feat: script to run llama8b training (#14239)

examples/mlperf/training_submission_v6.0/tinycorp/benchmarks/llama8b/implementations/tinybox_8xMI350X/dev_run.sh

@@ -0,0 +1,23 @@
+#!/usr/bin/env bash
+
+export PYTHONPATH="." AMD=1
+export IGNORE_OOB=1
+export REWRITE_STACK_LIMIT=5000000 HCQDEV_WAIT_TIMEOUT_MS=240000
+
+export DEBUG=${DEBUG:-0}
+export FLASH_ATTENTION=1
+
+export DEFAULT_FLOAT="bfloat16" OPTIM_DTYPE="bfloat16"
+export DP=8 BS=8 EVAL_BS=8
+
+export MODEL="llama3"
+export BASEDIR="/raid/datasets/c4-8b/"
+export SMALL=1 LLAMA3_SIZE="8B"
+export EVAL_TARGET=3.3 EVAL_FREQ=12288
+export LR="1e-3" END_LR="1e-4" WARMUP_STEPS=1024 MAX_STEPS=1200000
+
+export SEED=5760
+
+export JITBEAM=3 BEAM_UOPS_MAX=6000 BEAM_UPCAST_MAX=256 BEAM_LOCAL_MAX=1024 BEAM_MIN_PROGRESS=5
+
+python3 examples/mlperf/model_train.py