bdbj
/

Llama-3.1-70B-qpal-df-msq-3.4bit

Model card Files Files and versions

Model Card

Base model: meta-llama/Llama-3.1-70B
Quantization method: Memory constrained MSQ with Q-Palette
Target bit-width: 3.4
Backend kernel: Q-Palette kernel
Calibration data: N/A (data-free)

How to run

Follow the instruction in https://github.com/snu-mllab/Q-Palette.

References

Model Paper

Downloads last month: 8

Safetensors

Model size

16B params

Tensor type

I32

·

F16

·

I16

·

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for bdbj/Llama-3.1-70B-qpal-df-msq-3.4bit

Base model

meta-llama/Llama-3.1-70B

Quantized

(84)

this model

Collection including bdbj/Llama-3.1-70B-qpal-df-msq-3.4bit

Data-free quantization w/ Q-Palette

30 items • Updated 5 days ago