You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

ํ•œ๊ตญ ํŠนํ—ˆ-KOS ์†Œ๋ถ„๋ฅ˜ ๊ฒ€์ƒ‰ ์ž„๋ฒ ๋”ฉ ๋ชจ๋ธ

Embedding-Factory์—์„œ ํ•œ๊ตญ ํŠนํ—ˆ์™€ KOS ์†Œ๋ถ„๋ฅ˜ ์„ค๋ช…์˜ ์˜๋ฏธ์  ๊ฒ€์ƒ‰์„ ์œ„ํ•ด fine-tuningํ•œ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.

๋ชจ๋ธ ์ •๋ณด

ํ•ญ๋ชฉ ๋‚ด์šฉ
Base model Qwen/Qwen3-Embedding-0.6B
ํ•™์Šต ๋ฐ์ดํ„ฐ ํŠนํ—ˆ 2,002๊ฑด
ํ•™์Šต epoch 3
cutoff length 32,768
์ตœ์ข… ํ•™์Šต loss 0.2761
Loss Cached Multiple Negatives Ranking Loss
๊ฒ€์ƒ‰ ๋Œ€์ƒ KOS ์†Œ๋ถ„๋ฅ˜ 2,522๊ฐœ
positive ์ •์˜ ํŠนํ—ˆ์— ๋งคํ•‘๋œ ์†Œ๋ถ„๋ฅ˜์ฝ”๋“œ ์ „์ฒด

์ž…๋ ฅ ํ˜•์‹

ํŠนํ—ˆ anchor๋Š” ์ดˆ๋ก + ๊ธฐ์ˆ ๊ตฌ์„ฑ + ๋ฐฐ๊ฒฝ๊ธฐ์ˆ ์„ ์ด์–ด ๋ถ™์ธ ํ…์ŠคํŠธ์ž…๋‹ˆ๋‹ค.

query:

Instruct: Given a patent abstract, retrieve the relevant technology classification description.\nQuery: {patent_anchor}

taxonomy document:

{taxonomy_text}  # raw document; prefix ์—†์Œ

Nemotron์€ query์— query: , document์— passage: ๋ฅผ ์‚ฌ์šฉํ•ฉ๋‹ˆ๋‹ค. Qwen3๋Š” query instruction์„ ์ •ํ™•ํžˆ ํ•œ ๋ฒˆ๋งŒ ์‚ฌ์šฉํ•˜๋ฉฐ taxonomy document์—๋Š” prefix๋ฅผ ๋ถ™์ด์ง€ ์•Š์Šต๋‹ˆ๋‹ค.

์‚ฌ์šฉ ์˜ˆ์‹œ

from sentence_transformers import SentenceTransformer

model = SentenceTransformer("elplaguister/patent_classifier_en_qwen3_embedding_0_6b", trust_remote_code=True)
model.max_seq_length = 32768

patent_anchor = "์ดˆ๋ก ๋‚ด์šฉ ๊ธฐ์ˆ ๊ตฌ์„ฑ ๋‚ด์šฉ ๋ฐฐ๊ฒฝ๊ธฐ์ˆ  ๋‚ด์šฉ"
taxonomy_text = "๋Œ€๋ถ„๋ฅ˜๋ช…:...-์ค‘๋ถ„๋ฅ˜๋ช…:...-์†Œ๋ถ„๋ฅ˜๋ช…:...:์†Œ๋ถ„๋ฅ˜ ์„ค๋ช…..."

query = [f"Instruct: Given a patent abstract, retrieve the relevant technology classification description.\nQuery: {{patent_anchor}}"]
document = [taxonomy_text]

q_vec = model.encode(query, normalize_embeddings=True)
d_vec = model.encode(document, normalize_embeddings=True)
similarity = q_vec @ d_vec.T

ํ‰๊ฐ€ ์ •๋ณด

์ „์ฒด taxonomy๋ฅผ ๊ฒ€์ƒ‰ ํ’€๋กœ ์‚ฌ์šฉํ•˜๊ณ , ๊ฐ ํŠนํ—ˆ์˜ ๋งคํ•‘ ์†Œ๋ถ„๋ฅ˜ ์ „์ฒด๋ฅผ ์ •๋‹ต(positive)์œผ๋กœ ํ‰๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค. ํ‰๊ฐ€ ์ง€ํ‘œ๋Š” nDCG@10, MRR@10, Hit@1, Recall@10์ž…๋‹ˆ๋‹ค.

์—…๋กœ๋“œ ํŒŒ์ผ ๊ตฌ์„ฑ

์ด ํด๋”๋Š” inference์— ํ•„์š”ํ•œ ํŒŒ์ผ๋งŒ ํฌํ•จํ•ฉ๋‹ˆ๋‹ค. optimizer.pt, scheduler.pt, rng_state.pth, trainer_state.json, training_args.bin์€ ์ œ์™ธํ–ˆ์Šต๋‹ˆ๋‹ค.

Downloads last month
16
Safetensors
Model size
0.6B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for elplaguister/patent_classifier_qwen

Finetuned
(270)
this model

Collection including elplaguister/patent_classifier_qwen