Skip to content

Commit

Permalink
Minor, documentation
Browse files Browse the repository at this point in the history
  • Loading branch information
@Const-me
Const-me committed Dec 11, 2023
1 parent 070a986 commit f04fb95
Showing 1 changed file with 3 additions and 2 deletions.
5 changes: 3 additions & 2 deletions Mistral/MistralChat/Readme.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -34,8 +34,9 @@ On my laptop with integrated AMD GPU I'm getting about 2 tokens/second.
It seems the performance bottleneck is memory bandwidth.<br/>
VRAM in nVidia 1080 Ti delivers up to 484 GB/second, dual-channel DDR4 in the laptop up to 50 GB/second.

I have tested on another desktop PC with nVidia GeForce 1650 GPU, which only has 4GB VRAM.<br/>
The software works, but because the model doesn’t fit in 4GB, it only managed to generate 1 token/second.
I have tested on another desktop PC with nVidia GeForce 1650 GPU, which only has 4 GB VRAM.<br/>
The software works, but because the model doesn’t fit in 4GB, it only managed to generate 1 token/second.<br/>
Apparently, to run this program at optimal speed the GPU needs to have at least 6 GB VRAM.

## Converting Original Model

Expand Down

0 comments on commit f04fb95

Please sign in to comment.