bohrium-job

SKILL: Bohrium Job Management

Overview

Manage compute jobs on the Bohrium platform. Prefer bohr CLI; fall back to the API only for advanced operations not supported by the CLI.

Authentication

BOHR_ACCESS_KEY and PROJECT_ID are read from the OpenClaw config ~/.openclaw/openclaw.json:

"bohrium-job": {
  "enabled": true,
  "apiKey": "YOUR_BOHR_ACCESS_KEY",
  "env": {
    "BOHR_ACCESS_KEY": "YOUR_BOHR_ACCESS_KEY",
    "PROJECT_ID": "YOUR_PROJECT_ID"
  }
}

OpenClaw automatically injects env variables into the runtime. The skill only requires BOHR_ACCESS_KEY; helper scripts handle any legacy CLI compatibility internally.

Prerequisites: Install bohr CLI

# macOS
/bin/bash -c "$(curl -fsSL https://dp-public.oss-cn-beijing.aliyuncs.com/bohrctl/1.0.0/install_bohr_mac_curl.sh)"

# Linux
/bin/bash -c "$(curl -fsSL https://dp-public.oss-cn-beijing.aliyuncs.com/bohrctl/1.0.0/install_bohr_linux_curl.sh)"

# Verify
source ~/.bashrc  # or source ~/.zshrc
export PATH="$HOME/.bohrium:$PATH"
bohr version

The installer auto-configures OPENAPI_HOST and TIEFBLUE_HOST. If they don't take effect, set manually:

export OPENAPI_HOST=https://open.bohrium.com
export TIEFBLUE_HOST=https://tiefblue.dp.tech

---

Submit Jobs

Method 1: CLI Arguments (Recommended)

bohr job submit \
  -m "registry.dp.tech/dptech/deepmd-kit:3.1.1" \
  -t "c4_m15_1 * NVIDIA T4" \
  -c "python train.py" \
  -p ./input_dir/ \
  --project_id 154 \
  -n "my-job-name"

Parameters:

Parameter	Short	Required	Description
--image_address	-m	Yes	Full image URL (e.g. registry.dp.tech/dptech/xxx:tag)
--machine_type	-t	Yes	Machine spec
--command	-c	Yes	Command to execute
--input_directory	-p	No	Input file directory (default ./)
--project_id		Yes	Project ID
--job_name	-n	No	Job name
--log_file	-l	No	Log file path
--result_path	-r	No	Auto-download results path (only /data, /personal, /share)
--job_group_id	-g	No	Job group ID (create with bohr job_group create)
--max_run_time		No	Max run time (minutes); auto-terminates on timeout
--max_reschedule_times		No	Auto-retry count after abnormal interruption
--nnode		No	Parallel compute node count (default 1)

Method 2: Config File (For Complex Scenarios)

bohr job submit -i job.json -p ./input_dir/

Full job.json example:

{
  "job_name": "my-training-job",
  "command": "python train.py --epochs 10",
  "log_file": "train.log",
  "backward_files": ["model.pt", "results/"],
  "project_id": 154,
  "machine_type": "c4_m15_1 * NVIDIA T4",
  "image_address": "registry.dp.tech/dptech/deepmd-kit:3.1.1",
  "job_type": "container",
  "disk_size": 50,
  "dataset_path": ["/bohr/my-dataset/v1"],
  "result_path": "/personal",
  "max_reschedule_times": 2,
  "max_run_time": 120,
  "nnode": 1
}

job.json field reference:

Field	Description	Example
job_name	Job name	"DeePMD-kit test"
command	Command; use relative paths	"cd se_e2_a && dp train input.json"
log_file	Log viewable in real-time	"train.log"
backward_files	Files to download; empty = keep all	["model.pt", "results/"]
project_id	Project ID	154
machine_type	Machine spec	"c4_m15_1 * NVIDIA T4"
image_address	Full image URL (not just name)	"registry.dp.tech/dptech/deepmd-kit:2.1.5-cuda11.6"
job_type	Must be "container"	"container"
dataset_path	Mounted dataset paths	["/bohr/my-dataset/v1"]
result_path	Auto-collect results to data disk	"/personal"
max_run_time	Max run time (minutes)	120
max_reschedule_times	Retry count on interruption	2
disk_size	Disk size (GB)	50

