Problems with mteb_meta for german evaluation

[achibb]

Hi everyone, I am having troubles with generating the mteb_meta for German, with just running the script.

I am currently trying to format results but it does not seem to work straight away with "mteb_meta.py" - any idea? I just get a blank metadata file:

tags:

mteb
model-index:
name: gbert-large

results:

it gives this for every dataset:

WARNING:mteb.evaluation.MTEB:Passing task names as strings is deprecated and will be removed in the next release. Please use tasks = mteb.get_tasks(tasks=[...]) method to get tasks instead.
INFO:main:Skipping AmazonCounterfactualClassification as split test not present.
WARNING:mteb.evaluation.MTEB:Passing task names as strings is deprecated and will be removed in the next release. Please use tasks = mteb.get_tasks(tasks=[...]) method to get tasks instead.
INFO:main:Skipping AmazonReviewsClassification as split test not present.

Do I need to modify something on the code?

[imenelydiaker]

Due to recent updates #826 and #806:

For the WARNING message, if you're using a python script then your code should look like this:

import mteb
from sentence_transformers import SentenceTransformer

# Define the sentence-transformers model name
model_name = "average_word_embeddings_komninos"
# or directly from huggingface:
# model_name = "sentence-transformers/all-MiniLM-L6-v2"

model = SentenceTransformer(model_name)
tasks = mteb.get_tasks(tasks=["Banking77Classification"])
evaluation = mteb.MTEB(tasks=tasks)
results = evaluation.run(model, output_folder=f"results/{model_name}")

[imenelydiaker]

Else for these 2 messages:

INFO:main:Skipping AmazonCounterfactualClassification as split test not present.
INFO:main:Skipping AmazonReviewsClassification as split test not present.

It's likely a bug on our side, we'll check, thank you for reporting!

[KennethEnevoldsen]

@imenelydiaker I believe it is due to the new results format introduced in #759. mteb_meta.py will need to be rewritten for the new format.

We should probably make it a CLI with a test (otherwise it is impossible to know if it breaks).

[KennethEnevoldsen]

@achibb, we have currently updated the CLI and well as the benchmark lists. I believe the new CLI should suit your purpose

[achibb]

Thank you very much! Will test the next days and feedback. I was wondering can I also compute something for the German benchmark for other models like mdeberta, and somehow add it to the leaderboard ? Gesendet von Outlook für iOS<https://aka.ms/o0ukef>

…

________________________________ Von: Kenneth Enevoldsen ***@***.***> Gesendet: Tuesday, June 11, 2024 1:33:36 PM An: embeddings-benchmark/mteb ***@***.***> Cc: achibb ***@***.***>; Mention ***@***.***> Betreff: Re: [embeddings-benchmark/mteb] Problems with mteb_meta for german evaluation (Issue #847) @achibb<https://github.com/achibb>, we have currently updated the CLI and well as the benchmark lists. I believe the new CLI should suit your purpose — Reply to this email directly, view it on GitHub<#847 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AKBF2KRCQFGA6NY7WYMTW73ZG3OBBAVCNFSM6AAAAABIPBGHUWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNRQGUZDKMRRGA>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[imenelydiaker]

Thank you very much! Will test the next days and feedback. I was wondering can I also compute something for the German benchmark for other models like mdeberta, and somehow add it to the leaderboard ?

Yes you can evaluate any model and submit results to this repo via PR so they can be added to the leaderboard (check the guide on opening a PR on HF here).