deepmd-train

Name: deepmd-train
Author: jinzhezenggroup

113stars23forks

How to install deepmd-train?

jinzhezenggroup/computational-chemistry-agent-skills/deepmd-train

$npx -y skills add jinzhezenggroup/computational-chemistry-agent-skills --skill deepmd-train

Installs into the current project.

Summary

Train DeePMD-kit models with progressive disclosure. Use when the user wants to train a DeePMD-kit potential, prepare an input.json, choose between model families such as se_e2_a/DeepPot-SE and DPA3, run dp train, monitor learning curves, freeze checkpoints, or test trained mod

Files · 1

View on GitHub

SKILL.md

1	# DeePMD-kit Training
2
3	Use this skill to guide DeePMD-kit model training without loading every model-specific recipe up front.
4	The workflow is intentionally progressive:
5
6	1. Understand the user's data, target accuracy, compute budget, and deployment backend.
7	1. Choose an appropriate model family.
8	1. Read only the reference file for the selected model under [`models/`](models/).
9	1. Generate or edit `input.json`, run training, monitor, freeze, and test.
10
11	## Progressive disclosure protocol
12
13	Do not start by reading every model document. First classify the request:
14
15	- If the user already named a model, read only that model reference.
16	- If the user asks for a recommendation, collect the decision inputs below, choose a model, then read only the selected reference.
17	- If model-specific parameters are not needed yet, stay in this top-level workflow.
18
19	Available model references:
20
21	\| Model reference \| Read when \|
22	\| ---------------------------------------- \| ---------------------------------------------------------------------------------------------------------------------------------------- \|
23	\| [`models/se-e2-a.md`](models/se-e2-a.md) \| The user wants a classical DeepPot-SE baseline, broad compatibility, or a smaller/established production model. \|
24	\| [`models/dpa3.md`](models/dpa3.md) \| The user wants a high-accuracy DPA3/LAM workflow, large/diverse datasets, dynamic neighbor selection, or pretrained DPA3-style training. \|
25
26	## Model selection
27
28	Ask only for missing information that changes the choice. Prefer reasonable defaults when the answer is obvious from context.
29
30	Key inputs:
31
32	- Data format and size: deepmd/npy, deepmd/hdf5, mixed type, number of systems/frames/elements.
33	- Target: quick baseline, production accuracy, large atomic model, transfer/fine-tuning, or deployment in MD.
34	- Compute: CPU/GPU, available memory, single-node vs. distributed training.
35	- Backend/deployment: PyTorch/TensorFlow/JAX/Paddle training; LAMMPS, Python inference, or other downstream use.
36	- Labels: energy/force only or also virial/stress.
37	- System diversity: single chemistry/phase vs. diverse multi-domain datasets.
38
39	Recommended defaults:
40
41	- Choose se_e2_a for a robust baseline, small to medium systems, compatibility-focused workflows, or when compute is limited.
42	- Choose DPA3 for high accuracy on diverse datasets, LAM-style training, or when the user explicitly asks for DPA3, DPA-3, LiGS, dynamic neighbor selection, or pretrained DPA3 variants.
43
44	## Common workflow
45
46	### 1. Confirm environment
47
48	```bash
49	dp --version
50	```
51
52	For PyTorch training, use `dp --pt ...`; for TensorFlow, use `dp ...`; for other backends, confirm the installed backend first.
53
54	### 2. Confirm training data
55
56	Training data should be in DeePMD format, typically deepmd/npy or deepmd/hdf5. If the user has raw electronic-structure outputs, convert them first with dpdata before writing the training input.
57
58	Minimum information needed to build `input.json`:
59
60	- `type_map`
61	- training system paths
62	- validation system paths
63	- whether virial labels are present and should be trained
64	- target number of steps or accuracy/time budget
65	- model choice
66
67	### 3. Read the selected model reference
68
69	After selecting a model, read the corresponding file under [`models/`](models/) and apply its model-specific configuration, hyperparameters, and caveats.
70
71	### 4. Train
72
73	```bash
74	dp --pt train input.json
75	```
76
77	Use the backend-specific command if not using PyTorch.
78
79	Restart from a checkpoint when needed:
80
81	```bash
82	dp --pt train input.json --restart model.ckpt.pt
83	```
84
85	### 5. Monitor
86
87	Training progress is usually written to `lcurve.out`. Check for:
88
89	- decreasing validation RMSE
90	- NaN or exploding losses
91	- train/validation divergence
92	- learning-rate schedule behaving as expected
93
94	### 6. Freeze and test
95
96	```bash
97	dp --pt freeze -o model.pth
98	dp --pt test -m model.pth -s /path/to/test_system -n 30
99	```
100
101	Adjust the backend flags and output extension for non-PyTorch models.
102
103	## Agent checklist
104
105	- [ ] Model was selected before reading model-specific details.
106	- [ ] Only the selected model reference was

