-
Notifications
You must be signed in to change notification settings - Fork 20
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
set up Guardrails API to check LLM inputs/outputs #70
Comments
Change https://go.dev/cl/637977 mentions this issue: |
Change https://go.dev/cl/637976 mentions this issue: |
Change https://go.dev/cl/637978 mentions this issue: |
Change https://go.dev/cl/637975 mentions this issue: |
Change https://go.dev/cl/637979 mentions this issue: |
A PolicyChecker is used to check the inputs and outputs of an LLM against safety policies. We will implement this interface with the GCP Checks Guardrails API. For #70 Change-Id: I52a776cc94900cef4c0b56f284c56e44f5136d4b Reviewed-on: https://go-review.googlesource.com/c/oscar/+/637975 Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Use the GCP Checks Guardrails API to implement llm.PolicyChecker. Don't use it anywhere yet. For #70 Change-Id: I788eb138ed1f887e53ab154a8b569abe70cadf34 Reviewed-on: https://go-review.googlesource.com/c/oscar/+/637976 LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
Add an optional policy checker to the overviews client. When a policy checker is configured, all LLM inputs and outputs will be checked for safety against the configured policy. Not yet used by Gaby or anywhere else. For #70 Change-Id: I8d48048eae9651499ec937a8804ab554baca2316 Reviewed-on: https://go-review.googlesource.com/c/oscar/+/637977 LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
If the -enforcepolicy flag is set (default false), check all safety categories when generating overviews. Since overviews are not yet published anywhere, this would only affect users of the web UI. For #70 Change-Id: I100070b5726ca0ff21cea4dec9f7f68e74018f08 Reviewed-on: https://go-review.googlesource.com/c/oscar/+/637978 Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com> LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
When policies are enforced (-enforcepolicy), display policy evaluation results on the Overview web UI page. To support this, return the detailed results from the llmapp.Overview functions. For #70 Change-Id: I5642236f48d205be25f7359b540a72ff86903e4f Reviewed-on: https://go-review.googlesource.com/c/oscar/+/637979 LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com> Reviewed-by: Hyang-Ah Hana Kim <hyangah@gmail.com>
Add a package that calls the Guardrails API (https://developers.google.com/checks/guide/ai-safety/guardrails), which will be used to check LLM inputs and outputs against policies.
The text was updated successfully, but these errors were encountered: