Kamiak HPC Cluster

---

✨ Key Features

🖥️ Cluster Management

• Secure Auth: SSH-based login with password or private key support.
• File Manager: deeply integrated file browsing, editing (code/syntax highlighting), and safe deletion.
• Web Terminal: Instant access to a remote shell for quick commands.

⚡ Job Orchestration (Slurm)

• Visual Job Composer: Create SBATCH scripts using templates (Standard/GPU) without memorizing flags.
• Live Monitoring: Real-time view of active queues (squeue) and historical jobs (sacct).
• Control: Cancel running jobs instantly with one click.

🤖 LLM Studio (Gems New!)

Turn your HPC allocation into a private AI playground.

• One-Click Server: Deploys a Flask-based LLM server on a GPU node automatically.
• Custom Models: Run Llama 3, Mistral, Gemma 7B, and more.
• RAG / Context: Upload documents (.pdf, .docx, .txt) to chat with your data.
• Markdown Chat: Rich text formatting for code blocks, tables, and lists.
• Manage Models: View and delete cached HuggingFace models to save disk space.
• Gems: Custom, specialized AI assistants designed to handle repetitive tasks or provide expert, tailored help on specific topics.

---

https://github.com/user-attachments/assets/aefeb9b5-d653-4b43-8fcb-be5b7f99b37b

---

🧠 LLM Integration Guide

The LLM Tab provides a complete interface for running local LLMs on the cluster.

1. Start Server

Select your desired model (e.g., meta-llama/Meta-Llama-3-8B) and click Start LLM Server.

• An SBATCH job is submitted to allocate a GPU node. Once running, an SSH tunnel is automatically established to your local machine.

2. Chat & RAG

• Chat: Interact naturally with the model.
• Upload Context: Click the + icon to upload file (PDFs or word files). The model will use this content to answer your questions.
• Clear Chat: Use the "Clear Chat / Reset Context" button to wipe history and forget uploaded files.

3. Manage Cache

Navigate to the Manage LLMs sub-tab to view models stored in $HOME/.cache/huggingface/hub. You can delete old models here to free up your user quota.

---

🛠️ Installation

Prerequisites

• Must have Kamiak Access: To request access visit https://hpc.wsu.edu/kamiak-hpc/requesting-access/
• Hugging Face Access Token: https://huggingface.co/docs/hub/en/security-tokens
• Node.js 18+

---

📦 Deployment (Standalone)

Windows Helper Script

For Windows users, a helper script run_standalone.bat is included. It automatically handles the copying of assets and starting the server.

Manually

For production deployment without node_modules dependency :

1. Build

   npm run build
   

2. Deploy Copy the .next/standalone folder to your server.

   • Note: Ensure you also copy .next/static -> .next/standalone/.next/static and public -> .next/standalone/public.

3. Run

   node server.js
   

---

Quick Start

1. Clone the Repo

   git clone <repository-url>
   cd <repository-directory>
   

2. Install Dependencies

   npm install
   

3. Run Development Server

   npm run dev
   

   Open http://localhost:3000 to launch.

---

💻 Technology Stack

• Frontend: Next.js 16 (App Router), React, Tailwind CSS
• Backend Ops: SSH2 (Node.js), Slurm Workload Manager
• AI Backend: Python, Flask, HuggingFace Transformers, PyTorch

---