GitHub Check: Test Python (2, 3.12)
• GitHub Check: Build wheels for cp311-macosx_arm64
• GitHub Check: Build C++ (rocm, rocm)
• GitHub Check: Build wheels for cp311-macosx_x86_64
• GitHub Check: Test Python (2, 3.9)
• GitHub Check: Analyze (python)
• GitHub Check: Build C library (2.14, >=2.5.0rc0,<2.15, libdeepmd_c_cu11.tar.gz)
• GitHub Check: Build C++ (cuda120, cuda)
• GitHub Check: Build wheels for cp311-manylinux_x86_64
• GitHub Check: Test Python (1, 3.12)
• GitHub Check: Analyze (javascript-typescript)
• GitHub Check: Build C library (2.18, libdeepmd_c.tar.gz)
• GitHub Check: Build C++ (cuda, cuda)
• GitHub Check: Test C++ (false)
• GitHub Check: Test Python (1, 3.9)
• GitHub Check: Build wheels for cp311-manylinux_x86_64
• GitHub Check: Analyze (c-cpp)
• GitHub Check: Test C++ (true)
• GitHub Check: Build C++ (cpu, cpu)

[HydrogenSulfate]

@njzjz This PR is ready for review

[iProzd]

@HydrogenSulfate Note that this PR not only adds DPA3 descriptor, but also modifies other parts (such as hessian loss, silut activation and other improvements). Please at least keep a list of modifications in this PR comment. The main modifications are LGTM.

[HydrogenSulfate]

@HydrogenSulfate Note that this PR not only adds DPA3 descriptor, but also modifies other parts (such as hessian loss, silut activation and other improvements). Please at least keep a list of modifications in this PR comment. The main modifications are LGTM.

Thanks for comments, I have added the summary of modification.

[njzjz]

@coderabbitai resolve