diff --git a/MLExamples/PyTorch_Profiling/README_ROCM_NIGHTLY_TESTING.md b/MLExamples/PyTorch_Profiling/README_ROCM_NIGHTLY_TESTING.md
new file mode 100644
index 00000000..e9a66f75
--- /dev/null
+++ b/MLExamples/PyTorch_Profiling/README_ROCM_NIGHTLY_TESTING.md
@@ -0,0 +1,129 @@
+# Testing a ROCm Nightly Build with PyTorch (CIFAR-100 train + profile)
+
+This guide walks you through testing a **nightly ROCm build** together with
+PyTorch on an AMD GPU. You will:
+
+1. Build a self-contained Python virtual environment with nightly ROCm, PyTorch,
+   and the ROCm profilers.
+2. Create a `setup_rocm.sh` script that activates that environment.
+3. Run the CIFAR-100 training workload through each profiling tool using the
+   provided SLURM scripts, and check that everything works end to end.
+
+The workload itself is the same `train_cifar_100.py` used throughout this
+directory — a small vision model trained on CIFAR-100. It is intentionally
+short so that a nightly build can be validated quickly.
+
+> This guide targets an AMD MI300A GPU (`gfx942`) on a SLURM cluster. For a
+> different GPU, change the architecture (`device-gfx942` / `--arch`) and the
+> SLURM `--partition` in the scripts accordingly.
+
+---
+
+## Step 1 — Build the virtual environment
+
+Follow [`ROCM_PYTORCH_PIP_VENV_SETUP.md`](./ROCM_PYTORCH_PIP_VENV_SETUP.md) to
+create the `rocm-pytorch-pip` venv. In short, it:
+
+- creates a venv at `~/venvs/rocm-pytorch-pip`,
+- installs nightly ROCm + PyTorch + profilers from the multi-arch nightly index,
+- installs `transformers` (required by the training script),
+- runs `rocm-sdk init` to extract the development headers and device code.
+
+To test a **specific nightly**, set the ROCm version pin in that guide, e.g.:
+
+```bash
+ROCM_VERSION=7.15.0a20260721
+```
+
+Change this value to the nightly date you want to validate.
+
+## Step 2 — Verify `setup_rocm.sh`
+
+The SLURM scripts in each sub-directory activate the environment by sourcing
+`../setup_rocm.sh` — i.e. the `setup_rocm.sh` shipped in **this**
+(`PyTorch_Profiling/`) directory. It is already provided; just verify (and edit
+if needed) that its `VENV` points at the venv you built in Step 1. Its full
+contents, and a GPU-node sanity check, are covered in steps 5-7 of
+[`ROCM_PYTORCH_PIP_VENV_SETUP.md`](./ROCM_PYTORCH_PIP_VENV_SETUP.md).
+
+Once verified, a quick check that the nightly build runs GPU kernels through
+PyTorch:
+
+```bash
+source setup_rocm.sh
+srun -n1 --gpus=1 python3 -c "import torch; print('torch', torch.__version__); \
+x = torch.ones(4, device='cuda:0'); print('device ok:', (x+1).sum().item())"
+```
+
+If you see `device ok:`, you are ready to profile.
+
+## Step 3 — Run the SLURM scripts
+
+Each sub-directory contains a single-process SLURM script that sources
+`../setup_rocm.sh`, pre-downloads the dataset if needed, and runs the workload
+under one tool. All of them use a single GPU and a short run
+(`--batch-size 32 --max-steps 5`) so a nightly can be checked quickly.
