Add TB

sbatch_4b284bc4perplexity.sh

@@ -0,0 +1,163 @@
+#!/bin/bash
+#SBATCH --nodes=32
+#SBATCH --ntasks-per-node=1
+#SBATCH --cpus-per-task=40
+#SBATCH --mem=256G
+#SBATCH -p standard-g
+#SBATCH -t 48:00:00
+#SBATCH --gpus-per-node=mi250:8
+#SBATCH --exclusive=user
+#SBATCH --hint=nomultithread
+#SBATCH --account=project_462000119
+#SBATCH -o logs/%j.out
+#SBATCH -e logs/%j.err
+
+
+VARIANT=4b284bc4perplexity
+
+# if run without sbatch, invoke here
+if [ -z $SLURM_JOB_ID ]; then
+    mkdir -p logs
+    sbatch "$0"
+    exit
+fi
+
+set -euo pipefail
+
+# symlink logs/latest.out and logs/latest.err
+ln -f -s $SLURM_JOB_ID.out logs/latest.out
+ln -f -s $SLURM_JOB_ID.err logs/latest.err
+
+KILL_SWITCH_PATH=kill-switch-$VARIANT
+CHECKPOINT_PATH=checkpoints_$VARIANT
+TENSORBOARD_PATH=tensorboard_$VARIANT
+
+# Data
+VOCAB_FILE="gpt2/vocab.json"
+MERGE_FILE="gpt2/merges.txt"
+DATA_PATH="/scratch/project_462000119/data/c4perplexity/gpt2tok_perplexity_text_document"
+TRAIN_DATA_PATH=trainperplexity.txt
+# "train: 1.0 0:1 /scratch/project_462000119/data/c4perplexity/gpt2tok_perplexity_text_document"
+VALID_DATA_PATH=val.txt
+# "validation: 1.0 0:1 /scratch/project_462000119/data/c4_validation/gpt2tok_c4validation_rerun_text_document"
+
+PP_SIZE=1
+TP_SIZE=2
+
+MICRO_BATCH_SIZE=2
+GRADIENT_ACCUMULATION_STEPS=1
+WORLD_SIZE=$((SLURM_GPUS_ON_NODE*SLURM_JOB_NUM_NODES))
+GLOBAL_BATCH_SIZE=$((MICRO_BATCH_SIZE*WORLD_SIZE*GRADIENT_ACCUMULATION_STEPS))
+
+# Model parameters
+source model_params.sh
+MODEL_PARAM=("${PARAM_2980M[@]}")
+NHIDDEN=${MODEL_PARAM[0]}
+FFN_HIDDEN_SIZE=${MODEL_PARAM[1]}
+KV_SIZE=${MODEL_PARAM[2]}
+NHEADS=${MODEL_PARAM[3]}
+NLAYERS=${MODEL_PARAM[4]}
+SEQ_LEN=2048
+
+echo "Model parameters: d_model $NHIDDEN ffw_size $FFN_HIDDEN_SIZE kv_size $KV_SIZE n_heads $NHEADS n_layers $NLAYERS"
+
+SAVE_INTERVAL=10000
+
+# Tokens: 84_000_000_000
+# -> Samples: 41_015_625.0
+TRAIN_SAMPLES=41_015_625
+
+OPTIMIZER_ARGS=" \
+    --optimizer adam \
+    --adam-beta1 0.9 \
+    --adam-beta2 0.95 \
+    --adam-eps 1e-8 \
+    --lr 2e-4 \
+    --min-lr 2e-5 \
+    --lr-decay-style cosine \
+    --lr-decay-samples $TRAIN_SAMPLES \
+    --lr-warmup-samples 410_156 \
+    --clip-grad 1.0 \
+    --weight-decay 1e-1 \
+    "
+
+GPT_ARGS=" \
+    --num-layers $NLAYERS \
+    --hidden-size $NHIDDEN \
+    --num-attention-heads $NHEADS \
+    --kv-channels $KV_SIZE \
+    --ffn-hidden-size $FFN_HIDDEN_SIZE \
+    --seq-length $SEQ_LEN \
+    --max-position-embeddings $SEQ_LEN \
+    --micro-batch-size $MICRO_BATCH_SIZE \
+    --global-batch-size $GLOBAL_BATCH_SIZE \
+    --train-samples $TRAIN_SAMPLES \
+    --vocab-file $VOCAB_FILE \
+    --merge-file $MERGE_FILE \
+    --clip-grad 1.0 \
+    --kill-switch-path $KILL_SWITCH_PATH \
+    --bf16 \
+    $OPTIMIZER_ARGS \
+    "
+
+OUTPUT_ARGS=" \
+    --log-interval 10 \
+    --save-interval $SAVE_INTERVAL \
+    --eval-interval 1000 \
+    --eval-iters 100 \
+    --tensorboard-dir $TENSORBOARD_PATH \
+    --tensorboard-queue-size 5 \
+    --log-timers-to-tensorboard \
+    --log-batch-size-to-tensorboard \
+    --log-validation-ppl-to-tensorboard \
+    "
+
+ZERO_STAGE=0
+
+mkdir -p ds_configs
+DS_CONFIG_PATH="ds_configs/$SLURM_JOB_ID.json"
+
+cat <<EOF > $DS_CONFIG_PATH
+{
+    "train_micro_batch_size_per_gpu": $MICRO_BATCH_SIZE,
+    "train_batch_size": $GLOBAL_BATCH_SIZE,
+    "gradient_clipping": 1.0,
+    "zero_optimization": {
+        "stage": $ZERO_STAGE
+    },
+    "bf16": {
+        "enabled": true
+    },
+    "steps_per_print": 2000,
+    "wall_clock_breakdown": false
+}
+EOF
+
+DEEPSPEED_ARGS=" \
+    --deepspeed \
+    --deepspeed_config $DS_CONFIG_PATH \
+    --zero-stage $ZERO_STAGE \
+    "
+
+CMD=" \
+    Megatron-DeepSpeed/pretrain_gpt.py \
+    --tensor-model-parallel-size $TP_SIZE \
+    --pipeline-model-parallel-size $PP_SIZE \
+    $GPT_ARGS \
+    $OUTPUT_ARGS \
+    --save $CHECKPOINT_PATH \
+    --load $CHECKPOINT_PATH \
+    --train-weighted-split-paths-path $TRAIN_DATA_PATH \
+    --valid-weighted-split-paths-path $VALID_DATA_PATH \
+    --data-impl mmap \
+     $DEEPSPEED_ARGS \
+    "
+
+echo $CMD
+
+echo "START $SLURM_JOBID: $(date)"
+
+# bash launch_srun.sh $CMD
+srun --label launch.sh $CMD
+
+echo "END $SLURM_JOBID: $(date)"

