Hugging Face's logo

FluidInference
/
system-one-gemma-coreml

Core ML
conversion-toolkit
gated-model
Model card Files
xet
 Community
system-one-gemma-coreml / STATUS.md
alexwengg's picture
alexwengg
Publish pinned blocked Gemma conversion toolkit without weights
bea429c verified
preview code
|
Raw History Blame Contribute Delete
938 Bytes

System One Gemma status — 2026-09-22

  • blocked_gated_base: authenticated hf_hub_download("google/gemma-3-270m", "config.json", revision="9b0cfec892e2bc2afd938c98eabe4e4a7b1e0ca1") returned GatedRepoError HTTP 403.
  • The exact upstream trained adapter, score-head tensor, SHA-256 hashes, serving renderer, 256-token limit, and 2.35 temperature are pinned and source-audited.
  • The Core ML exporter and native-vs-Core ML parity runner are prepared but have not executed. There is no validated Core ML package, speed number, ANE placement, or Decision Index score.
  • Trained-weight redistribution needs an explicit rights determination. This public toolkit excludes the adapter and base weights.
  • Next: account owner accepts the Gemma access terms. Then fetch exact base, run conversion and parity, review Gemma and upstream weight terms before publishing any converted weights.