+
+**Submit each script from its own directory** (the scripts use
+`SLURM_SUBMIT_DIR` to locate themselves):
+
+| Tool | Directory | Script | What it produces |
+|------|-----------|--------|------------------|
+| None (baseline) | `no-profiling/` | `slurm_single_process_noprofile.sh` | Plain training run — confirms the workload runs without a profiler. |
+| ROCm Compute Profiler (profile) | `rocm-compute-profiler/` | `slurm_single_process_profile.sh` | Hardware-counter profile under `workloads/`. |
+| ROCm Compute Profiler (analyze) | `rocm-compute-profiler/` | `slurm_single_process_analyze.sh` | Analysis report from the profile above. |
+| RocProfiler (kernels) | `rocprofv3/` | `slurm_single_process_kernels.sh` | Kernel stats + trace CSVs under `single_process/`. |
+| RocProfiler (traces) | `rocprofv3/` | `slurm_single_process_traces.sh` | System timeline trace (`.pftrace`) under `single_process/`. |
+| ROCm Systems Profiler | `rocm-systems-profiler/` | `slurm_single_process.sh` | Sampling profile + trace under `rocprofsys-python3-output/`. |
+| Roofline Extractor | `roofline-extractor/` | `slurm_single_process.sh` | Per-kernel roofline analysis + interactive HTML plot under `output/`. |
+
+Example (baseline sanity check first, then a profiler):
+
+```bash
+cd no-profiling
+sbatch slurm_single_process_noprofile.sh
+
+cd ../rocprofv3
+sbatch slurm_single_process_kernels.sh
+```
+
+Check job status and output:
+
+```bash
+squeue --me
+# stdout/stderr land in the submit directory as <name>_<jobid>.out / .err
+```
+
+### Analyzing the results
+
+> **Note — running `rocprof-compute analyze`:** run it only **after** its
+> profile job has finished (the counter database must exist). Because it needs
+> `numpy==1.26.4` (vs the shared venv's `numpy>=2.0`), the analysis script uses
+> its own isolated venv (`~/venvs/rocprof-compute-analyze`) and never touches
+> the shared venv.
+
+- **ROCm Compute Profiler:** submit the companion analysis job from
+  `rocm-compute-profiler/` (it locates the workload and runs the analysis for you):
+
+```bash
+cd rocm-compute-profiler
+sbatch slurm_single_process_analyze.sh
+```
+
+  Or run it by hand: `rocprof-compute analyze -p rocm-compute-profiler/workloads/cifar_100_single_proc/<subdir>`
+- **RocProfiler:** open the CSVs (kernels) or load the `.pftrace` in
+  [ui.perfetto.dev](https://ui.perfetto.dev) (traces).
+- **ROCm Systems Profiler:** load the `perfetto-trace-*.proto` from
+  `rocprofsys-python3-output/<timestamp>/` in [ui.perfetto.dev](https://ui.perfetto.dev).
+- **Roofline Extractor:** open the generated `.html` in `roofline-extractor/output/`.
+
+---
+
+## Notes
+
+- If a job fails to start, check the SLURM `--partition` and time limits in the
+  script headers match your cluster.
+- The scripts derive a per-job rendezvous port from the SLURM job ID, so
+  multiple jobs can share a node without port collisions.
+- To validate a different nightly, rebuild the venv (Step 1) with a new
+  `ROCM_VERSION` and re-run the scripts.
diff --git a/MLExamples/PyTorch_Profiling/ROCM_PYTORCH_PIP_VENV_SETUP.md b/MLExamples/PyTorch_Profiling/ROCM_PYTORCH_PIP_VENV_SETUP.md
new file mode 100644
index 00000000..1d023d59
--- /dev/null
+++ b/MLExamples/PyTorch_Profiling/ROCM_PYTORCH_PIP_VENV_SETUP.md
@@ -0,0 +1,123 @@
+# Creating the `rocm-pytorch-pip` venv (nightly ROCm + PyTorch, MI300A / gfx942)
+
+This guide walks you through building a Python virtual environment with ROCm,
+PyTorch, and the ROCm profiling tools, so you can train and profile a model on
+an AMD GPU. Follow the steps in order.
+
+Everything installs from pip into one self-contained venv, so PyTorch and the
+profilers use the same ROCm. Each step below is a single command block you can
+copy and run.
+
+> This guide targets an AMD MI300A GPU (`gfx942`). If you have a different GPU,
+> change `device-gfx942` to your architecture.
+
+---
+
+## 1. Create and activate the venv
+
+First choose where the venv should live. Set `VENV_BASE` to the directory that
+will hold the `venvs` folder (defaults to your home directory). Keep this shell
+open for the remaining steps, which reuse the variable.