sbatch_4b284bc4perplexity25.sh

@@ -0,0 +1,163 @@
+#!/bin/bash
+#SBATCH --nodes=32
+#SBATCH --ntasks-per-node=1
+#SBATCH --cpus-per-task=40
+#SBATCH --mem=256G
+#SBATCH -p standard-g
+#SBATCH -t 48:00:00
+#SBATCH --gpus-per-node=mi250:8
+#SBATCH --exclusive=user
+#SBATCH --hint=nomultithread
+#SBATCH --account=project_462000119
+#SBATCH -o logs/%j.out
+#SBATCH -e logs/%j.err
+
+
+VARIANT=4b284bc4perplexity25
+
+# if run without sbatch, invoke here
+if [ -z $SLURM_JOB_ID ]; then
+    mkdir -p logs
+    sbatch "$0"
+    exit
+fi
+
+set -euo pipefail
+
+# symlink logs/latest.out and logs/latest.err
+ln -f -s $SLURM_JOB_ID.out logs/latest.out
+ln -f -s $SLURM_JOB_ID.err logs/latest.err
+
+KILL_SWITCH_PATH=kill-switch-$VARIANT
+CHECKPOINT_PATH=checkpoints_$VARIANT
+TENSORBOARD_PATH=tensorboard_$VARIANT
+
+# Data
+VOCAB_FILE="gpt2/vocab.json"
+MERGE_FILE="gpt2/merges.txt"
+DATA_PATH="/scratch/project_462000119/data/c4perplexity/gpt2tok_perplexity_text_document"
+TRAIN_DATA_PATH=trainperplexity25.txt
+# "train: 1.0 0:1 /scratch/project_462000119/data/c4perplexity/gpt2tok_c4_en_pplx_25_text_document"
+VALID_DATA_PATH=val.txt
+# "validation: 1.0 0:1 /scratch/project_462000119/data/c4_validation/gpt2tok_c4validation_rerun_text_document"
+
+PP_SIZE=1
+TP_SIZE=2
+
+MICRO_BATCH_SIZE=2
+GRADIENT_ACCUMULATION_STEPS=1
+WORLD_SIZE=$((SLURM_GPUS_ON_NODE*SLURM_JOB_NUM_NODES))
+GLOBAL_BATCH_SIZE=$((MICRO_BATCH_SIZE*WORLD_SIZE*GRADIENT_ACCUMULATION_STEPS))
+
+# Model parameters
+source model_params.sh
+MODEL_PARAM=("${PARAM_2980M[@]}")
+NHIDDEN=${MODEL_PARAM[0]}
+FFN_HIDDEN_SIZE=${MODEL_PARAM[1]}
+KV_SIZE=${MODEL_PARAM[2]}
+NHEADS=${MODEL_PARAM[3]}
+NLAYERS=${MODEL_PARAM[4]}
+SEQ_LEN=2048
+
+echo "Model parameters: d_model $NHIDDEN ffw_size $FFN_HIDDEN_SIZE kv_size $KV_SIZE n_heads $NHEADS n_layers $NLAYERS"
+
+SAVE_INTERVAL=10000
+
+# Tokens: 84_000_000_000
+# -> Samples: 41_015_625.0
+TRAIN_SAMPLES=41_015_625
+
+OPTIMIZER_ARGS=" \
+    --optimizer adam \
+    --adam-beta1 0.9 \
+    --adam-beta2 0.95 \
+    --adam-eps 1e-8 \
+    --lr 2e-4 \
+    --min-lr 2e-5 \
+    --lr-decay-style cosine \
