Prerequisites

• I am running the latest code. Mention the version if possible as well.
• I carefully followed the README.md.
• I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
• I reviewed the Discussions, and have a new and useful enhancement to share.

Feature Description

Update the Readme with a note on determining the best tensors to place on a GPU for a MoE using llama-gguf tool

Motivation

./build/bin/llama-gguf /path/to/model.gguf r n
(r: read, n: no check of tensor data)

It can be combined with a awk/sort one-liner to see tensors sorted by size decreasing, then by name:

./build/bin/llama-gguf /path/to/model.gguf r n
| awk '/read_0.+size =/ { gsub(/[=,]+/, "", $0); print $6, $4 }'
| sort -k1,1rn -k2,2
| less

I see testing emerging for GPU poor folks running large MoEs on modest hardware that placing the biggest tensor layers on GPU 0 via --override-tensor flag is best practice for speed.

Possible Implementation

No response