+
+```bash
+VENV_BASE=~
+mkdir -p "${VENV_BASE}/venvs"
+python -m venv "${VENV_BASE}/venvs/rocm-pytorch-pip"
+source "${VENV_BASE}/venvs/rocm-pytorch-pip/bin/activate"
+```
+
+## 2. Install ROCm + PyTorch from the nightly multi-arch index
+
+```bash
+# Pin the nightly ROCm version once and reuse it everywhere below.
+ROCM_VERSION=7.15.0a20260721
+
+pip install --index-url https://rocm.nightlies.amd.com/whl-multi-arch/ \
+    "rocm[profiler,devel,libraries,device-gfx942]==${ROCM_VERSION}" \
+    "torch[device-gfx942]" \
+    "torchvision[device-gfx942]"
+```
+
+The `rocm[...]` extras pull in the pieces this workflow needs:
+- `profiler`  — the ROCm profilers: `rocprof-compute`, `rocprofv3`, and
+  `rocprof-sys` (bundled `_rocm_profiler`)
+- `devel`     — development package (headers/device code, extracted in step 4)
+- `libraries` — math libraries (hipBLAS, rocBLAS, ...)
+- `device-gfx942` — the GPU-arch kernels for MI300A
+
+## 3. Install `transformers` (required by the training script)
+
+```bash
+pip install transformers
+```
+
+The training script builds its models with `transformers`, so this package is
+required.
+
+## 4. Extract development headers and device code
+
+```bash
+"${VENV_BASE}/venvs/rocm-pytorch-pip/bin/rocm-sdk" init
+```
+
+`rocm-sdk init` unpacks the `devel` payload (headers, LLVM device bitcode) into
+`_rocm_sdk_devel/` inside the venv — the device bitcode HIP needs to run kernels,
+and the paths `setup_rocm.sh` points at next.
+
+---
+
+## 5. Verify `setup_rocm.sh`
+
+The repo already ships `setup_rocm.sh` in `MLExamples/PyTorch_Profiling/` (the
+SLURM scripts source it as `../setup_rocm.sh`). It activates the venv and points
+the ROCm environment at the extracted `_rocm_sdk_devel` tree. It defaults
+`VENV_BASE` to your home directory; if you used a different `VENV_BASE` in
+step 1, export it before sourcing (or edit the default here). Its contents are:
+
+```bash
+#!/usr/bin/env bash
+# Source this to activate the ROCm venv and set ROCm env vars:
+#   source setup_rocm.sh
+VENV_BASE="${VENV_BASE:-$HOME}"
+VENV="$VENV_BASE/venvs/rocm-pytorch-pip"
+source "$VENV/bin/activate"
+DEVEL="$(python3 -c 'import site; print(site.getsitepackages()[0])')/_rocm_sdk_devel"
+export ROCM_PATH="$DEVEL"
+export HIP_PATH="$DEVEL"
+export HIP_DEVICE_LIB_PATH="$DEVEL/lib/llvm/amdgcn/bitcode"
+export PATH="$DEVEL/bin:$PATH"
+export LD_LIBRARY_PATH="$DEVEL/lib:$DEVEL/lib/rocm_sysdeps/lib:$LD_LIBRARY_PATH"
+echo "ROCm venv active: $VENV"
+```
+
+## 6. Re-source to pick up the ROCm env vars
+
+If the venv is already active from step 1, deactivate and source the script so
+the `ROCM_PATH` / `LD_LIBRARY_PATH` exports take effect (run from
+`MLExamples/PyTorch_Profiling/`):
+
+```bash
+deactivate
+source setup_rocm.sh
+```
+
+---
+
+## 7. Verify (on a GPU node)
+
+```bash
+source setup_rocm.sh
+srun -n1 --gpus=1 python3 -c "import torch; print('torch', torch.__version__); \
+x = torch.ones(4, device='cuda:0'); print('device ok:', (x+1).sum().item())"
+```
+
+Expected output resembles:
+
+```
+ROCm venv active: /.../rocm-pytorch-pip
+torch 2.12.0+rocm7.15.0a20260721
+device ok: 8.0
+```
+
+If you see `device ok:`, the GPU is working and your environment is ready to use.
