bge-m3-doc-r1-gguf

keisuke-miyako/bge-m3-doc-r1-gguf

bge-m3-doc-r1-gguf is a machine learning model from keisuke-miyako. Use it for the machine learning task on the model card, and read the license before you ship it in a product.

Fine-tuned on 4D doc document embeddings. Dataset: keisuke-miyako/doc-2026-0607

Downloads · 30 days

13% of all-time downloads

All-time downloads

168

Public

Repo size

635 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.gguf635 MB · 100%

At a glance

Access: Public
Created: Jun 7, 2026
Updated: Jun 15, 2026
SHA: daab1449

Base models

BAAI/bge-m3

Model card

From the Hugging Face model README

bge-m3-doc-r1 (GGUF q8_0)

Fine-tuned on 4D doc document embeddings. Dataset: keisuke-miyako/doc-2026-0607

Usage (llama.cpp)

./llama-embedding -m bge-m3-doc-r1-q8_0.gguf --prompt "your query"

Bechmarks

Original BGE M3

Relevance	Min	Max	Average
`3`	`0.33`	`0.83`	`0.63`
`2`	`0.28`	`0.81`	`0.60`
`1`	`0.31`	`0.82`	`0.57`
`0`	`0.21`	`0.69`	`0.43`

avg. spread: 0.25
lv. 3 vs 2 is separated by 0.03
lv. 3 vs 1 is separated by 0.03
lv. 0 is separated at 0.43

Threshold	Positive	Negative	Gap
0.63	`0.60`	`0.11`	`0.49`
0.64	`0.52`	`0.08`	`0.44`
0.65	`0.44`	`0.06`	`0.38`
0.66	`0.37`	`0.05`	`0.32`

LoRA r1

Relevance	Min	Max	Average
`3`	`0.42`	`0.89`	`0.67`
`2`	`0.23`	`0.85`	`0.63`
`1`	`0.27`	`0.83`	`0.58`
`0`	`0.05`	`0.70`	`0.31`

avg. spread: 0.36 (+0.11) 👍🏻
lv. 3 vs 2 is separated by 0.04 (+0.01) 👍🏻
lv. 3 vs 1 is separated by 0.05 (+0.02) 👍🏻
lv. 0 is separated at 0.31 (-0.09) 👍🏻

Threshold	Positive	Negative	Gap
0.58	`0.93`	`0.32`	`0.61`
0.59	`0.92`	`0.28`	`0.63`
0.60	`0.87`	`0.26`	`0.61`
0.61	`0.86`	`0.21`	`0.64`

gap peak: 0.59

Created: Jun 7, 2026
Updated: Jun 15, 2026