Training own entity disambiguation - Cannot find trained model in models directory

[KemalDrop]

I am trying to train my own entity disambiguation model. I used this script for the preprocessing

#!/bin/bash

# Copyright (c) Facebook, Inc. and its affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.


DATASET_PATH=$1
MODEL_PATH=$2

echo "Processing ${DATASET}"

cd ../fairseq

# BPE preprocessing.
for SPLIT in train dev; do
    for LANG in "source" "target"; do
        python -m examples.roberta.multiprocessing_bpe_encoder\
            --encoder-json "$MODEL_PATH/encoder.json" \
            --vocab-bpe "$MODEL_PATH//vocab.bpe" \
            --inputs "$DATASET_PATH/$SPLIT.$LANG" \
            --outputs "$DATASET_PATH/$SPLIT.bpe.$LANG" \
            --workers 60 \
            --keep-empty;
    done
done

cd ..

# Binarize the dataset.
fairseq-preprocess --joined-dictionary \
    --source-lang "source" --target-lang "target" \
    --trainpref "$DATASET_PATH/train.bpe" \
    --validpref "$DATASET_PATH/dev.bpe" \
    --destdir "$DATASET_PATH/bin" \
    --workers 60

• I created a dataset where train.source contains I am going to the [START_ENT] States [END_ENT]. and train.target has United States
• Since it is not reported I used GPT-2 encoder.json and vocab.bpe but open to suggestions here, I might have done wrong there.
• I used --joined-dictionary option otherwise I got an error at training stating that source and target should be joined dictionary.
• Without --source-lang and --target-lang I also get an error that is why I put those in.

After doing that I used this bash file for training

#!/bin/bash
# Copyright (c) Facebook, Inc. and its affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.

fairseq-train /home/ubuntu/kemal/entity_disambiguation_data/v0/bin/ \
    --save-dir models/$NAME \
    --tensorboard-logdir tensorboard_logs/$NAME \
    --restore-file models/bart.large/model.pt \
    --arch bart_large  \
    --task translation \
    --criterion label_smoothed_cross_entropy  \
    --source-lang source  \
    --target-lang target  \
    --truncate-source  \
    --label-smoothing 0.1  \
    --max-tokens 1024  \
    --update-freq 1  \
    --max-update 200000  \
    --required-batch-size-multiple 1  \
    --dropout 0.1  \
    --attention-dropout 0.1  \
    --relu-dropout 0.0  \
    --weight-decay 0.01  \
    --optimizer adam  \
    --adam-betas "(0.9, 0.999)"  \
    --adam-eps 1e-08  \
    --clip-norm 0.1  \
    --lr-scheduler polynomial_decay  \
    --lr 3e-05  \
    --total-num-update 200000  \
    --warmup-updates 500  \
    --ddp-backend no_c10d  \
    --num-workers 20  \
    --reset-meters  \
    --reset-optimizer \
    --layernorm-embedding \
    --share-decoder-input-output-embed  \
    --share-all-embeddings \
    --skip-invalid-size-inputs-valid-test  \
    --log-format json  \
    --log-interval 10  \
    --patience 200

and training started. However what I want to ask is:

1. Are these configurations correct for disambiguation? Especially pre-processing part?
2. When I checked the checkpoint.pt files I can see there is a data.pkl which i will use later for inference I think as specified here https://github.com/facebookresearch/GENRE/tree/main/examples_genre. However I cannot find the actual model.pt file in the checkpoint.pt. When I go into the checkpoint file there is an archive file and when I open that there is data.pkl and data folder which has additional binary files inside but there is no model.pt thus I cannot import the model and use it. How can I solve this problem?
3. The evaluation metric is perplexity I want to know the KB accuracy scores how can I change it to that?