Immers.cloud

💾 Need More Disk Space for Your Virtual Machine? Expanding It in immers.cloud Is Easier Than Ever!

Here’s how you can increase disk space for different types of VMs—Volume-backed and Local. Manage your resources quickly and efficiently.

Whether you’re working with a Volume-backed or Local VM, expanding disk space takes just a few clicks in immers.cloud. Follow our simple guide and eliminate storage limitations effortlessly!

🛠 Easy Connection to a Windows Server Virtual Machine: A Step-by-Step Guide

By default, we provide connection via Remote Desktop Protocol (RDP). If you need to connect to your Virtual Machine (VM) using alternative methods (e.g., SSH for Linux), you’ll need to configure them manually.

1. Connecting from Windows

Decrypt the password:

- In the "Actions" menu, select Get Password.

- On the page that opens, click Show to reveal the password.

- Download the RDP file: Click the IP address displayed on the VM’s page to download the file named vmname.rdp to your computer.

- Open the RDP file:

- Double-click the downloaded file to launch it.

- Log in using the following credentials:

Username: admin

Password: the one you decrypted earlier.

2. Connecting from macOS

- Install Microsoft Remote Desktop: Download it from the Mac App Store.

- Add the PC: In the application, click Add PC, enter the VM’s IP address in the "PC name" field, and click Add.

- Decrypt the password: In the "Actions" menu, select Get Password.

- On the page that opens, click Show to reveal the password.

- Connect to the VM: Use the saved configuration to access the VM.

- Log in using the following credentials:

Username: admin

Password: the one you decrypted earlier.

3. To Shut Down the VM

Use the Stop option in the "Actions" dropdown menu to properly shut down the virtual machine.

‼️ Important: Billing for the VM’s vCPU and RAM stops only if the VM is shut down using this method ‼️

👨‍💻 Qwen2.5 Coder: The Ultimate Open-Source Code Generation Model

The newly released large language model, Qwen2.5 Coder, sets the state of the art in code generation. This collection of models offers a range of features that make it highly appealing to developers.

Versatile Model Sizes: Qwen2.5 Coder is available in multiple configurations—0.5B, 3B, 14B, and 32B parameters. The largest version, which delivers benchmark-leading performance, rivals GPT-4 in code generation quality. Deploy it on immers.cloud to create a powerful private server for code analysis or a self-hosted alternative to GitHub Copilot.

Optimized for Local Use: For those looking for a compact coding assistant on local devices, smaller versions of the model are perfect and fully compatible with Ollama.

Enhanced Context and Versatility: With a 128k-token context window and support for 92 programming languages, Qwen2.5 Coder is designed to tackle a wide range of programming tasks. Developers worldwide are now exploring its potential.

Experience the power of Qwen2.5 Coder—whether for cloud deployments or as a localized assistant. The community is eager to see how it transforms coding workflows!

🔔 GPU Availability Notifications

We’re excited to tell you about a convenient feature: GPU Configuration Availability Notifications!

If your preferred Virtual Machine configuration is temporarily unavailable, you can set up an alert to stay updated on its status. With pay-as-you-go billing, GPU availability changes dynamically, so you’ll be notified immediately when the desired configuration becomes available.

To set up a notification:

1️⃣ Visit the GPU page using this link [https://en.immers.cloud/gpu/.

2️⃣ Click the bell icon (🔔) next to your desired configuration.

Once the equipment is available, you’ll receive an email notification, ensuring you can quickly access the resources you need.

🚀 Porting Llama 3.2 to Android and iOS Made Easy!

AI developers can now convert weights for Llama 3.2 models (1B and 3B) into a format compatible with mobile devices. The official open-source implementation of the new post-training quantization method, SpinQuant, optimizes these models for mobile platforms by reducing memory usage and enhancing inference speed.

This method is simpler than QLoRA because it doesn't require fine-tuning the model. All you need are the original uncompressed weights, a preliminary step to rotate weight and activation matrices (a script is provided in the repository), and a GPU. Any GPU on immers.cloud will work, and the RTX 3090 is more than enough for lightweight 1B and 3B models.

Once the rotation step is complete, you can perform standard post-training quantization using a provided script and then export the model for mobile deployment. But how do you actually run Llama on Android?

That's where ExecuTorch comes in—a framework designed to create PyTorch programs for mobile processors. The SpinQuant repository even provides scripts to export weights in a format compatible with ExecuTorch, so you’re ready to go!