+From now on, just run `source setup_rocm.sh` in any new shell to activate it.
diff --git a/MLExamples/PyTorch_Profiling/no-profiling/slurm_single_process_noprofile.sh b/MLExamples/PyTorch_Profiling/no-profiling/slurm_single_process_noprofile.sh
new file mode 100755
index 00000000..11fef41b
--- /dev/null
+++ b/MLExamples/PyTorch_Profiling/no-profiling/slurm_single_process_noprofile.sh
@@ -0,0 +1,59 @@
+#!/bin/bash
+#SBATCH --job-name=rpc-single-noprofile
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=8
+#SBATCH --gpus=1
+#SBATCH --time=00:30:00
+#SBATCH --output=rpc_single_process_noprofile_%j.out
+#SBATCH --error=rpc_single_process_noprofile_%j.err
+
+# ---------------------------------------------------------------------------
+# Run the single-process CIFAR-100 workload directly with python3 (no profiler).
+# Sources ../setup_rocm.sh to activate the ROCm PyTorch venv.
+# ---------------------------------------------------------------------------
+
+set -e
+
+# Resolve this script's dir; under sbatch prefer SLURM_SUBMIT_DIR. Submit from
+# MLExamples/PyTorch_Profiling/no-profiling/.
+if [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    SCRIPT_DIR="$(dirname "$(readlink -fm "$0")")"
+fi
+PROFILER_TOP_DIR="$(dirname "${SCRIPT_DIR}")"
+echo "SCRIPT_DIR=${SCRIPT_DIR}"
+echo "PROFILER_TOP_DIR=${PROFILER_TOP_DIR}"
+
+# ../setup_rocm.sh activates the ROCm PyTorch venv and exports ROCm env vars.
+source ${PROFILER_TOP_DIR}/setup_rocm.sh
+
+# Distributed bootstrap variables expected by train_cifar_100.py (single rank).
+export NPROCS=1
+export MASTER_ADDR=${MASTER_ADDR:-$(hostname)}
+# Derive a per-job port so concurrent jobs don't collide.
+export MASTER_PORT=${MASTER_PORT:-$((20000 + SLURM_JOB_ID % 20000))}
+
+# Make sure the dataset is present.
+if [ ! -d ${PROFILER_TOP_DIR}/data/cifar-100-python ]; then
+    python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+        --data-path ${PROFILER_TOP_DIR}/data --download-only
+fi
+
+cd ${SCRIPT_DIR}
+
+# Run the workload directly (no profiler).
+echo
+echo "==================================================================="
+echo "Running (no profiler): python3 train_cifar_100.py"
+echo "==================================================================="
+srun -n 1 --gpus=1 --cpus-per-task=8 \
+    python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+        --batch-size 32 --max-steps 5 \
+        --data-path ${PROFILER_TOP_DIR}/data
+
+echo
+echo "==================================================================="
+echo "Workload run complete."
+echo "==================================================================="
diff --git a/MLExamples/PyTorch_Profiling/rocm-compute-profiler/slurm_single_process_analyze.sh b/MLExamples/PyTorch_Profiling/rocm-compute-profiler/slurm_single_process_analyze.sh
new file mode 100755
index 00000000..e587d8ae
--- /dev/null
+++ b/MLExamples/PyTorch_Profiling/rocm-compute-profiler/slurm_single_process_analyze.sh
@@ -0,0 +1,139 @@
+#!/bin/bash
+#SBATCH --job-name=rpc-single-analyze
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=8
+#SBATCH --time=00:15:00
+#SBATCH --output=rpc_single_process_analyze_%j.out
+#SBATCH --error=rpc_single_process_analyze_%j.err
+
+# ---------------------------------------------------------------------------
+# Analyze the profiled workload with rocprof-compute analyze.
+#
+# rocprof-compute analyze needs numpy 1.26.x, which conflicts with the shared
+# venv's numpy 2.x. So we use a separate venv holding only rocprof-compute's
+# pinned requirements, while reusing the tool bundled in the shared venv (the
+# shared venv is never modified). CPU-only; run after the profile job finishes.