+    --lr-decay-samples $TRAIN_SAMPLES \
+    --lr-warmup-samples 410_156 \
+    --clip-grad 1.0 \
+    --weight-decay 1e-1 \
+    "
+
+GPT_ARGS=" \
+    --num-layers $NLAYERS \
+    --hidden-size $NHIDDEN \
+    --num-attention-heads $NHEADS \
+    --kv-channels $KV_SIZE \
+    --ffn-hidden-size $FFN_HIDDEN_SIZE \
+    --seq-length $SEQ_LEN \
+    --max-position-embeddings $SEQ_LEN \
+    --micro-batch-size $MICRO_BATCH_SIZE \
+    --global-batch-size $GLOBAL_BATCH_SIZE \
+    --train-samples $TRAIN_SAMPLES \
+    --vocab-file $VOCAB_FILE \
+    --merge-file $MERGE_FILE \
+    --clip-grad 1.0 \
+    --kill-switch-path $KILL_SWITCH_PATH \
+    --bf16 \
+    $OPTIMIZER_ARGS \
+    "
+
+OUTPUT_ARGS=" \
+    --log-interval 10 \
+    --save-interval $SAVE_INTERVAL \
+    --eval-interval 1000 \
+    --eval-iters 100 \
+    --tensorboard-dir $TENSORBOARD_PATH \
+    --tensorboard-queue-size 5 \
+    --log-timers-to-tensorboard \
+    --log-batch-size-to-tensorboard \
+    --log-validation-ppl-to-tensorboard \
+    "
+
+ZERO_STAGE=0
+
+mkdir -p ds_configs
+DS_CONFIG_PATH="ds_configs/$SLURM_JOB_ID.json"
+
+cat <<EOF > $DS_CONFIG_PATH
+{
+    "train_micro_batch_size_per_gpu": $MICRO_BATCH_SIZE,
+    "train_batch_size": $GLOBAL_BATCH_SIZE,
+    "gradient_clipping": 1.0,
+    "zero_optimization": {
+        "stage": $ZERO_STAGE
+    },
+    "bf16": {
+        "enabled": true
+    },
+    "steps_per_print": 2000,
+    "wall_clock_breakdown": false
+}
+EOF
+
+DEEPSPEED_ARGS=" \
+    --deepspeed \
+    --deepspeed_config $DS_CONFIG_PATH \
+    --zero-stage $ZERO_STAGE \
+    "
+
+CMD=" \
+    Megatron-DeepSpeed/pretrain_gpt.py \
+    --tensor-model-parallel-size $TP_SIZE \
+    --pipeline-model-parallel-size $PP_SIZE \
+    $GPT_ARGS \
+    $OUTPUT_ARGS \
+    --save $CHECKPOINT_PATH \
+    --load $CHECKPOINT_PATH \
+    --train-weighted-split-paths-path $TRAIN_DATA_PATH \
+    --valid-weighted-split-paths-path $VALID_DATA_PATH \
+    --data-impl mmap \
+     $DEEPSPEED_ARGS \
+    "
+
+echo $CMD
+
+echo "START $SLURM_JOBID: $(date)"
+
+# bash launch_srun.sh $CMD
+srun --label launch.sh $CMD
+
+echo "END $SLURM_JOBID: $(date)"

