- add adjustments for BPT (63b50fd46f481502c415d2887aebd462e8d29f7f)
- add adjustments for BPT (0ad4b4114e39bd13d6bebec9c9f7173a67be9fa9)
- add adjustments for BPT (06166c3c75538a943cefddb39cad56a1d062f3e5)
- document the image-to-text request body again (168531e73e2051cca32634c63ae7a4faca1d3319)
- make the CI gate this PR introduces pass on its own code (adccba448af62d7273263c9d3cf9705e523fc341)
- address review findings 2-6 (dd1c842ded80e8c8699f1f118c94195a51558816)
- spike: trusted-org gate for remote code, and one cached copy per model (a6e8fc2d224d4c1b94f0387c905ee81c272e293e)
- bound the model cache with LRU eviction (e8e94d2ad776cbe6c664461677827b3b54daaf00)
- bring the README in line with the service (15faa546ba1b9b1b894ef5a83480918b0255081f)
- add runnable request samples to the embedding endpoints (94b59c6176288dcf13a9f63e30a2447db8cf2303)
Co-authored-by: Lukas Mzigot <lukmzig@users.noreply.huggingface.co>
