Spaces:

HuggingFaceH4

/

open_llm_leaderboard

like 9.62k

Running on CPU Upgrade

@Dampfinchen I was wondering the same about Dolphin so I went to its model card and it says "trust_remote_code is required". I've heard evaluations fail if remote code is required.

Hi, the models failed during download, probably because it took too long to download, I re-added them to the queue.

Evaluation failed again.
https://huggingface.co/datasets/open-llm-leaderboard/requests/blob/main/cognitivecomputations/dolphin-2.6-mixtral-8x7b_eval_request_False_bfloat16_Original.json
https://huggingface.co/datasets/open-llm-leaderboard/requests/blob/main/cognitivecomputations/dolphin-2.5-mixtral-8x7b_eval_request_False_float16_Original.json
https://huggingface.co/datasets/open-llm-leaderboard/requests/blob/main/cognitivecomputations/dolphin-2.7-mixtral-8x7b_eval_request_False_bfloat16_Original.json

For some reason, these Dolphin Mixtrals have issues getting evaluated and when they are in the queue, it seems the whole process gets stuck :(

Hi!
These models all failed at download again. Could you make sure their weights are in safetensors?
(As requested by the submit form :) )

That's good to know. These are not my models but it seems they are not in the safetensors format. I guess that explains it then, thank you.

@Dampfinchen Perhaps try submitting Dolphin's GGUF versions. They should score virtually the same as their unquantized versions and safetensors shouldn't be an issue.

https://huggingface.co/TheBloke/dolphin-2.7-mixtral-8x7b-GGUF

Edit: Or GPTQ

https://huggingface.co/TheBloke/dolphin-2.7-mixtral-8x7b-GPTQ