More flexible PR experiment requests.

[DonggeLiu]

1. Pass arbitrary flags to run_all_experiments.py (Support passing arbitrary flags to run_all_experiments.py via request_pr_exp.py.  #245).
2. Customize more fields in [oss-fuzz-gen/ci/k8s/pr-exp.yaml](https://github.com/google/oss-fuzz-gen/blob/main/ci/k8s/pr-exp.yaml). E.g.:
   • Hardcode each model's supported locations, and dynamically switch between them based on model names.

[Snehil-Shah]

@DonggeLiu I would like to help with the first issue. Can I be assigned?

[DonggeLiu]

@DonggeLiu I would like to help with the first issue. Can I be assigned?

Sure, thanks!

[Snehil-Shah]

@DonggeLiu for some clarification, the goal with the second task is to dynamically set the VERTEX_AI_LOCATIONS variable in the yaml file based on the model, right? (which is currently set to all locations?).

[DonggeLiu]

@DonggeLiu for some clarification, the goal with the second task is to dynamically set the VERTEX_AI_LOCATIONS variable in the yaml file based on the model, right? (which is currently set to all locations?).

Yes.
You can probably hardcode the model-locations pairs in ci/request_pr_exp.py to begin with, then also allow users to overwrite locations.
For example, we can support both:

# Use the default location set of model_1
/gcbrun exp ... --llm model_1

# Use the specified locations (l1, l2, l3)
/gcbrun exp ... --llm model_1 --llm-locations l1,l2,l3

Note the current location flag is for GKE, not models.