Important Notes

Topic	Details
Working directory	Bohrium auto-cds into the extracted input dir; use relative paths
Do NOT cd /root/input	Actual path is /home/input_lbg-{userId}-{jobId}/, unpredictable
image_address format	Must be full URL registry.dp.tech/dptech/xxx:tag
machine_type format	CPU: c2_m4_cpu; GPU: c4_m15_1 * NVIDIA T4
WAF blocking	If command triggers Alibaba Cloud WAF (405), write to script, use bash run.sh
Large files in -p	No output? Check for large hidden files causing slow compression
Auto-retry	max_reschedule_times retries on interruption (full re-run)
job_type	Must be "container"; VM jobs deprecated since 2023

---

View Jobs

bohr job list -n 10              # Recent 10
bohr job list -n 5 --json        # JSON output
bohr job list -r                 # Running only
bohr job list -f                 # Failed only
bohr job list -i                 # Finished only
bohr job list -p                 # Pending only
bohr job list -j 15954383        # Jobs in a specific group

Job Details

bohr job describe -j 22153612 --json
bohr job describe -j 22153612 -l        # Full details

View/Download Logs

bohr job log -j 22153612                # View
bohr job log -j 22153612 -o ./logs/     # Download

Download Results

bohr job download -j 22153612 -o ./results/
bohr job_group download -j 15954383 -o ./results/

---

Manage Jobs

bohr job terminate 22153612             # Terminate (keep results -> completed)
bohr job kill 22153612                  # Force stop (discard results, keep record)
bohr job delete 22153612                # Delete (remove everything)
bohr job terminate 22153612 22153613    # Batch terminate
bohr job delete 22153612 22153613       # Batch delete

terminate vs kill vs delete:

Action	Result files	Record	Status
terminate	Kept	Kept	-> completed
kill	Discarded	Kept	-> failed
delete	Removed	Removed	Disappears

---

Job Group Management

bohr job_group list -n 10 --json
bohr job_group list -s 2026-01-01 -e 2026-03-14     # Date range
bohr job_group create -n "experiment-v1" -p 154
bohr job_group terminate 15954383
bohr job_group delete 15954383
bohr job_group download -j 15954383 -o ./results/

Note: job_group_id from bohr job_group create is for CLI only (bohr job submit -g <id>). It differs from the web UI job group ID.

---

Scientific Software Examples

DeePMD-kit Training

{
  "job_name": "DeePMD-kit test",
  "command": "cd se_e2_a && dp train input.json > tmp_log 2>&1 && dp freeze -o graph.pb",
  "log_file": "se_e2_a/tmp_log",
  "backward_files": ["se_e2_a/lcurve.out", "se_e2_a/graph.pb"],
  "project_id": 154,
  "machine_type": "c4_m15_1 * NVIDIA T4",
  "job_type": "container",
  "image_address": "registry.dp.tech/dptech/deepmd-kit:2.1.5-cuda11.6"
}

LAMMPS Molecular Dynamics

{
  "job_name": "lammps_tutorial",
  "command": "mpirun -n 32 lmp_mpi -i in.shear > log",
  "log_file": "log",
  "backward_files": [],
  "project_id": 154,
  "machine_type": "c32_m64_cpu",
  "job_type": "container",
  "image_address": "registry.dp.tech/dptech/lammps:29Sep2021"
}

ABACUS First-Principles

{
  "job_name": "ABACUS test",
  "command": "OMP_NUM_THREADS=1 mpirun -np 8 abacus > log",
  "log_file": "log",
  "backward_files": [],
  "project_id": 154,
  "machine_type": "c16_m32_cpu",
  "job_type": "container",
  "image_address": "registry.dp.tech/dptech/abacus:3.0.0"
}

GROMACS Molecular Simulation

{
  "job_name": "bohrium-gmx-example",
  "command": "bash rungmx.sh > log",
  "log_file": "log",
  "backward_files": [],
  "project_id": 154,
  "machine_type": "c16_m62_1 * NVIDIA T4",
  "job_type": "container",
  "image_address": "registry.dp.tech/dptech/gromacs:2022.2"
}

CP2K Quantum Chemistry

{
  "job_name": "CP2K_Si_opt",
  "command": "source /cp2k-7.1/tools/toolchain/install/setup && mpirun -n 16 --allow-run-as-root --oversubscribe cp2k.popt -i input.inp -o output.log",
  "log_file": "output.log",
  "backward_files": ["output.log"],
  "project_id": 154,
  "machine_type": "c16_m32_cpu",
  "job_type": "container",
  "image_address": "registry.dp.tech/dptech/cp2k:7.1"
}

---

Container / VM Image Mapping

Software	VM Image	Container Image
DeePMD-kit	LBG_DeePMD-kit_2.1.4_v1	registry.dp.tech/dptech/deepmd-kit:2.1.5-cuda11.6
DPGEN	LBG_DP-GEN_0.10.6_v3	registry.dp.tech/dptech/dpgen:0.10.6
LAMMPS	LBG_LAMMPS_stable_23Jun2022_v1	registry.dp.tech/dptech/lammps:29Sep2021
GROMACS	gromacs-dp:2020.2	registry.dp.tech/dptech/gromacs:2022.2
Quantum-Espresso	LBG_Quantum-Espresso_7.1	registry.dp.tech/dptech/quantum-espresso:7.1
CP2K	-	registry.dp.tech/dptech/cp2k:7.1
ABACUS	-	registry.dp.tech/dptech/abacus:3.0.0
Base (CPU)	LBG_Common_v1/v2	registry.dp.tech/dptech/ubuntu:20.04-py3.10
Base (GPU)	LBG_base_image_ubun20.04	registry.dp.tech/dptech/ubuntu:20.04-py3.10-cuda11.6
Intel OneAPI	LBG_oneapi_2021_v1	registry.dp.tech/dptech/ubuntu:20.04-py3.10-intel2022-cuda11.6

---

API Supplement (CLI Unsupported Operations)

import os, requests

AK = os.environ.get("BOHR_ACCESS_KEY", "")
BASE = "https://open.bohrium.com/openapi/v1"
HEADERS = {"Authorization": f"Bearer {AK}"}

# Filter by status (0=pending, 1=running, 2=finished, 3=scheduling, -1=failed)
r = requests.get(f"{BASE}/job/list", headers=HEADERS,
    params={"page": 1, "pageSize": 10, "status": 1})

# Job config (includes file access token)
r = requests.get(f"{BASE}/job/view/conf/{job_id}", headers=HEADERS)
# Returns: {state, baseDir, tempDir, token, host, expires}

# View snapshot
r = requests.get(f"{BASE}/job/{job_id}/snapshot", headers=HEADERS)

# Rename job
requests.post(f"{BASE}/job/{job_id}/modify",
    headers={**HEADERS, "Content-Type": "application/json"},
    json={"name": "new-name"})

# Rename job group
requests.post(f"{BASE}/job_group/{job_group_id}/modify",
    headers={**HEADERS, "Content-Type": "application/json"},
    json={"name": "new-group-name"})

---

Job Status Codes

status	Meaning	CLI Display
0	Pending	Pending
1	Running	Running
2	Finished	Finished
3	Scheduling	Scheduling
-1	Failed	Failed

Common Machine Types

machine_type	Description
c2_m4_cpu	2 cores, 4G RAM
c4_m8_cpu	4 cores, 8G RAM
c8_m32_cpu	8 cores, 32G RAM
c16_m32_cpu	16 cores, 32G RAM
c32_m64_cpu	32 cores, 64G RAM
c4_m15_1 * NVIDIA T4	4 cores, 15G + 1x T4
c16_m62_1 * NVIDIA T4	16 cores, 62G + 1x T4
c8_m32_1 * NVIDIA V100	8 cores, 32G + 1x V100
c32_m128_4 * NVIDIA V100	32 cores, 128G + 4x V100

Troubleshooting

Problem	Cause	Solution
cd /root/input: No such file	Absolute path in command	Use relative paths
unsupported protocol scheme ""	Missing env vars	Set OPENAPI_HOST and TIEFBLUE_HOST
(200, '/account/login', None)	Old pip lbg	Use Go CLI (~/.bohrium/bohr)
WAF 405	Shell keywords in command	Write to script, use bash run.sh
Permission error	Wrong user	Verify BOHR_ACCESS_KEY
jobId vs jobGroupId	Different concepts	CLI uses jobId for kill/terminate/delete
No output after submit	Large hidden files in -p dir	Check dir size
Variable performance	~30% algorithm variance	Normal
Long scheduling	Image caching or resource shortage	Wait or contact support
System disk full	Output to system disk	Write to working directory
Abnormal interruption	Machine reclaim (rare)	Set max_reschedule_times