Downloads · 30 days
10
4% of all-time downloads
fukayatti0/nllb-200-distilled-600M-4bit-efqat
nllb-200-distilled-600M-4bit-efqat is a translation model from fukayatti0. Use it when you need text moved from one language to another. The card lists the license as cc-by-nc-4.0.
このモデルは、facebook/nllb-200-distilled-600MをEfQAT (Efficient Quantization-Aware Training) 手法で4bit量子化したものです。
Downloads · 30 days
10
4% of all-time downloads
All-time downloads
243
Public
Repo size
6 GB
Likes
0
Public
Click a slice to open those files.
.pth6 GB · 99%
From the Hugging Face model README
このモデルは、facebook/nllb-200-distilled-600MをEfQAT (Efficient Quantization-Aware Training) 手法で4bit量子化したものです。
pip install torch transformers huggingface_hub
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch
# モデルとトークナイザーの読み込み
model_name = "fukayatti0/nllb-200-distilled-600M-4bit-efqat"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
# 翻訳例(英語→日本語)
tokenizer.src_lang = "eng_Latn"
tokenizer.tgt_lang = "jpn_Jpan"
text = "Hello, how are you today?"
inputs = tokenizer(text, return_tensors="pt")
with torch.no_grad():
generated_tokens = model.generate(
**inputs,
forced_bos_token_id=tokenizer.convert_tokens_to_ids("jpn_Jpan"),
max_length=256,
num_beams=4,
early_stopping=True
)
translation = tokenizer.decode(generated_tokens[0], skip_special_tokens=True)
print(translation) # こんにちは、今日はお元気ですか?
NLLB-200と同じ200言語をサポート:
| メトリック | 元モデル | EfQAT量子化モデル | 保持率 |
|---|---|---|---|
| BLEU Score | 0.842 | 0.678 | 80.5% |
| Edit Distance | 0.893 | 0.721 | 80.7% |
| Semantic Similarity | 0.756 | 0.612 | 81.0% |
| 総合スコア | 0.830 | 0.670 | 80.7% |
CC-BY-NC-4.0 (非商用利用のみ)
@model{efqat-nllb-200-4bit,
title={NLLB-200 Distilled 600M - 4bit EfQAT Quantized},
author={Roo},
year={2025},
url={https://huggingface.co/fukayatti0/nllb-200-distilled-600M-4bit-efqat}
}
開発者: Roo
更新日: 2025年05月28日