Japanese CLIP model clip‑japanese‑base‑v2, upgraded with ~2 B image‑text pairs and distillation, pairs an Eva02‑B image encoder with a 12‑layer BERT text encoder to reach higher ImageNet‑1k accuracy (0.708) than its predecessor.
Description in other languages
Statistics
Statistics as of Nov 16, 2026
Hugging Face downloads (last 30 days)
2,000
Hugging Face likes
18
Listed
Jan 6, 2026
Downloads (30-day window) over time
Show as a table
| Month | DL (30d) |
|---|---|
| Dec 2025 | — |
| Jan 2026 | 2,000 |
| Feb 2026 | 5,000 |
| Mar 2026 | 5,000 |
| Apr 2026 | 14,000 |
| May 2026 | 2,000 |
| Jun 2026 | 23,000 |
| Jul 2026 | 9,000 |
| Aug 2026 | 6,000 |
| Sep 2026 | 4,000 |
| Oct 2026 | 4,000 |
| Nov 2026 | 2,000 |
Downloads over the 30 days before the last snapshot of each month.
Popular in the same category
- clip-japanese-baseline-corporation34kLY Corporation’s clip‑japanese‑base is a Japanese CLIP model trained on ~1 B image‑text pairs, using an Eva02‑B transformer image encoder with a 12‑layer BERT text encoder, achieving R@1 0.30 on STAIR, 0.89 accuracy on Recruit and 0.58 accuracy on ImageNet‑1K, and supporting zero‑shot image classification and retrieval.
- japanese-clip-vit-b-32-roberta-baserecruit-jp1kA Japanese CLIP model combining the Laion ViT‑B‑32 image encoder with rinna/japanese‑roberta‑base, trained on a Japanese subset of LAION2B‑multi for zero‑shot classification, retrieval, and feature extraction under a CC‑BY‑4.0 license.
- sentence-bert-base-ja-mean-tokens-v2sonoisa53kA Japanese Sentence‑BERT v2, fine‑tuned on cl‑tohoku/bert‑base‑japanese‑whole‑word‑masking with MultipleNegativesRankingLoss, boosts accuracy by ~1.5–2 % over v1 and is released as sonoisa/sentence‑bert‑base‑ja‑mean‑tokens‑v2.
- japanese-clip-vit-b-16rinna40krinna/japanese-clip‑vit‑b‑16 is an Apache‑2.0 licensed Japanese CLIP model based on ViT‑B/16, trained on CC12M captions translated to Japanese and released on May 12 2022.
- transformers-ud-japanese-electra-base-ginza-510megagonlabs22kja_ginza_electra is a spaCy v3 Python package offering a Japanese ELECTRA model fine‑tuned on mC4 and UD_Japanese_BCCWJ r2.8 (based on megagonlabs/transformers‑ud‑japanese‑electra‑base‑discrimininator) with custom bunsetu‑phrase detection, distributed under the MIT license.
- sentence-bert-base-ja-mean-tokenssonoisa14kJapanese Sentence‑BERT (v1) model for generating sentence embeddings, with an improved v2 available and sample usage via Hugging Face Transformers and a custom SentenceBertJapanese class.
Resources in similar research fields
- sup-simcse-ja-basecl-nagoya2kA Japanese BERT‑base model fine‑tuned with supervised SimCSE on JSNLI, exposed via Sentence‑Transformers or HuggingFace with CLS pooling, trained on 1 M examples at 512‑batch size, 5 × 10⁻⁵ learning rate, 5 × 10⁻⁵ temperature, 64‑token limit, and BFloat16 precision.Shared: Representation Learning
- japanese-hubert-baseyky-h28kJapanese HuBERT Base, a 12‑layer transformer mirroring rinna’s original, trained on ~19,000 hrs of ReazonSpeech v1 Japanese speech and released under the Apache 2.0 license.Shared: Representation Learning
- sarashina-embedding-v2-1bsbintuitions2kSarashina‑Embedding‑v2‑1B is a 1,792‑dimensional Japanese sentence transformer trained with multi‑stage contrastive learning that achieves state‑of‑the‑art JMTEB scores, and can be used for semantic similarity, search, paraphrase mining, classification, and clustering via Sentence‑Transformers with optional instruction prefixes.Shared: Representation Learning
- ruri-v3-310mcl-nagoya634kRuri v3 is a state‑of‑the‑art Japanese text‑embedding model built on ModernBERT‑Ja, supporting up to 8,192‑token inputs, a 100K‑token vocabulary, FlashAttention‑accelerated inference, and multiple size variants for fast sentence‑transformer usage.Shared: Representation Learning
- ruri-v3-30mcl-nagoya126kRuri v3 is a state‑of‑the‑art Japanese text embedding model built on ModernBERT‑Ja, supporting up to 8,192 tokens, a 100 k‑token vocabulary, FlashAttention acceleration, and multiple sizes from 37 M to 315 M parameters.Shared: Representation Learning
- ruri-v3-130mcl-nagoya93kRuri v3 is a state‑of‑the‑art Japanese text‑embedding model built on ModernBERT‑Ja, supporting up to 8192‑token sequences, a 100K‑token vocabulary, FlashAttention, and released in sizes from 30 M to 310 M parameters for use with sentence‑transformers.Shared: Representation Learning