dragonkue-BGE-m3-ko-mlx-affine-4bit

flaglow/dragonkue-BGE-m3-ko-mlx-affine-4bit

dragonkue-BGE-m3-ko-mlx-affine-4bit is a machine learning model from flaglow. Use it for the machine learning task on the model card, and read the license before you ship it in a product.

Converted to MLX format for Apple Silicon.

Downloads · 30 days

2% of all-time downloads

All-time downloads

167

Public

Parameters

89M

337 MB on disk

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors320 MB · 95%

Parameter types

How the weights are stored.

U3270.9M · 80%

Model card

From the Hugging Face model README

Conversion

Converted to MLX format for Apple Silicon.

python -m mlx_embeddings.convert --hf-path /Volumes/970EvoPlus/Users/mac_ex/github/mlx_converter/models/dragonkue--BGE-m3-ko --mlx-path /Volumes/970EvoPlus/Users/mac_ex/github/mlx_converter/models/dragonkue-BGE-m3-ko-mlx-affine-4bit -q --q-bits 4 --q-group-size 64

Parameter	Value
Converter	`mlx_embeddings.convert`
Source	`dragonkue/BGE-m3-ko`
Output	`flaglow/dragonkue-BGE-m3-ko-mlx-affine-4bit`
Quantized	`yes`
q-mode	`affine`
q-bits	`4`
q-group-size	`64`

Benchmark Results

Evaluated on AutoRAG Korean Embedding Benchmark (dataset: ocr_corpus_v3 / qa_v4). Metrics: NDCG · F1 · Recall · Precision · mAP · mRR at top-k retrieval. For these retrieval metrics, higher is better. Bold values mark the best score for each metric at the same Top-k.

Quantization	Model	Top-k	NDCG	F1	Recall	Precision	mAP	mRR
original	`BGE-m3-ko`	top_1	0.6404	0.6404	0.6404	0.6404	0.6404	0.6404
affine-4bit	`dragonkue-BGE-m3-ko-mlx-affine-4bit`	top_1	0.6404	0.6404	0.6404	0.6404	0.6404	0.6404
fp16	`dragonkue-BGE-m3-ko-mlx-fp16`	top_1	0.6754	0.6754	0.6754	0.6754	0.6754	0.6754
original	`BGE-m3-ko`	top_3	0.7409	0.3991	0.7982	0.2661	0.7208	0.7208
affine-4bit	`dragonkue-BGE-m3-ko-mlx-affine-4bit`	top_3	0.7619	0.4211	0.8421	0.2807	0.7339	0.7339
fp16	`dragonkue-BGE-m3-ko-mlx-fp16`	top_3	0.7564	0.4211	0.8421	0.2807	0.7266	0.7266
original	`BGE-m3-ko`	top_5	0.7811	0.2982	0.8947	0.1789	0.7428	0.7428
affine-4bit	`dragonkue-BGE-m3-ko-mlx-affine-4bit`	top_5	0.777	0.2924	0.8772	0.1754	0.7427	0.7427
fp16	`dragonkue-BGE-m3-ko-mlx-fp16`	top_5	0.7964	0.3012	0.9035	0.1807	0.7601	0.7601
original	`BGE-m3-ko`	top_10	0.7918	0.1707	0.9386	0.0939	0.7448	0.7448
affine-4bit	`dragonkue-BGE-m3-ko-mlx-affine-4bit`	top_10	0.792	0.1675	0.9211	0.0921	0.7493	0.7493
fp16	`dragonkue-BGE-m3-ko-mlx-fp16`	top_10	0.8009	0.1722	0.9474	0.0947	0.7539	0.7539

Type: xlm-roberta
Created: May 21, 2026
Updated: May 21, 2026