FLUX Compute

Flux Compute lets students run model training remotely on cloud GPUs while continuing local development in VS Code.

Setup Guide

• Detailed setup and credential instructions: SETUP_AND_CREDENTIALS.md

What it does

• Handles login/authentication from inside VS Code.
• Adds a dedicated Jobs panel (tree view) in the Flux Compute activity bar.
• Adds a click-first Control Center with action buttons in the Flux Compute activity bar.
• Scans the workspace and suggests likely training scripts.
• Prompts for:
  • Training file to run
  • GPU type
  • Budget in USD
• Shows estimated cost before submit and warns if account balance is low.
• Shows a mandatory warning before upload not to send personal or regulated data.
• Packages project code using ignore rules similar to .gitignore, excluding common secrets and bulky/generated files by default.
• Uploads package and queues a remote training job.
• Streams remote logs into the Flux Compute output channel.
• Notifies on completion or failure.
• Shows account balance (command + status bar).
• Downloads completed job artifacts (model, logs, outputs) into the workspace.
• Lets users delete remote uploaded source, logs, and outputs at any time.

Commands

• Flux Compute: Submit Training Job
• Flux Compute: Download Training Artifacts
• Flux Compute: Login
• Flux Compute: Logout
• Flux Compute: Show Account Balance
• Flux Compute: Refresh Jobs
• Flux Compute: Open Job Details
• Flux Compute: Delete Remote Job Data

Tip:

• Students can use the Control Center view with inline form fields and buttons (no input pop-up prompts) for login, submit, refresh status, balance, and artifact download.

Settings

• cloudTrainer.apiBaseUrl: Base API URL (required)
• cloudTrainer.apiKey: Bearer token for API auth
• cloudTrainer.defaultGpuType: Default GPU shown in prompt
• cloudTrainer.defaultBudgetUsd: Default budget shown in prompt
• cloudTrainer.pollIntervalSeconds: Job polling interval
• cloudTrainer.excludePatterns: Extra packaging excludes

Default sensitive-file exclusions include:

• .env*, .aws, .ssh, private key files (*.pem, *.key, *.p12, *.pfx)
• credential-like file names (*secret*, *token*, *credentials*.json, *service-account*.json)
• local databases (*.db, *.sqlite, *.sqlite3)

Auth notes:

• Login stores a token in VS Code Secret Storage.
• You can also set cloudTrainer.apiKey manually for non-interactive auth.

Expected API Contract

Submit job

POST /v1/jobs

Request body:

{
  "metadata": {
    "trainingFile": "train.py",
    "gpuType": "NVIDIA T4",
    "budgetUsd": 20
  },
  "archiveBase64": "<base64-zip>"
}

Response body:

{
  "jobId": "job_123"
}

Login

POST /v1/auth/login

Request body:

{
  "email": "[email protected]",
  "password": "secret"
}

Response body:

{
  "apiKey": "token_abc123",
  "userName": "student"
}

Cost estimate

POST /v1/jobs/estimate

Request body:

{
  "trainingFile": "train.py",
  "gpuType": "NVIDIA T4",
  "budgetUsd": 20
}

Response body:

{
  "estimatedCostUsd": 12.75,
  "estimatedDurationHours": 3.5
}

Account

GET /v1/account

Response body:

{
  "balanceUsd": 45.2,
  "currency": "USD"
}

Job state

GET /v1/jobs/:jobId

Response body:

{
  "status": "running",
  "logs": ["epoch 1 ...", "epoch 2 ..."],
  "artifactUrl": "https://api.example.com/v1/jobs/job_123/artifact"
}

status can be: queued, running, completed, failed.

Delete job data

DELETE /v1/jobs/:jobId

Deletes the job's remote uploaded source bundle, logs, outputs, and artifact for the authenticated user.

Development

npm install
npm run compile

Press F5 in VS Code to launch an Extension Development Host and test commands.