+# ---------------------------------------------------------------------------
+
+set -e
+
+# Resolve this script's dir; under sbatch prefer SLURM_SUBMIT_DIR. Submit from
+# MLExamples/PyTorch_Profiling/rocm-compute-profiler/.
+if [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    SCRIPT_DIR="$(dirname "$(readlink -fm "$0")")"
+fi
+PROFILER_TOP_DIR="$(dirname "${SCRIPT_DIR}")"
+echo "SCRIPT_DIR=${SCRIPT_DIR}"
+echo "PROFILER_TOP_DIR=${PROFILER_TOP_DIR}"
+
+# Locate the shared ROCm venv WITHOUT activating it; we only need the ROCm
+# install inside it (bundled rocprof-compute launcher + ROCm libraries).
+MAIN_VENV="${HOME}/venvs/rocm-pytorch-pip"
+if [[ ! -x "${MAIN_VENV}/bin/python3" ]]; then
+    echo "ERROR: shared ROCm venv not found at ${MAIN_VENV}" >&2
+    exit 1
+fi
+# site-packages of the shared venv (avoids hard-coding the python3.x version).
+MAIN_SP="$("${MAIN_VENV}/bin/python3" -c 'import site; print(site.getsitepackages()[0])')"
+ROCM_PROFILER="${MAIN_SP}/_rocm_profiler"
+ROCM_CORE="${MAIN_SP}/_rocm_sdk_core"
+ROCM_DEVEL="${MAIN_SP}/_rocm_sdk_devel"
+COMPUTE_LIBEXEC="${ROCM_PROFILER}/libexec/rocprofiler-compute"
+REQ_FILE="${COMPUTE_LIBEXEC}/requirements.txt"
+RPC="${ROCM_PROFILER}/bin/rocprof-compute"
+
+for p in "${RPC}" "${REQ_FILE}" "${ROCM_CORE}" "${ROCM_DEVEL}"; do
+    if [[ ! -e "${p}" ]]; then
+        echo "ERROR: expected ROCm component not found: ${p}" >&2
+        exit 1
+    fi
+done
+
+# Create/refresh the dedicated analysis venv with rocprof-compute's pinned
+# requirements (numpy 1.26.x). Reinstalled only when requirements.txt changes.
+ANALYZE_VENV="${HOME}/venvs/rocprof-compute-analyze"
+if [[ ! -x "${ANALYZE_VENV}/bin/python3" ]]; then
+    echo "Creating analysis venv at ${ANALYZE_VENV}"
+    # A venv stays isolated from its creator's site-packages, so using the
+    # shared venv's python3 keeps this independent while avoiding system python3.
+    "${MAIN_VENV}/bin/python3" -m venv "${ANALYZE_VENV}"
+fi
+source "${ANALYZE_VENV}/bin/activate"
+
+STAMP="${ANALYZE_VENV}/.rocprof_compute_reqs_installed"
+if [[ ! -f "${STAMP}" || "${REQ_FILE}" -nt "${STAMP}" ]]; then
+    echo "Installing rocprof-compute requirements into ${ANALYZE_VENV}"
+    python3 -m pip install --upgrade pip
+    python3 -m pip install -r "${REQ_FILE}"
+    cp "${REQ_FILE}" "${STAMP}"
+else
+    echo "Analysis venv already satisfies ${REQ_FILE}"
+fi
+
+# Point the isolated venv at the ROCm install in the shared venv (bundled
+# rocprof-compute + libs). ROCm bin goes LAST on PATH so it can't shadow the
+# analysis venv's python3, keeping numpy 1.26.x in effect.
+export PYTHONPATH="${COMPUTE_LIBEXEC}:${PYTHONPATH}"
+export LD_LIBRARY_PATH="${ROCM_CORE}/lib:${ROCM_CORE}/lib/rocm_sysdeps/lib:${ROCM_DEVEL}/lib:${ROCM_DEVEL}/lib/rocm_sysdeps/lib:${LD_LIBRARY_PATH}"
+export ROCM_PATH="${ROCM_DEVEL}"
+export HIP_PATH="${ROCM_DEVEL}"
+export PATH="${PATH}:${ROCM_DEVEL}/bin"
+
+echo "Analysis python : $(which python3)  (numpy $(python3 -c 'import numpy; print(numpy.__version__)'))"
+"${RPC}" --version
+
+# Locate the workload written by slurm_single_process_profile.sh (must match
+# WORKLOAD_NAME / WORK_ROOT there).
