kobotoolbox / kpi / 23757408200 / 1

82%

master: 76%

push

github

feat(qual): get model id from settings DEV-1917 (#6877)

### Summary
Remove unnecessary information from audit logs and replace with bedrock
request id.


### 👷 Description for instance maintainers
The model id used for LLM requests is now configured via environment
variables.


### Notes
This is a combination of a few issues:
1. Model ids need to be configurable so we can use specific ARNs for
tracking purposes (DEV-1917)
2. LLM prompts and responses are no longer logged for privacy reasons. 
3. Project history logs should only show the request id and the model,
not the number of tokens. If we have the request id we can get the
number of tokens from bedrock logs, and it is not information we need
users to know (DEV-1936)

Model ids are still hardcoded to avoid having to make an extra AWS call
every time when the model id for a given ARN is unlikely to ever change.
This does open up the possibility that someone could change the ARN in
settings and the model id will no longer be correct, but that is highly
unlikely during the initial rollout. We are aware that we need to
rethink model configurations in general and make them more dynamic.

### 👀 Preview steps


1. ℹ️ Update your local environment to set `AUTOQA_OSS120_MODEL_AIP_ARN`
to a specific ARN (internal: ask Becca)
4. have an account and a project with an audio question and at least one
submission
5. Add a transcript
6. Add a QA question
7. "Generate with AI"
8. Look at the project history logs for the asset
9. 🔴 [on main] notice that the log contains information about the number
of tokens
10. 🟢 [on PR] the token information is no longer there, but there is a
bedrock request id. also note the model id has not changed

3570 of 11836 branches covered (30.16%)

18995 of 35948 relevant lines covered (52.84%)

0.53 hits per line