SKILL: Bohrium Dataset Management

Overview

Manage datasets on the Bohrium platform. Prefer bohr CLI; fall back to the API for version management, quota checks, etc.

bohr dataset create advantages over web upload: no size limit and resumable upload.

Datasets solve common pain points:

• Repeated file upload on every job submission -> mount datasets to avoid re-upload
• Large input files with slow upload -> datasets support resumable upload
• Need to share data with collaborators -> datasets support project-level sharing

Authentication

"bohrium-dataset": {
  "enabled": true,
  "apiKey": "YOUR_ACCESS_KEY",
  "env": { "ACCESS_KEY": "YOUR_ACCESS_KEY" }
}

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)"
source ~/.bashrc && export PATH="$HOME/.bohrium:$PATH"
export OPENAPI_HOST=https://open.bohrium.com

List Datasets

bohr dataset list                       # Default: recent 50
bohr dataset list -n 10 --json          # JSON, top 10
bohr dataset list -p 154                # Filter by project ID
bohr dataset list -t "my-dataset"       # Search by title

JSON fields: id, title, path (mount path like /bohr/my-dataset/v1), projectName, creatorName, updateTime, desc

Create Dataset (Upload Data)

bohr dataset create \
  -n "my-dataset" \
  -p "my-dataset" \
  -i 154 \
  -l "/path/to/local/data"

Resumable upload: If interrupted (network issues, etc.), re-run the same command and enter y to resume from breakpoint.

Using Datasets

Mount in Compute Jobs

Add dataset_path to job.json:

{
  "job_name": "DeePMD-kit test",
  "command": "cd se_e2_a && dp train input.json",
  "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",
  "dataset_path": ["/bohr/my-dataset/v1", "/bohr/another-dataset/v2"]
}

dataset_path and -p (input directory) can be used simultaneously.

Mount on Dev Nodes

Select datasets when creating a container node; access via path (e.g. /bohr/my-dataset/v1).

• Adds 2-4s boot delay (regardless of count)
• Use df -a | grep bohr to view mount points

Use in Notebooks

1. Expand side panel in Notebook editor -> Select existing datasets
2. Hover dataset name -> click copy to get path
3. Use in code: cd /bohr/testdataset-6xwt/v1/

Datasets must be added before connecting to the node. Adding afterward requires a node restart.

Version Management

Datasets support multi-version management. Files within a version are immutable once created.

Create New Version

Via Web UI: Dataset details -> "New Version" -> system imports latest version files -> add/remove files -> Create.

Via API:

requests.post(f"{BASE}/{dataset_id}/version", headers=HEADERS_JSON,
    json={"versionDesc": "v2 update"})

Preparation time depends on file size and count.

Delete Datasets

bohr dataset delete 138201              # Single
bohr dataset delete 138201 108601       # Batch

Deleted versions cannot be recovered.

Permission Model

"Usable" permission can be granted to other projects or users via editing.

API Supplement (CLI Unsupported)

import os, requests

AK = os.environ.get("ACCESS_KEY", "")
BASE = "https://open.bohrium.com/openapi/v1/ds"
HEADERS = {"accessKey": AK}
HEADERS_JSON = {**HEADERS, "Content-Type": "application/json"}

# Dataset details
r = requests.get(f"{BASE}/{dataset_id}", headers=HEADERS)

# Version list
r = requests.get(f"{BASE}/{dataset_id}/version", headers=HEADERS)
# Returns: [{version, totalCount, totalSize, downloadUri, datasetPath, ...}]

# Specific version
r = requests.get(f"{BASE}/{dataset_id}/version/{version_id}", headers=HEADERS)

# Create via API
r = requests.post(f"{BASE}/", headers=HEADERS_JSON, json={
    "title": "my-dataset", "projectId": 154,
    "identifier": "my-dataset",  # Required, unique ID
})
# Returns: {datasetId, tiefbluePath, requestId}
# Then upload files via tiefblue, then call commit

# Commit
requests.put(f"{BASE}/commit", headers=HEADERS_JSON,
    json={"datasetId": dataset_id})

# New version
requests.post(f"{BASE}/{dataset_id}/version", headers=HEADERS_JSON,
    json={"versionDesc": "v2 update"})

# Update info
requests.put(f"{BASE}/{dataset_id}", headers=HEADERS_JSON,
    json={"title": "new-title"})

# Delete version
requests.delete(f"{BASE}/{dataset_id}/version/{version_id}", headers=HEADERS)

# Check quota
r = requests.get(f"{BASE}/quota/check", headers=HEADERS,
    params={"projectId": 154})
# Returns: {result: true, limit: 30, used: 5}

# Upload token (for tiefblue)
r = requests.get(f"{BASE}/input/token", headers=HEADERS,
    params={"projectId": 154, "path": "/bohr/my-dataset"})

# Permissions
r = requests.get(f"{BASE}/{dataset_id}/permission", headers=HEADERS)

# Associated projects
r = requests.get(f"{BASE}/project", headers=HEADERS)

Important: The dataset list API path is GET /v1/ds/ (with trailing slash), not /v1/ds/list (/list gets caught by the /:id route).

Status Codes

Troubleshooting