huggingface/lighteval

[FT] showing count in Markdown summary table

Ouverte

#804 ouverte le 9 juin 2025

 (5 commentaires) (1 réaction) (0 personne assignée)Python (514 forks)auto 404
featuregood first issue

Métriques du dépôt

Stars
 (2 496 étoiles)
Métriques de merge PR
 (Métriques PR en attente)

Description

Issue encountered

I am seeing this output from lighteval endpoint litellm with lighteval==0.10.0:

|          Task          |Version|    Metric     |Value|   |Stderr|
|------------------------|------:|---------------|----:|---|-----:|
|all                     |       |ether0_accuracy|    0|±  |     0|
|community:ether0:loose:0|      0|ether0_accuracy|    0|±  |     0|

At a glance, it's unclear if I have misconfigured LightevalTaskConfig.hf_filter, or if accuracy truly was 0%.

(Note: in actuality, there were 10 questions, all incorrect.)

Solution/Feature

Showing a count of questions run, to clarify the size of the evaluation.

Possible alternatives

None

Guide contributeur