DeepSeek v3
New Deepseek v3 scores really high on Aider leaderboards:
https://aider.chat/docs/leaderboards/
Model: https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/tree/main
Would probably be the new best model on the platform.
Is 600B+ params a problem ? Would the pricing be similar to LLama 3 405B ?
Merry Christmas!
Log in to comment and vote
Comments6
Reem Ehab
May 2, 2025
This is a Project Report. I want you to make a powerpoint presentation of it:
Mikael Godal
Jan 1, 2025
Its on both deep infra and fireworks now, really impressive model!! Especiay the output token size which is basically unlimited
Mikael Godal
Dec 27, 2024
Chat model is here:
https://huggingface.co/deepseek-ai/DeepSeek-V3
Mikael Godal
Dec 27, 2024
It would make a dramatic difference for coding! But I think it’s too much for Nebius infrastructure to handle, it’s 700GB large!
Gaël CADO
Dec 27, 2024
It’s a MoE so yeah heavy to load the model, but only part of the parameters activate on inference.
So heavy on VRAM but not so much on compute
Gaël CADO
Dec 26, 2024
Note: It’s a MoE