People who liked this also viewed

6 picks

Type

deepmd-train

113stars23forks

View on GitHub

How to install deepmd-train?

jinzhezenggroup/computational-chemistry-agent-skills/deepmd-train

$npx -y skills add jinzhezenggroup/computational-chemistry-agent-skills --skill deepmd-train

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# DeePMD-kit Training
2
3	Use this skill to guide DeePMD-kit model training without loading every model-specific recipe up front.
4	The workflow is intentionally progressive:
5
6	1. Understand the user's data, target accuracy, compute budget, and deployment backend.
7	1. Choose an appropriate model family.
8	1. Read only the reference file for the selected model under [`models/`](models/).
9	1. Generate or edit `input.json`, run training, monitor, freeze, and test.
10
11	## Progressive disclosure protocol
12
13	Do not start by reading every model document. First classify the request:
14
15	- If the user already named a model, read only that model reference.
16	- If the user asks for a recommendation, collect the decision inputs below, choose a model, then read only the selected reference.
17	- If model-specific parameters are not needed yet, stay in this top-level workflow.
18
19	Available model references:
20
21	\| Model reference \| Read when \|
22	\| ---------------------------------------- \| ---------------------------------------------------------------------------------------------------------------------------------------- \|
23	\| [`models/se-e2-a.md`](models/se-e2-a.md) \| The user wants a classical DeepPot-SE baseline, broad compatibility, or a smaller/established production model. \|
24	\| [`models/dpa3.md`](models/dpa3.md) \| The user wants a high-accuracy DPA3/LAM workflow, large/diverse datasets, dynamic neighbor selection, or pretrained DPA3-style training. \|
25
26	## Model selection
27
28	Ask only for missing information that changes the choice. Prefer reasonable defaults when the answer is obvious from context.
29
30	Key inputs:
31
32	- Data format and size: deepmd/npy, deepmd/hdf5, mixed type, number of systems/frames/elements.
33	- Target: quick baseline, production accuracy, large atomic model, transfer/fine-tuning, or deployment in MD.
34	- Compute: CPU/GPU, available memory, single-node vs. distributed training.
35	- Backend/deployment: PyTorch/TensorFlow/JAX/Paddle training; LAMMPS, Python inference, or other downstream use.
36	- Labels: energy/force only or also virial/stress.
37	- System diversity: single chemistry/phase vs. diverse multi-domain datasets.
38
39	Recommended defaults:
40
41	- Choose se_e2_a for a robust baseline, small to medium systems, compatibility-focused workflows, or when compute is limited.
42	- Choose DPA3 for high accuracy on diverse datasets, LAM-style training, or when the user explicitly asks for DPA3, DPA-3, LiGS, dynamic neighbor selection, or pretrained DPA3 variants.
43
44	## Common workflow
45
46	### 1. Confirm environment
47
48	```bash
49	dp --version
50	```
51
52	For PyTorch training, use `dp --pt ...`; for TensorFlow, use `dp ...`; for other backends, confirm the installed backend first.
53
54	### 2. Confirm training data
55
56	Training data should be in DeePMD format, typically deepmd/npy or deepmd/hdf5. If the user has raw electronic-structure outputs, convert them first with dpdata before writing the training input.
57
58	Minimum information needed to build `input.json`:
59
60	- `type_map`
61	- training system paths
62	- validation system paths
63	- whether virial labels are present and should be trained
64	- target number of steps or accuracy/time budget
65	- model choice
66
67	### 3. Read the selected model reference
68
69	After selecting a model, read the corresponding file under [`models/`](models/) and apply its model-specific configuration, hyperparameters, and caveats.
70
71	### 4. Train
72
73	```bash
74	dp --pt train input.json
75	```
76
77	Use the backend-specific command if not using PyTorch.
78
79	Restart from a checkpoint when needed:
80
81	```bash
82	dp --pt train input.json --restart model.ckpt.pt
83	```
84
85	### 5. Monitor
86
87	Training progress is usually written to `lcurve.out`. Check for:
88
89	- decreasing validation RMSE
90	- NaN or exploding losses
91	- train/validation divergence
92	- learning-rate schedule behaving as expected
93
94	### 6. Freeze and test
95
96	```bash
97	dp --pt freeze -o model.pth
98	dp --pt test -m model.pth -s /path/to/test_system -n 30
99	```
100
101	Adjust the backend flags and output extension for non-PyTorch models.
102
103	## Agent checklist
104
105	- [ ] Model was selected before reading model-specific details.
106	- [ ] Only the selected model reference was

People who liked this also viewed

6 picks

Type