+WORKLOAD_NAME=cifar_100_single_proc
+WORK_ROOT=${SCRIPT_DIR}/workloads
+WORK_DIR=${WORK_ROOT}/${WORKLOAD_NAME}
+
+if [[ ! -d "${WORK_DIR}" ]]; then
+    echo "ERROR: workload directory not found: ${WORK_DIR}" >&2
+    echo "Run slurm_single_process_profile.sh first." >&2
+    exit 1
+fi
+
+# The profile is written into a single subdirectory named after the architecture
+# (e.g. `MI300A_*`) or numerically (`0`, `1`, ...). Pick the first one.
+ARCH_DIR=$(find ${WORK_DIR} -mindepth 1 -maxdepth 1 -type d | sort | head -1)
+if [[ -z "${ARCH_DIR}" ]]; then
+    echo "ERROR: no workload subdirectory found under ${WORK_DIR}" >&2
+    exit 1
+fi
+
+# ---------------------------------------------------------------------------
+# Analyze.
+# ---------------------------------------------------------------------------
+# Optional: list per-kernel stats and dispatch IDs. Uncomment only if you want
+# to narrow the analysis to a specific kernel/dispatch (see the NOTE at the
+# bottom of this file).
+# echo
+# echo "==================================================================="
+# echo "rocprof-compute analyze --list-stats -p ${ARCH_DIR}"
+# echo "==================================================================="
+# STATS_FILE=${SCRIPT_DIR}/stats_${SLURM_JOB_ID}.txt
+# "${RPC}" analyze --list-stats -p ${ARCH_DIR} >& ${STATS_FILE}
+# echo "Stats and dispatch IDs written to ${STATS_FILE}"
+
+echo
+echo "==================================================================="
+echo "rocprof-compute analyze -p ${ARCH_DIR}"
+echo "==================================================================="
+ANALYSIS_FILE=${SCRIPT_DIR}/analysis_${SLURM_JOB_ID}.txt
+"${RPC}" analyze -p ${ARCH_DIR} > ${ANALYSIS_FILE} 2>&1
+echo "Analysis written to ${ANALYSIS_FILE}"
+
+echo
+echo "-------------------------------------------------------------------"
+echo "NOTE: The analysis above aggregates all kernels. To focus on a"
+echo "specific kernel, uncomment the --list-stats block above to list the"
+echo "kernel names and dispatch IDs, then re-run rocprof-compute analyze"
+echo "narrowing to that kernel with --kernel <kernel-id>, or to a specific"
+echo "invocation with --dispatch <dispatch-id> (you may pass either, or both):"
+echo "  ${RPC} analyze -p ${ARCH_DIR} --kernel <kernel-id>"
+echo "  ${RPC} analyze -p ${ARCH_DIR} --dispatch <dispatch-id>"
+echo "-------------------------------------------------------------------"
diff --git a/MLExamples/PyTorch_Profiling/rocm-compute-profiler/slurm_single_process_profile.sh b/MLExamples/PyTorch_Profiling/rocm-compute-profiler/slurm_single_process_profile.sh
new file mode 100755
index 00000000..be3974ed
--- /dev/null
+++ b/MLExamples/PyTorch_Profiling/rocm-compute-profiler/slurm_single_process_profile.sh
@@ -0,0 +1,84 @@
+#!/bin/bash
+#SBATCH --job-name=rpc-single-profile
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=8
+#SBATCH --gpus=1
+#SBATCH --time=02:00:00
+#SBATCH --output=rpc_single_process_profile_%j.out
+#SBATCH --error=rpc_single_process_profile_%j.err
+
+# ---------------------------------------------------------------------------
+# Profile the single-process CIFAR-100 workload with rocprof-compute profile.
+# Sources ../setup_rocm.sh to activate the ROCm PyTorch venv.
