|
--- |
|
license: cc-by-sa-4.0 |
|
language: |
|
- ja |
|
--- |
|
|
|
The following data set was vectorized with the [intfloat/multilingual-e5-base](https://huggingface.co/intfloat/multilingual-e5-base) model and an index file created by faiss. |
|
|
|
[oshizo/japanese-wikipedia-paragraphs](https://huggingface.co/datasets/oshizo/japanese-wikipedia-paragraphs) |
|
|
|
|
|
|
|
## Usage |
|
|
|
First, download index_me5-base_IVF2048_PQ192.faiss from this repository. |
|
|
|
```python |
|
import faiss |
|
import datasets |
|
from sentence_transformers import SentenceTransformer |
|
|
|
ds = datasets.load_dataset("oshizo/japanese-wikipedia-paragraphs", split="train") |
|
|
|
index = faiss.read_index("./index_me5-base_IVF2048_PQ192.faiss") |
|
|
|
model = SentenceTransformer("intfloat/multilingual-e5-base") |
|
|
|
question = "日本で二番目に高い山は？" |
|
emb = model.encode(["query: " + question]) |
|
scores, indexes = index.search(emb, 10) |
|
scores = scores[0] |
|
indexes = indexes[0] |
|
|
|
results = [] |
|
for idx, score in zip(indexes, scores): |
|
idx = int(idx) |
|
passage = ds[idx] |
|
passage["score"] = score |
|
results.append((passage)) |
|
|
