Modelpublic

yujuyeon/internvl3_5-1b-korean-347k

sourceHugging Faceotherupdated 3mo agoView on Hugging Face

1likes65downloads

Model Card

internvl3_5-1b-korean-347k

OpenGVLab/InternVL3_5-1B 를 한국어 멀티모달 데이터로 파인튜닝한 InternVL3.5 specialist.

항목	값
Base model	`OpenGVLab/InternVL3_5-1B`
Method	Full FT
Domain	한국어 종합 347k

Hyperparameters

—numtrainepochs: 5
—steps: 10000 / max 13575
—trainbatchsize: 2
—peak learning_rate: 3.999999772287927e-05

Training Loss

—init 1.5136 → final 0.3429 (min 0.3178)

step	loss
10	1.5136
1010	0.7029
2010	0.6294
3010	0.5353
4010	0.4988
5010	0.5090
6010	0.4190
7010	0.4065
8010	0.3964
9010	0.3448
10000	0.3429

Training Data

구성: 18개 서브셋 (한국어 specialist SFT)

subset	repeat
`aihub_visual_ShortQA_30k`	1
`hf_korLlava_Caption_20k`	1
`llava_ko_recap_30k`	1
`out_kor_llava_20k`	1
`chartRqa1_30k`	1
`chartRqa2_20k`	1
`tableVqa_Reason_20k`	1
`tableVqa_Caption_20k`	1
`aihub_subjectTxt_OCR_20k`	1
`aihub_visual_OCR_15k`	1
`kisti_arxiv_OCR_15k`	1
`kisti_hanbat_Reason_30k`	1
`kisti_documen_Reason_10k`	1
`aihub_mathMultiple_kor_M0`	1
`aihub_mathSubjective_kor_M0`	1
`kisti_hanbat_Vqa_25k`	1
`hf_latexUpdate_15k`	1
`aihub_subjectImg_Parse_10k`	1

Usage

python

from transformers import AutoModel, AutoTokenizer
import torch
m = AutoModel.from_pretrained("yujuyeon/internvl3_5-1b-korean-347k", torch_dtype=torch.bfloat16,
                              trust_remote_code=True).eval().cuda()
tok = AutoTokenizer.from_pretrained("yujuyeon/internvl3_5-1b-korean-347k", trust_remote_code=True, use_fast=False)