sbatch_4b284bc4perplexity50.sh

@@ -0,0 +1,163 @@
+#!/bin/bash
+#SBATCH --nodes=32
+#SBATCH --ntasks-per-node=1
+#SBATCH --cpus-per-task=40
+#SBATCH --mem=256G
+#SBATCH -p standard-g
+#SBATCH -t 48:00:00
+#SBATCH --gpus-per-node=mi250:8
+#SBATCH --exclusive=user
+#SBATCH --hint=nomultithread
+#SBATCH --account=project_462000119
+#SBATCH -o logs/%j.out
+#SBATCH -e logs/%j.err
+
+
+VARIANT=4b284bc4perplexity50
+
+# if run without sbatch, invoke here
+if [ -z $SLURM_JOB_ID ]; then
+    mkdir -p logs
+    sbatch "$0"
+    exit
+fi
+
+set -euo pipefail
+
+# symlink logs/latest.out and logs/latest.err
+ln -f -s $SLURM_JOB_ID.out logs/latest.out
+ln -f -s $SLURM_JOB_ID.err logs/latest.err
+
+KILL_SWITCH_PATH=kill-switch-$VARIANT
+CHECKPOINT_PATH=checkpoints_$VARIANT
+TENSORBOARD_PATH=tensorboard_$VARIANT
+
+# Data
+VOCAB_FILE="gpt2/vocab.json"
+MERGE_FILE="gpt2/merges.txt"
+DATA_PATH="/scratch/project_462000119/data/c4perplexity/gpt2tok_perplexity_text_document"
+TRAIN_DATA_PATH=trainperplexity50.txt
+# "train: 1.0 0:1 /scratch/project_462000119/data/c4perplexity/gpt2tok_c4_en_pplx_50_text_document"
+VALID_DATA_PATH=val.txt
+# "validation: 1.0 0:1 /scratch/project_462000119/data/c4_validation/gpt2tok_c4validation_rerun_text_document"
+
+PP_SIZE=1
+TP_SIZE=2
+
+MICRO_BATCH_SIZE=2
+GRADIENT_ACCUMULATION_STEPS=1
+WORLD_SIZE=$((SLURM_GPUS_ON_NODE*SLURM_JOB_NUM_NODES))
+GLOBAL_BATCH_SIZE=$((MICRO_BATCH_SIZE*WORLD_SIZE*GRADIENT_ACCUMULATION_STEPS))
+
+# Model parameters
+source model_params.sh
+MODEL_PARAM=("${PARAM_2980M[@]}")
+NHIDDEN=${MODEL_PARAM[0]}
+FFN_HIDDEN_SIZE=${MODEL_PARAM[1]}
+KV_SIZE=${MODEL_PARAM[2]}
+NHEADS=${MODEL_PARAM[3]}
+NLAYERS=${MODEL_PARAM[4]}
+SEQ_LEN=2048
+
+echo "Model parameters: d_model $NHIDDEN ffw_size $FFN_HIDDEN_SIZE kv_size $KV_SIZE n_heads $NHEADS n_layers $NLAYERS"
+
+SAVE_INTERVAL=10000
+
+# Tokens: 84_000_000_000
+# -> Samples: 41_015_625.0
+TRAIN_SAMPLES=41_015_625
+
+OPTIMIZER_ARGS=" \
+    --optimizer adam \
+    --adam-beta1 0.9 \
+    --adam-beta2 0.95 \
+    --adam-eps 1e-8 \
+    --lr 2e-4 \
+    --min-lr 2e-5 \
+    --lr-decay-style cosine \
+    --lr-decay-samples $TRAIN_SAMPLES \
+    --lr-warmup-samples 410_156 \
+    --clip-grad 1.0 \
+    --weight-decay 1e-1 \
+    "
+
+GPT_ARGS=" \
+    --num-layers $NLAYERS \
+    --hidden-size $NHIDDEN \
+    --num-attention-heads $NHEADS \
+    --kv-channels $KV_SIZE \
+    --ffn-hidden-size $FFN_HIDDEN_SIZE \
+    --seq-length $SEQ_LEN \
+    --max-position-embeddings $SEQ_LEN \
+    --micro-batch-size $MICRO_BATCH_SIZE \
+    --global-batch-size $GLOBAL_BATCH_SIZE \
+    --train-samples $TRAIN_SAMPLES \
+    --vocab-file $VOCAB_FILE \
+    --merge-file $MERGE_FILE \
+    --clip-grad 1.0 \
+    --kill-switch-path $KILL_SWITCH_PATH \
+    --bf16 \
+    $OPTIMIZER_ARGS \
+    "
+
+OUTPUT_ARGS=" \
+    --log-interval 10 \
+    --save-interval $SAVE_INTERVAL \
+    --eval-interval 1000 \
+    --eval-iters 100 \
+    --tensorboard-dir $TENSORBOARD_PATH \
+    --tensorboard-queue-size 5 \
+    --log-timers-to-tensorboard \
+    --log-batch-size-to-tensorboard \
+    --log-validation-ppl-to-tensorboard \
+    "
+
+ZERO_STAGE=0
+
+mkdir -p ds_configs
+DS_CONFIG_PATH="ds_configs/$SLURM_JOB_ID.json"
+
+cat <<EOF > $DS_CONFIG_PATH
+{
+    "train_micro_batch_size_per_gpu": $MICRO_BATCH_SIZE,
+    "train_batch_size": $GLOBAL_BATCH_SIZE,
+    "gradient_clipping": 1.0,
+    "zero_optimization": {
+        "stage": $ZERO_STAGE
+    },
+    "bf16": {
+        "enabled": true
+    },
+    "steps_per_print": 2000,
+    "wall_clock_breakdown": false
+}
+EOF
+
+DEEPSPEED_ARGS=" \
+    --deepspeed \
+    --deepspeed_config $DS_CONFIG_PATH \
+    --zero-stage $ZERO_STAGE \
+    "
+
+CMD=" \
+    Megatron-DeepSpeed/pretrain_gpt.py \
+    --tensor-model-parallel-size $TP_SIZE \
+    --pipeline-model-parallel-size $PP_SIZE \
+    $GPT_ARGS \
+    $OUTPUT_ARGS \
+    --save $CHECKPOINT_PATH \
+    --load $CHECKPOINT_PATH \
+    --train-weighted-split-paths-path $TRAIN_DATA_PATH \
+    --valid-weighted-split-paths-path $VALID_DATA_PATH \
+    --data-impl mmap \
+     $DEEPSPEED_ARGS \
+    "
+
+echo $CMD
+
+echo "START $SLURM_JOBID: $(date)"
+
+# bash launch_srun.sh $CMD
+srun --label launch.sh $CMD
+
+echo "END $SLURM_JOBID: $(date)"

