Spaces:
Running
on
CPU Upgrade
Models with multiple submissions.
Some models are being submitted multiple times.
These are just the 7B ones I saw with 2 submissions or more.
ehartford/dolphin-2.1-mistral-7b
Open-Orca/Mistral-7B-SlimOrca
TheBloke/Llama-2-7B-GPTQ
TheTravellingEngineer/llama2-7b-chat-hf-v2
TheTravellingEngineer/llama2-7b-chat-hf-v3
TheTravellingEngineer/llama2-7b-chat-hf-v4
codellama/CodeLlama-7b-Instruct-hf
codellama/CodeLlama-7b-Python-hf
garage-bAInd/Platypus2-7B
kfkas/Llama-2-ko-7b-Chat
kittn/mistral-7B-v0.1-hf
lmsys/vicuna-7b-v1.5
lmsys/vicuna-7b-v1.5-16k
meta-llama/Llama-2-7b-hf
mosaicml/mpt-7b
mosaicml/mpt-7b-8k-chat
mosaicml/mpt-7b-8k-instruct
mosaicml/mpt-7b-storywriter
PocketDoc/Dans-TotSirocco-7b
tiiuae/falcon-7b-instruct
togethercomputer/LLaMA-2-7B-32K
togethercomputer/Llama-2-7B-32K-Instruct
wenge-research/yayi-7b-llama2
Hi!
Are these models the same at the precision and commit level?
The way we are dealing with it is by having filters on precision. If, however, the same model has two different categories (like the model you just showed), this is a mistake in the request file.
There's probably not going to be any useful difference between float16 and bfloat16 evals though.
Also filtering by precision doesn't exactly solve this since you don't really get to compare all models, since some may be submitted only as float16 or bfloat16 or 8bit, so either you filter it to one of those and then you can't see some models, or you don't filter and you see duplicates.
@xzuyn
We don't plan on changing this mechanism - we understand that it brings a bit of redundancy between the bfloat16 and float16 models, but since you can hide the quantized models from a given search, it should still allow people to compare models quite fast. Thank you for your interest in the leaderboard!
Closing as it is not an issue but a feature.