NetsPresso_QA / pyserini /resources /index-metadata /faiss.mrtydi-v1.1-finnish.20220207.5df364.README.md
geonmin-kim's picture
Upload folder using huggingface_hub
d6585f5

A newer version of the Gradio SDK is available: 4.37.2

Upgrade

mrtydi-v1.1-finnish

Faiss flat index for Mr.TyDi v1.1 (Finnish), using mDPR fine-tuned on NQ.

This index was generated on 2022/02/07 at commit 5df364 on narval with the following command:

lang=finnish

tarfn=mrtydi-v1.1-$lang.tar.gz
encoder=models/mdpr-context-encoder
corpus=mrtydi-v1.1-$lang/collection/docs.jsonl
index_dir=mrtydi-mdpr-dindex/$lang

wget https://git.uwaterloo.ca/jimmylin/mr.tydi/-/raw/master/data/$tarfn
tar –xvf $tarfn
gzip -cvf $corpus.gz > $corpus

mkdir -p $index_dir

python -m pyserini.encode   input   --corpus $corpus \
                                    --fields title text \
                                    --delimiter "\n\n" \
                            output  --embeddings  $index_dir \
                                    --to-faiss \
                            encoder --encoder $encoder \
                                    --fields title text \
                                    --batch 128 \
                                    --fp16

Note that the delimiter was manually changed from "\n" into "\n\n" in pyserini.encode. This was later generalized into a command-line option in Pyserini #1000.

Here's a sample retrieval command (on the test set):

set_name=test
python -m pyserini.dsearch \
  --encoder castorini/mdpr-question-nq \
  --topics mrtydi-v1.1-${lang}-${set_name} \
  --index ${index_dir} \
  --output runs/run.mrtydi-v1.1-$lang.${set_name}.txt
  --batch-size 36 \
  --threads 12