Official AQLM quantization of meta-llama/Llama-3.2-3B finetuned with PV-Tuning.

For this quantization, we used 2 codebooks of 8 bits and groupsize of 8.

Results:

Model	Quantization	MMLU (5-shot)	ArcC	ArcE	Hellaswag	PiQA	Winogrande	Model size, Gb
meta-llama/Llama-3.2-3B	fp16	0.5642	0.4224	0.7449	0.5526	0.7666	0.6993	6.4
	2x8g8	0.3876	0.3328	0.6763	0.4817	0.7421	0.5975	1.5

Downloads last month: 14

Safetensors

Model size

1.1B params

Tensor type

FP16

Inference Examples

Text Generation

This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for ISTA-DASLab/Llama-3.2-3B-AQLM-PV-2Bit-2x8

Base model

meta-llama/Llama-3.2-3B

Quantized

(30)

this model

Collection including ISTA-DASLab/Llama-3.2-3B-AQLM-PV-2Bit-2x8

AQLM+PV

Collection

Official AQLM quantizations for "PV-Tuning: Beyond Straight-Through Estimation for Extreme LLM Compression": https://arxiv.org/abs/2405.14852 • 25 items • Updated 17 days ago • 19