Skip to content

v0.001 The First Sprint Release, "Working Prototype"

Latest
Latest

Choose a tag to compare

View all tags
@myungjoo myungjoo released this 27 Jan 06:51
· 50 commits to main since this release
v0.001
71eabfa

The first prototype that "works"

  • server-1: an OpenPCC router server.

  • server-2: an OpenPCC compute server. w/ ollama + llama3.1 1B 8bit. for c6a.4xlarge

  • client: CLI fake-attestation supported

  • Note: this requires fake-attestation and TPM-simulator only. (not very secure. real attestation and real TPM not supported)

  • With almost everything hardcoded, this is to setup the first building block for hybrid AI research.

  • When you deploy this be careful:

    • fake-attestation (at build-pack) is required.
    • server-2 requires to "know" server-1 (router) address. this is injected in the deploy process. To do so, EIF should be created at deploy.yml, not in build-pack.yml
      • DON'T enable "Build Nitro Enclave EIF for server-2" at build-pack.yml
      • DON'T add "S3 URI to upload compute EIF" at build-pack.yml
      • DO add "optional compute_boot build tags" with include_fake_attestation at build-pack.yml
      • CPU count < 14 (update /etc/nitro-enclave/allocate.yaml to increase this) at deploy.tml
      • 10000 < Memory < 24000 (update /etc/nitro-enclave/allocate.yaml to increase this) at deploy.tml
      • CID should be kept 16 at deploy.tml
      • Image tag should be kept the same between deploy.yml and build-pack.yml
      • Compute instance should be "c6a.4xlarge" or better.

Full Changelog: https://github.com/nnstreamer/hybrid/commits/v0.001

Assets 2
Loading