Generate.md

create embeddings

multilingual-e5-small + passage

python datasets_to_embs.py \
  --target="data" \
  --model_name="intfloat/multilingual-e5-small" \
  --output_name="multilingual-e5-small-passage" \
  --input_prefix="passage: "
python datasets_to_embs.py \
  --target="chunked" \
  --model_name="intfloat/multilingual-e5-small" \
  --output_name="multilingual-e5-small-passage" \
  --input_prefix="passage: "

multilingual-e5-small + query

python datasets_to_embs.py \
  --target="data" \
  --model_name="intfloat/multilingual-e5-small" \
  --output_name="multilingual-e5-small-query" \
  --input_prefix="query: "
python datasets_to_embs.py \
  --target="chunked" \
  --model_name="intfloat/multilingual-e5-small" \
  --output_name="multilingual-e5-small-query" \
  --input_prefix="query: "

GLuCoSE-base-ja

python datasets_to_embs.py \
  --target="data" \
  --model_name="pkshatech/GLuCoSE-base-ja" \
  --output_name="GLuCoSE-base-ja"
python datasets_to_embs.py \
  --target="chunked" \
  --model_name="pkshatech/GLuCoSE-base-ja" \
  --output_name="GLuCoSE-base-ja"

cl-nagoya/sup-simcse-ja-base

python datasets_to_embs.py \
  --target="data" \
  --model_name="cl-nagoya/sup-simcse-ja-base" \
  --output_name="sup-simcse-ja-base"
python datasets_to_embs.py \
  --target="chunked" \
  --model_name="cl-nagoya/sup-simcse-ja-base" \
  --output_name="sup-simcse-ja-base"

create faiss indexes

python embs_to_faiss.py \
  --target="data" \
  --input_name="multilingual-e5-small-passage" \
  --use_gpu

generate all by ./embs/

python all_embs_to_faiss.py --use_gpu

XXX

• 同じ output があったら skip する