Curious about the training cost behind MinerU2.5-Pro
#4
by wwjiang - opened
Hi! Thanks for the great work on MinerU2.5-Pro — the results are really impressive, especially considering it's still a 1.2B model.
I’m curious about the training setup behind this model. Would you be able to share some high-level insights about the training cost or compute scale (e.g., approximate GPU hours, number of GPUs, or training duration)?
Totally understand if exact numbers aren’t convenient to disclose — even rough estimates or comparisons would be very helpful for the community.
Thanks again for sharing this amazing work!