GLEN-model / scripts /eval_make_docid_glen_marco.sh
QuanTH02's picture
Phase 1
3d5551b
raw
history blame contribute delete
684 Bytes
NUM_MULTI_VECTORS=7
MAX_OUTPUT_LENGTH=$(($NUM_MULTI_VECTORS + 1))
# you can use --infer_dir or --infer_ckpt for other checkpoints
# e.g., --infer_dir logs/model_glen_nq/GLEN_P1_base/checkpoint-1234
# e.g., --infer_ckpt logs/model_glen_nq/GLEN_P1_base/checkpoint-1234/pytorch_model.bin
CUDA_VISIBLE_DEVICES=0 python examples/glen_phase2/makeid_glen.py \
--model_name_or_path t5-base \
--dataset_name marco_passage \
--max_input_length 156 \
--per_device_eval_batch_size 16 \
--id_class t5_bm25_3 \
--docid_file_name glen_marco_docid \
--max_output_length $MAX_OUTPUT_LENGTH \
--num_return_sequences 10 \
--infer_dir checkpoint/glen_p2_marco \
--output_dir output