DpkGen vs Alternative Tools: Which One Is Best?

To troubleshoot DP-GEN (Deep Potential Generator) errors quickly, you must first inspect the master tracking file record.dpgen to pinpoint exactly which stage of the concurrent learning iteration (make_train, run_model_devi, or run_fp) failed. Because DP-GEN orchestrates multiple tools, isolating whether the issue is a syntax error in your JSON configuration or a cluster submission failure is key to a fast resolution. 🛠️ Step 1: Check record.dpgen to Isolate the Stage

Open record.dpgen in your working directory. It outputs two numbers per line (e.g., 0 4): First number: The iteration index (starts at 0).

Second number: The active process stage ranging from 0 to 9. 0–2 = Training Stage (make_train, run_train, post_train)

3–5 = Exploration/MD Stage (make_model_devi, run_model_devi, post_model_devi) 6–8 = Labeling/DFT Stage (make_fp, run_fp, post_fp)

Navigate straight to the log files of the failing step to bypass irrelevant data. ❌ Common Errors and Quick Fixes 1. Configuration & JSON Formatting Errors

dargs.ArgumentKeyError: undefined key xxx is not allowed in strict mode

Cause: DP-GEN (v0.10.7+) enforces strict schema validation and rejects outdated parameters in param.json.

Fix: Remove or comment out the legacy keys mentioned in the error trace. Check the latest ⁠DP-GEN Examples on GitHub for valid formatting. OSError: [Error cannot find valid a data system] Cause: Incorrect array format or pathing for datasets.

Fix: Ensure init_data_sys is formatted as a 1D list of paths, while sys_configs must be formatted as a two-dimensional list. sys_configs_prefix formatting issues Cause: Strict trailing and leading slash requirements.

Fix: Format your prefix without a leading slash and with a trailing slash (e.g., “sys_configs_prefix”: “SystemConfigs/”). 2. Remote Job Submission & dpdispatcher Failures RuntimeError: job:xxxxxxx failed 3 times

Cause: DP-GEN cannot execute the task on the High Performance Cluster (HPC) via dpdispatcher.

Fix: Log directly into your remote cluster and check the deep-level task logs (e.g., train.log for DeePMD-kit, or standard slurm/PBS error logs) to find the explicit crash reason. RuntimeError: find too many unsuccessfully terminated jobs

Cause: The percentage of failed exploration or DFT tasks has exceeded your ratio_failure threshold.

Fix: Temporarily increase the ratio_failure value in param.json if minor unconverged jobs are acceptable, or check your input files for geometric overlaps crashing the simulation. FileNotFoundError: … /01.model_devi/graph.xxx.pb

Cause: The training step crashed silently, so no frozen graph model was generated for the exploration stage.

Fix: Re-examine your initial_data systems and the preceding train.log to fix the training data constraints. 3. Software Environment & Dependencies Command not found or Software unavailable

Cause: The cluster node did not load the right paths or deep learning packages.

Fix: Verify your environment commands in machine.json. Ensure your remote cluster configuration block activates the correct Conda environment (e.g., conda activate deepmd) before execution. ⏱️ Quick Triage Checklist Medium·Flavius Dinu The Most Common Kubernetes Errors and How to Fix Them