+# ---------------------------------------------------------------------------
+
+set -e
+
+# Resolve this script's dir; under sbatch prefer SLURM_SUBMIT_DIR. Submit from
+# MLExamples/PyTorch_Profiling/rocm-compute-profiler/.
+if [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    SCRIPT_DIR="$(dirname "$(readlink -fm "$0")")"
+fi
+PROFILER_TOP_DIR="$(dirname "${SCRIPT_DIR}")"
+echo "SCRIPT_DIR=${SCRIPT_DIR}"
+echo "PROFILER_TOP_DIR=${PROFILER_TOP_DIR}"
+
+# ../setup_rocm.sh activates the ROCm PyTorch venv and exports ROCm env vars.
+source ${PROFILER_TOP_DIR}/setup_rocm.sh
+rocprof-compute --version
+
+# Distributed bootstrap vars for train_cifar_100.py (single rank). Derive a
+# per-job port so concurrent jobs don't collide.
+export NPROCS=1
+export MASTER_ADDR=${MASTER_ADDR:-$(hostname)}
+export MASTER_PORT=${MASTER_PORT:-$((20000 + SLURM_JOB_ID % 20000))}
+
+# Pre-download the dataset so the profiled step doesn't spend time on it.
+if [ ! -d ${PROFILER_TOP_DIR}/data/cifar-100-python ]; then
+    python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+        --data-path ${PROFILER_TOP_DIR}/data --download-only
+fi
+
+WORKLOAD_NAME=cifar_100_single_proc
+WORK_ROOT=${SCRIPT_DIR}/workloads
+WORK_DIR=${WORK_ROOT}/${WORKLOAD_NAME}
+
+mkdir -p ${WORK_ROOT}
+rm -rf ${WORK_DIR}
+cd ${SCRIPT_DIR}
+
+# Profile. Notes:
+#  * --no-roof skips roofline capture (not needed here).
+#  * rocprof-compute replays the app multiple times to cover all counters, so
+#    keep the workload short.
+echo
+echo "==================================================================="
+echo "rocprof-compute profile -- python3 train_cifar_100.py"
+echo "==================================================================="
+srun -n 1 --gpus=1 --cpus-per-task=8 \
+    rocprof-compute profile \
+        --name ${WORKLOAD_NAME} \
+        --no-roof \
+        -- \
+        python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+            --batch-size 32 --max-steps 5 \
+            --data-path ${PROFILER_TOP_DIR}/data
+
+# The profile subdir is named by arch (e.g. MI300A_*) or numerically; pick the first.
+ARCH_DIR=$(find ${WORK_DIR} -mindepth 1 -maxdepth 1 -type d | sort | head -1)
+if [[ -z "${ARCH_DIR}" ]]; then
+    echo "ERROR: no workload subdirectory found under ${WORK_DIR}" >&2
+    exit 1
+fi
+
+echo
+echo "==================================================================="
+echo "Profile complete."
+echo "Workload directory: ${ARCH_DIR}"
+echo
+echo "Analyze with, e.g.:"
+echo "  rocprof-compute analyze -p ${ARCH_DIR}"
+echo "==================================================================="
diff --git a/MLExamples/PyTorch_Profiling/rocm-systems-profiler/slurm_single_process.sh b/MLExamples/PyTorch_Profiling/rocm-systems-profiler/slurm_single_process.sh
new file mode 100755
index 00000000..c4e73749
--- /dev/null
+++ b/MLExamples/PyTorch_Profiling/rocm-systems-profiler/slurm_single_process.sh
@@ -0,0 +1,65 @@
+#!/bin/bash
+#SBATCH --job-name=rps-single-run
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=8
+#SBATCH --gpus=1
+#SBATCH --time=00:30:00
+#SBATCH --output=rps_single_process_%j.out
+#SBATCH --error=rps_single_process_%j.err
+
+# ---------------------------------------------------------------------------
+# Profile the single-process CIFAR-100 workload with the ROCm Systems Profiler
+# (rocprof-sys-run). Sources ../setup_rocm.sh to activate the ROCm venv.
