Paraphrase Multilingual MiniLM L12 V2.gguf

🚀 paraphrase-multilingual-MiniLM-L12-v2.gguf

This project provides a multilingual paraphrase model, which can generate embeddings for text in multiple languages and calculate the cosine distance between them.

🚀 Quick Start

The following is a sample code demonstrating how to use the paraphrase-multilingual-MiniLM-L12-v2.gguf model to generate text embeddings and calculate cosine distances.

Basic Usage

import torch
from llama_cpp import Llama
from sentence_transformers import SentenceTransformer
from scipy.spatial.distance import cosine

model = SentenceTransformer(
    "paraphrase-multilingual-MiniLM-L12-v2",
    model_kwargs={"torch_dtype": torch.float16}
)
llm = Llama.from_pretrained(
    "mykor/paraphrase-multilingual-MiniLM-L12-v2.gguf",
    filename="paraphrase-multilingual-MiniLM-L12-118M-v2-F16.gguf",
    embedding=True,
    verbose=False,
)

text = "움츠러든 어깨를 따라서 다시 저물어가는 오늘의 끝 밤이 조용히 나를 안으면 무너져가는 날 잊어버릴 수 있어"
embed1 = model.encode(text)
embed2 = llm.embed(text)
print(cosine(embed1, embed2))

Output Result

0.005355771841081269

📄 License

This project is licensed under the Apache-2.0 license.