Push details to hub does not work

[NathanHB]

Removing the all section of the result file has broken the create metadata card mechanism.

To reproduce:

 python  run_evals_accelerate.py --model_args "pretrained=HuggingFaceH4/zephyr-7b-beta" --tasks "lighteval|triviaqa|0|0"  --override_batch_size 1  --output_dir="tmp/"  --max_samples 1 --save_details --push_details_to_hub --results_org open-llm-leaderboard --push_results_to_hub

Also, having datasets version below 2.18.0 yields an error when fetching some files for the details dataset.

[clefourrier]

Good catch!