The Rapid Evolution of GPT Models: Capabilities, Challenges, and Hybrid Solutions

GPT models, with their advanced conversational capabilities, face challenges in local deployment due to high hardware and memory requirements, with larger models needing over 40 GB of GPU memory. A hybrid cloud approach, combining local and cloud resources, offers an efficient solution by balancing workload management, cost, and scalability while adhering to data compliance standards.

Read more

In the future, the definition of "real" will have no meaning

One could argue that we’re currently close to a huge technological revolution that will happen within the next 10 to 15 years. One that has slowly been accelerated in quiet research pockets of academia and startups before being finally introduced to the masses. Machine learning technology is rapidly advancing, especially in natural language text processing and text generation. For example, the GPT-2 model developed by OpenAI can now generate text based on seed text, which is virtually indistinguishable to human generated text, and which is not only syntactically and grammatically correct, but also conveys coherent content.

  • gpt/2

  • machine learning

  • fake news generator

  • etc

These output mechanisms have been iteratively refined over many generations and the models have become much more reliable in producing output that is able to quantifiably extract meaning from source inputs.

But of course there can be inherent danger in unsupervised generative output, especially given social media platforms’ proclivity for promoting algorithmically generated misinformation and algorithmically promoted fake news.

Upgrading the VM lab

As I recently wrote in a previous post about the ESXi virtualization lab setup, we needed to upgrade the host hypervisor from the originally installed 6.5.0 to 6.7 U2. Following the instructions on TinkerTry: 

   https://tinkertry.com/easy-update-to-latest-esxi

This site outlines simple easy to follow steps for upgrading a vSphere ESXi installation to the latest. However it should be noted that the following command simply did not work for my installation:

   esxcli software profile update -p ESXi-6.7.0-20190402001-standard -d   
   https://hostupdate.vmware.com/software/VUM/PRODUCTION/main/vmw-depot-index.xml

This command simply resulted in the download hanging. Network access to the Internet was not an issue, as:

   esxcli software sources profile list --depot=https://hostupdate.vmware.com/
   software/VUM/PRODUCTION/main/vmw-depot-index.xml

Worked to display the available updates, just fine, as well as the command

   esxcli software profile update -p ESXi-6.7.0-20190402001-standard -d 
   https://hostupdate.vmware.com/software/VUM/PRODUCTION/main/vmw-depot-index.xml --dry-run

produced output which indicated that downloading the installation package itself was successful.

We were finally able to upgrade the lab by manually download the update package directly from VMware’s web site: 

   https://my.vmware.com/en/group/vmware/evalcenter?p=free-esxi6

Once downloaded and the file has been uploaded to the host datastore, you can install via the command: 

   esxcli software profile update -p <profile name> -d <location of ZIP file> 
   esxcli software profile update -p ESXi-6.7.0-20190402001-standard -d <path to update.zip>

The latest update includes mitigations for recent speculative execution side-channel attacks targeting Intel chips.

Virtualization Lab with ESXi hypervisor 6.5.x

This is a bit of a dated post, but I’m going to discuss virtualization lab configuration for development and experimentation using ESXi hypervisor 6.5.x. It’s particular relevant now as I’ve setup a Kubernetes cluster at home to develop some orchestration scripts and cluster architecture design for scalable applications. This is a much cheaper solution that deploying a cluster into a cloud-based provider. The only resource cost is power which comes in at a significantly cheaper expense than the per-hour provisioning costs of a cloud provider.

Read more