MedHelm: Implement medcalc bench scenario, metrics and specs #3207
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
sashimono-san
changed the title
sashimono-san
force-pushed
the
feat/medcalc_bench_scenario
branch
from
98d3f83 to
792fb4f
Compare
sashimono-san
changed the title
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR has the adds support to MedCalc-Bench benchmarking.
As much as possible, I maintained the implementation as done originally. For example, data is downloaded directly from huggingFace, and that the evaluation metric is basically a copy of the original benchmark repo. But in some cases, changes were necessary, as discussed in the following paragraphs.
The original benchmark repo) expects the model to answer in JSON format. But considering we do not want to necessarily evaluate models' ability to output JSON, prompts were adapted so the output is given in natural language. It would be great if someone can review the prompts, make sure they align with other Helm scenarios.
Originally, one-shot examples come from a pre-defined JSON, and a specific example is used for each question type (defined by the field
Calculator Idin the dataset. I couldn't figure how to pass information, at runtime, from theScenarioto theRunSpec. Namely I did not find a way to extract from theInstancebeing used theCalculator Idand use it when building the inference prompt. Is this supported by the current implementation of Helm?The original implementation has a special truncation logic for the one-shot method. They truncates the Patient note and the step by step explanation of the output depending on the model. This is currently not implemented, but already during local tests (with gpt2) I had issues with input length. Any suggestions on how to solve this?