Merge branch 'maanug/1gpu-pretrain-examples' into 'main'

Fix pretraining examples

See merge request ADLR/megatron-lm!513

examples/pretrain_bert.sh

 #!/bin/bash
 
-RANK=0
-WORLD_SIZE=1
-DATA_PATH=<Specify path and file prefix>_text_sentence
+export CUDA_DEVICE_MAX_CONNECTIONS=1
+
 CHECKPOINT_PATH=<Specify path>
+VOCAB_FILE=<Specify path to file>/bert-vocab.txt
+DATA_PATH=<Specify path and file prefix>_text_sentence
+
+BERT_ARGS="
+    --num-layers 24 \
+    --hidden-size 1024 \
+    --num-attention-heads 16 \
+    --seq-length 512 \
+    --max-position-embeddings 512 \
+    --micro-batch-size 4 \
+    --global-batch-size 8 \
+    --lr 0.0001 \
+    --train-iters 2000000 \
+    --lr-decay-iters 990000 \
+    --lr-decay-style linear \
+    --min-lr 0.00001 \
+    --weight-decay 1e-2 \
+    --lr-warmup-fraction .01 \
+    --clip-grad 1.0 \
+    --fp16
+"
+
+DATA_ARGS="
+    --data-path $DATA_PATH \
+    --vocab-file $VOCAB_FILE \
+    --data-impl mmap \
+    --split 949,50,1
+"
+
+OUTPUT_ARGS="
+    --log-interval 100 \
+    --save-interval 10000 \
+    --eval-interval 1000 \
+    --eval-iters 10
+"
 
-python pretrain_bert.py \
-       --num-layers 24 \
-       --hidden-size 1024 \
-       --num-attention-heads 16 \
-       --micro-batch-size 4 \
-       --global-batch-size 8 \
-       --seq-length 512 \
-       --max-position-embeddings 512 \
-       --train-iters 2000000 \
-       --lr-decay-iters 990000 \
-       --save $CHECKPOINT_PATH \
-       --load $CHECKPOINT_PATH \
-       --data-path $DATA_PATH \
-       --vocab-file bert-vocab.txt \
-       --data-impl mmap \
-       --split 949,50,1 \
-       --lr 0.0001 \
-       --min-lr 0.00001 \
-       --lr-decay-style linear \
-       --lr-warmup-fraction .01 \
-       --weight-decay 1e-2 \
-       --clip-grad 1.0 \
-       --log-interval 100 \
-       --save-interval 10000 \
-       --eval-interval 1000 \
-       --eval-iters 10 \
-       --fp16
+torchrun pretrain_bert.py \
+    $BERT_ARGS \
+    $DATA_ARGS \
+    $OUTPUT_ARGS \
+    --save $CHECKPOINT_PATH \
+    --load $CHECKPOINT_PATH

examples/pretrain_gpt.sh

-#! /bin/bash
+#!/bin/bash
 
 # Runs the "345M" parameter model
 
-RANK=0
-WORLD_SIZE=1
+export CUDA_DEVICE_MAX_CONNECTIONS=1
 
-DATA_PATH=<Specify path and file prefix>_text_document
 CHECKPOINT_PATH=<Specify path>
+VOCAB_FILE=<Specify path to file>/gpt2-vocab.json
+MERGE_FILE=<Specify path to file>/gpt2-merges.txt
+DATA_PATH=<Specify path and file prefix>_text_document
+
+GPT_ARGS="
+    --num-layers 24 \
+    --hidden-size 1024 \
+    --num-attention-heads 16 \
+    --seq-length 1024 \
+    --max-position-embeddings 1024 \
+    --micro-batch-size 4 \
+    --global-batch-size 8 \
+    --lr 0.00015 \
+    --train-iters 500000 \
+    --lr-decay-iters 320000 \
+    --lr-decay-style cosine \
+    --min-lr 1.0e-5 \
+    --weight-decay 1e-2 \
+    --lr-warmup-fraction .01 \
+    --clip-grad 1.0 \
+    --fp16
+"
+
+DATA_ARGS="
+    --data-path $DATA_PATH \
+    --vocab-file $VOCAB_FILE \
+    --merge-file $MERGE_FILE \
+    --data-impl mmap \
+    --split 949,50,1
+"
 
+OUTPUT_ARGS="
+    --log-interval 100 \
+    --save-interval 10000 \
+    --eval-interval 1000 \
+    --eval-iters 10
+"
 
