After working with early GitLab CI customers we've shipped a new release of Actuated for GitLab - run your CI jobs securely in microVMs with very little management
#gitlabci#microvms#firecracker#supplychain
https://t.co/thjjPsUVxx
If you didn't watch it at the time, have a look at how
@AmpereComputing, @equinixmetal, and @selfactuated are helping @CloudNativeFdn projects do secure, native Arm64 builds on Ampere servers: https://t.co/IhSgajqj5C
I sent out some swag to customers and wanted to see what they were like for myself. This is the Gildan hoodie from @Printful
They're great value for one-off orders, including global shipping. https://t.co/H4XB5vMhxT
@swade1987@esotericmeans@laurentsenta@selfactuated Very likely, we've seen builds that took 6 hours before choking passing in 3-5 minutes, consistently.
Thanks for the mention.
https://t.co/1jwkXiIGEy
Using @selfactuated I ran the UNIX documentary from YouTube into @OpenAI's medium-sized Whisper model with 3 different GPUs.
Nvidia 3080: 2m14s
Nvidia 3060: 3m35
Nvidia A100 40GB: 4m5s
I guess $$$ isn't always better 🤔
Ran an SSH debug session with @selfactuated and asked llama2 why GPUs might be needed in CI, and what gaps we're leaving when we omit testing
The answer from 13b was much better than I've seen with self-hosted models in the past
What other models should we be testing?
Here's @ollama running e2e in GitHub Actions, with a dedicated GPU mounted into a microvM using @selfactuated
Example YAML below, and the responses from the model on an @nvidia 3060
https://t.co/aDSBKneJQl
I'll be in Paris from Tuesday evening for KubeCon, feel free to reach out if you'd like to meet - or to talk about microVMs, actuated, openfaas, inlets, etc. ☕🥖🇫🇷👋
An interesting pattern of Arm GitHub Actions jobs from @falco_org seen on the @selfactuated flightdeck this morning - is someone iterating on a PR, or is there a pending eBPF announcement for KubeCon?