A 4-billion-parameter open-source model outperformed the proprietary Mythos 5 on the CharXiv chart-understanding benchmark
Finbarr Timbers says open-source models are rapidly catching up.
Users sarcastically dismiss alignment in open models like Mythos as just adding safety filters later, implying the claims lack substance.
Most Activity
@huggingface Find more details here https://paperswithcode.co/paper/2406.18521
Find the model here (Apache 2.0 licensed) https://huggingface.co/baidu/Qianfan-OCR
On http://paperswithcode.co, you can see Mythos 5 getting beaten by a 4B open-source model on CharXiv, a popular chart understanding benchmark
A tiny model that is freely available on @huggingface, that you can deploy anywhere!
@NielsRogge @huggingface is this from this competition? https://datamfm.github.io/ how well are they performing wrt mythos / other SOTA closed source models for the doc to .md conversions apart from chart conversions
@finbarrtimbers theyll add a safety filter in the next update and call it alignment
