Skip to content
Trending storyDeveloping

Kindling AI releases exlr8 K2/K3/K4 quantized GLM-5.3 for local multi-node serving

1 article1 sourcesince Oct 15Last article 1h ago ·

Overview

AISummary of 1 article

Kindling AI releases an exlr8 K2/K3/K4 quantization of the 744B-parameter GLM-5.3 mixture-of-experts model, according to a ModelScope post on X. The memory footprint ranges from 219GB to 378GB. The K2, K3 and K4 expert weights ship together, so precision can be changed at load time.

The release is a preview. ModelScope says performance measurement and full BF16-alignment validation are still in progress.

Written by AI from the articles below · updated Oct 15, 6:13 PM ET

Check the sources:

Latest developmentOct 15, 6:00 PM ET
Kindling AI releases compressed GLM-5.3 for multi-node local serving

Article timeline

The articles in this story. Times are ET.

Oct 15
  1. ModelScopeOfficial
    Kindling AI releases compressed GLM-5.3 for multi-node local serving

    AIKindling AI releases an exlr8 K2/K3/K4 quantization of the 744B-parameter GLM-5.3 MoE model. Memory profiles range from 219GB to 378GB, and K2, K3, and K4 expert weights ship together so precision can be changed at load time. The release is a preview, and performance and full BF16-alignment validation are still in progress.

    Image from @ModelScope2022's post

Heat trend

Not enough continuous observations to show a trend yet.