PAL-1217 heatlhbench original

[pratacosmin]

this can be run like so
define in config the model_deployments.yaml
and run

 helm-run --run-entries "health_bench:model=jsl/medical-visual-lm-30b,jury_config_path=<PATH>/judges.yaml" --max-eval-instances 4 --suite my-suite --local-path ./config

helm-summarize --suite my-suite -o ./benchmark_output --schema-path  <PATH>/schema_medhelm.yaml

 helm-server --suite my-suite -o ./benchmark_output --port 8000

[chakravarthik27]

LGTM

[blidiselalin]

Looks good

[MiguelAFH]

Left a comment to create a separate metric for health_bench_professional. Everything else looks good

[MiguelAFH]

I would suggest to change the name of the metric to health_bench_professional_score. The annotators differ in that the health_bench uses only one prompt to evaluate and health_bench_professional uses one or two prompts, so it's not the same metric. For clarity, it would be best to have them as separate and include the new metric in the schema as well.

[MiguelAFH]

In order for health_bench and health_bench_professional to appear in the leaderboard, they need to be placed in one of the groups for the medhelm categories (Look at how we have benchmarks under patient_communication in the schema). Also, it seems health_bench_professional has not been added to the schema.

[MiguelAFH]

Did final edits on the schema. Looks good to me now