-python pretrain_gpt.py \
-       --num-layers 24 \
-       --hidden-size 1024 \
-       --num-attention-heads 16 \
-       --micro-batch-size 4 \
-       --global-batch-size 8 \
-       --seq-length 1024 \
-       --max-position-embeddings 1024 \
-       --train-iters 500000 \
-       --lr-decay-iters 320000 \
-       --save $CHECKPOINT_PATH \
-       --load $CHECKPOINT_PATH \
-       --data-path $DATA_PATH \
-       --vocab-file gpt2-vocab.json \
-       --merge-file gpt2-merges.txt \
-       --data-impl mmap \
-       --split 949,50,1 \
-       --distributed-backend nccl \
-       --lr 0.00015 \
-       --min-lr 1.0e-5 \
-       --lr-decay-style cosine \
-       --weight-decay 1e-2 \
-       --clip-grad 1.0 \
-       --lr-warmup-fraction .01 \
-       --activations-checkpoint-method uniform \
-       --log-interval 100 \
-       --save-interval 10000 \
-       --eval-interval 1000 \
-       --eval-iters 10 \
-       --fp16
+torchrun pretrain_gpt.py \
+    $GPT_ARGS \
+    $DATA_ARGS \
+    $OUTPUT_ARGS \
+    --save $CHECKPOINT_PATH \
+    --load $CHECKPOINT_PATH

examples/pretrain_t5.sh

 #!/bin/bash
 
-RANK=0
-WORLD_SIZE=1
-DATA_PATH=<Specify path and file prefix>
-VOCAB_FILE=<Specify path to vocab.txt>
+export CUDA_DEVICE_MAX_CONNECTIONS=1
+
 CHECKPOINT_PATH=<Specify path>
+VOCAB_FILE=<Specify path to file>/t5-vocab.txt
+DATA_PATH=<Specify path and file prefix>_text_sentence
+
+T5_ARGS="
+    --num-layers 12 \
+    --hidden-size 768 \
+    --num-attention-heads 12 \
+    --kv-channels 64 \
+    --ffn-hidden-size 3072 \
+    --encoder-seq-length 512 \
+    --decoder-seq-length 128 \
+    --max-position-embeddings 512 \
+    --micro-batch-size 16 \
+    --global-batch-size 16 \
+    --lr 0.0001 \
+    --train-iters 1000000 \
+    --lr-decay-iters 1000000 \
+    --lr-decay-style linear \
+    --min-lr 0.00001 \
+    --weight-decay 1e-2 \
+    --lr-warmup-fraction .01 \
+    --clip-grad 1.0 \
+    --fp16 \
+    --vocab-extra-ids 100
+"
+
+DATA_ARGS="
+    --data-path $DATA_PATH \
+    --vocab-file $VOCAB_FILE \
+    --data-impl mmap \
+    --split 949,50,1
+"
+
+OUTPUT_ARGS="
+    --log-interval 100 \
+    --save-interval 10000 \
+    --eval-interval 1000 \
+    --eval-iters 10
+"
 
-python pretrain_t5.py \
-       --num-layers 12 \
-       --hidden-size 768 \
-       --num-attention-heads 12 \
-       --kv-channels 64 \
-       --ffn-hidden-size 3072 \
-       --encoder-seq-length 512 \
-       --decoder-seq-length 128 \
-       --micro-batch-size 16 \
-       --global-batch-size 16 \
-       --max-position-embeddings 512 \
-       --train-iters 1000000 \
-       --lr-decay-iters 1000000 \
-       --save $CHECKPOINT_PATH \
-       --load $CHECKPOINT_PATH \
-       --data-path $DATA_PATH \
-       --vocab-file $VOCAB_FILE \
-       --data-impl mmap \
-       --split 949,50,1 \
-       --lr 0.0001 \
-       --min-lr 0.00001 \
-       --lr-decay-style linear \
-       --lr-warmup-fraction .01 \
-       --weight-decay 1e-2 \
-       --clip-grad 1.0 \
-       --log-interval 100 \
-       --save-interval 10000 \
-       --eval-interval 1000 \
-       --eval-iters 10 \
-       --fp16 \
-       --vocab-extra-ids 100
+torchrun pretrain_t5.py \
+    $T5_ARGS \
+    $DATA_ARGS \
+    $OUTPUT_ARGS \
+    --save $CHECKPOINT_PATH \
+    --load $CHECKPOINT_PATH