File size: 1,155 Bytes
6c639a4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 |
#!/bin/bash
WORLD_SIZE=8
DISTRIBUTED_ARGS="--nproc_per_node $WORLD_SIZE \
--nnodes 1 \
--node_rank 0 \
--master_addr localhost \
--master_port 6000"
TASK="LAMBADA"
VALID_DATA=<lambada path>
VOCAB_FILE=gpt2-vocab.json
MERGE_FILE=gpt2-merges.txt
CHECKPOINT=checkpoints/gpt2_345m
python -m torch.distributed.launch $DISTRIBUTED_ARGS ./tasks/main.py \
--task $TASK \
--valid-data $VALID_DATA \
--tokenizer-type GPT2BPETokenizer \
--strict-lambada \
--vocab-file $VOCAB_FILE \
--merge-file $MERGE_FILE \
--load $CHECKPOINT \
--tensor-model-parallel-size 1 \
--num-layers 24 \
--hidden-size 1024 \
--num-attention-heads 16 \
--batch-size 8 \
--activations-checkpoint-method uniform \
--seq-length 1024 \
--max-position-embeddings 1024 \
--log-interval 10 \
--fp16 \
--no-load-optim \
--no-load-rng
|