tensorboard/tensorboard_4b284bc4perplexity25/events.out.tfevents.1675983088.nid005401.17305.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d8ad51d2662fc24895413e38dcd9c514dfe6b0fc3ecf474882ce7c36ccde171
+size 40

tensorboard/tensorboard_4b284bc4perplexity25/events.out.tfevents.1675983591.nid005695.45960.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba73df888c7bec444c353e96219d2c90b6468f7e48cde329cf5961c7804ca406
+size 40

tensorboard/tensorboard_4b284bc4perplexity25/events.out.tfevents.1675984102.nid005695.54181.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c6c6d0f8ceb3fe6aa0d763830030d10e7ed3fc140d42c4ff4833720ff762d93
+size 143659632

tensorboard/tensorboard_4b284bc4perplexity25/events.out.tfevents.1676144755.nid005695.111249.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d37e79432b7f773b71ab04e8b788cd5ea8a709903f18d2534ab7408bdc740d56
+size 40

tensorboard/tensorboard_4b284bc4perplexity25/events.out.tfevents.1676145256.nid005695.119215.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5fa5ba5d3ec5ab32f5eb2e60176ab4bfeb7acb60b5c7285d007f4e978896285
+size 40

tensorboard/tensorboard_4b284bc4perplexity50/events.out.tfevents.1675975712.nid005946.123671.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dbc8cbe3c1e343f0646a8899ed0ba1bab44da4d75264e8f0b3f583b15f459be7
+size 40

tensorboard/tensorboard_4b284bc4perplexity50/events.out.tfevents.1675976219.nid006070.128633.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b10d68963f8bf0785210f9b4a8d2ded741e241c2e388f2aca7eb7b55ea20063
+size 143659632

tensorboard/tensorboard_4b284bc4perplexity50/events.out.tfevents.1676137082.nid006070.59097.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:456b9314a080fce1ffe7b780981b8391f90df85592eb21e2aec34f5fb101498d
+size 40

tensorboard/tensorboard_4b284bc4perplexity50/events.out.tfevents.1676137586.nid006070.64809.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6dc614a7079a07d00c2ee252a8faa5acabca95d514be7f04a4fcf973003387f
+size 40

tensorboard/tensorboard_4b284bc4perplexity50/events.out.tfevents.1676138087.nid006070.73224.0

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a61a6b497e46f1f6e28fb065b247136949b8d224bfc622e5daa828102a245f50
+size 40