[RecML] fix DLRM benchmark script issue and README instruction.

1. Reduce the batch size from 135168 to 4224. The existing batch size is too large as compared to the default dataset that's used, which causes parsing issues.
2. Fix the script file path typo, and make scripts executable before running it
3. Downgrade protobuf version to avoid incompatible error between python and protobuf library.

PiperOrigin-RevId: 775976394

Author: RecML authors

recml/inference/benchmarks/DLRM_DCNv2/ckpt_load_and_eval.sh

@@ -6,14 +6,14 @@ export XLA_FLAGS=
 
 export TPU_NAME=<TPU_NAME>
 export LEARNING_RATE=0.0034
-export BATCH_SIZE=135168
+export BATCH_SIZE=4224
 export EMBEDDING_SIZE=128
 export MODEL_DIR=/tmp/
 export FILE_PATTERN=gs://qinyiyan-vm/mlperf-dataset/criteo_merge_balanced_4224/train-*
 export NUM_STEPS=28000
 export CHECKPOINT_INTERVAL=1500
 export EVAL_INTERVAL=1500
-export EVAL_FILE_PATTER=gs://qinyiyan-vm/mlperf-dataset/criteo_merge_balanced_4224/eval-*
+export EVAL_FILE_PATTERN=gs://qinyiyan-vm/mlperf-dataset/criteo_merge_balanced_4224/eval-*
 export EVAL_STEPS=660
 export MODE=eval
 export EMBEDDING_THRESHOLD=21000
@@ -23,7 +23,6 @@ export RESTORE_CHECKPOINT=true
 
 
 python recml/inference/models/jax/DLRM_DCNv2/dlrm_main.py \
-
 --learning_rate=${LEARNING_RATE} \
 --batch_size=${BATCH_SIZE} \
 --embedding_size=${EMBEDDING_SIZE} \

recml/inference/benchmarks/DLRM_DCNv2/train_and_checkpoint.sh

@@ -6,22 +6,21 @@ export XLA_FLAGS=
 
 export TPU_NAME=<TPU_NAME>
 export LEARNING_RATE=0.0034
-export BATCH_SIZE=135168
+export BATCH_SIZE=4224
 export EMBEDDING_SIZE=128
 export MODEL_DIR=/tmp/
 export FILE_PATTERN=gs://qinyiyan-vm/mlperf-dataset/criteo_merge_balanced_4224/train-*
 export NUM_STEPS=28000
 export CHECKPOINT_INTERVAL=1500
 export EVAL_INTERVAL=1500
-export EVAL_FILE_PATTER=gs://qinyiyan-vm/mlperf-dataset/criteo_merge_balanced_4224/eval-*
+export EVAL_FILE_PATTERN=gs://qinyiyan-vm/mlperf-dataset/criteo_merge_balanced_4224/eval-*
 export EVAL_STEPS=660
 export MODE=train
 export EMBEDDING_THRESHOLD=21000
 export LOGGING_INTERVAL=1500
 export RESTORE_CHECKPOINT=true
 
 python recml/inference/models/jax/DLRM_DCNv2/dlrm_main.py \
-
 --learning_rate=${LEARNING_RATE} \
 --batch_size=${BATCH_SIZE} \
 --embedding_size=${EMBEDDING_SIZE} \

recml/inference/benchmarks/README.md

@@ -54,10 +54,10 @@ gcloud alpha compute tpus tpu-vm ssh ${TPU_NAME} --project ${PROJECT} --zone ${Z
 gcloud alpha compute tpus tpu-vm ssh ${TPU_NAME} --project ${PROJECT} --zone ${ZONE} --worker=all  --command="pip install -U tensorflow  dm-tree flax google-metrax"
 ```
 
-#### Run workload
+#### Make script executable & Run workload
 
 Note: Please update the MODEL_NAME & TASK_NAME before running the below command
 
 ```
-gcloud alpha compute tpus tpu-vm ssh ${TPU_NAME} --project ${PROJECT} --zone ${ZONE} --worker=all  --command="TPU_NAME=${TPU_NAME} ./inference/benchmarks/<MODEL_NAME>/<TASK_NAME>"
+gcloud alpha compute tpus tpu-vm ssh ${TPU_NAME} --project ${PROJECT} --zone ${ZONE} --worker=all  --command="cd RecML && chmod +x ./recml/inference/benchmarks/<MODEL_NAME>/<TASK_NAME> && TPU_NAME=${TPU_NAME} ./recml/inference/benchmarks/<MODEL_NAME>/<TASK_NAME>"
 ```

requirements.txt

@@ -63,7 +63,7 @@ platformdirs==4.3.7
 pluggy==1.5.0
 pre-commit==4.2.0
 promise==2.3
-protobuf==5.29.4
+protobuf==4.21.12
 psutil==7.0.0
 pyarrow==19.0.1
 pygments==2.19.1