+# ---------------------------------------------------------------------------
+
+set -e
+
+# Resolve this script's dir; under sbatch prefer SLURM_SUBMIT_DIR. Submit from
+# MLExamples/PyTorch_Profiling/rocm-systems-profiler/.
+if [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    SCRIPT_DIR="$(dirname "$(readlink -fm "$0")")"
+fi
+PROFILER_TOP_DIR="$(dirname "${SCRIPT_DIR}")"
+echo "SCRIPT_DIR=${SCRIPT_DIR}"
+echo "PROFILER_TOP_DIR=${PROFILER_TOP_DIR}"
+
+# ../setup_rocm.sh activates the ROCm PyTorch venv and exports ROCm env vars.
+source ${PROFILER_TOP_DIR}/setup_rocm.sh
+rocprof-sys-run --version
+
+# Distributed bootstrap variables expected by train_cifar_100.py (single rank).
+export NPROCS=1
+export MASTER_ADDR=${MASTER_ADDR:-$(hostname)}
+# Derive a per-job port so concurrent jobs don't collide.
+export MASTER_PORT=${MASTER_PORT:-$((20000 + SLURM_JOB_ID % 20000))}
+
+# Make sure the dataset is present before the profiled step.
+if [ ! -d ${PROFILER_TOP_DIR}/data/cifar-100-python ]; then
+    python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+        --data-path ${PROFILER_TOP_DIR}/data --download-only
+fi
+
+cd ${SCRIPT_DIR}
+
+# Remove any stale output directories from previous runs.
+rm -rf ${SCRIPT_DIR}/rocprofsys-python3-output
+
+# Profile with rocprof-sys-run (profiling + tracing).
+echo
+echo "==================================================================="
+echo "rocprof-sys-run --profile --trace -- python3 train_cifar_100.py"
+echo "==================================================================="
+srun -n 1 --gpus=1 --cpus-per-task=8 \
+    rocprof-sys-run --profile --trace -- \
+        python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+            --batch-size 32 --max-steps 5 \
+            --data-path ${PROFILER_TOP_DIR}/data
+
+echo
+echo "==================================================================="
+echo "Profile complete."
+echo "Output directory: ${SCRIPT_DIR}/rocprofsys-python3-output"
+echo "==================================================================="
diff --git a/MLExamples/PyTorch_Profiling/rocprofv3/slurm_single_process_kernels.sh b/MLExamples/PyTorch_Profiling/rocprofv3/slurm_single_process_kernels.sh
new file mode 100755
index 00000000..e4cb75a4
--- /dev/null
+++ b/MLExamples/PyTorch_Profiling/rocprofv3/slurm_single_process_kernels.sh
@@ -0,0 +1,66 @@
+#!/bin/bash
+#SBATCH --job-name=rpv3-single-kernels
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=8
+#SBATCH --gpus=1
+#SBATCH --time=00:30:00
+#SBATCH --output=rpv3_single_process_kernels_%j.out
+#SBATCH --error=rpv3_single_process_kernels_%j.err
+
+# ---------------------------------------------------------------------------
+# Profile the single-process CIFAR-100 workload with rocprofv3 kernel tracing
+# (--stats --kernel-trace). Sources ../setup_rocm.sh to activate the ROCm venv.
+# ---------------------------------------------------------------------------
+
+set -e
+
+# Resolve this script's dir; under sbatch prefer SLURM_SUBMIT_DIR. Submit from
+# MLExamples/PyTorch_Profiling/rocprofv3/.
+if [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    SCRIPT_DIR="$(dirname "$(readlink -fm "$0")")"
+fi
+PROFILER_TOP_DIR="$(dirname "${SCRIPT_DIR}")"
+echo "SCRIPT_DIR=${SCRIPT_DIR}"
+echo "PROFILER_TOP_DIR=${PROFILER_TOP_DIR}"
+
+# ../setup_rocm.sh activates the ROCm PyTorch venv and exports ROCm env vars.
+source ${PROFILER_TOP_DIR}/setup_rocm.sh
+rocprofv3 --version
+
+# Distributed bootstrap variables expected by train_cifar_100.py (single rank).
+export NPROCS=1
+export MASTER_ADDR=${MASTER_ADDR:-$(hostname)}
+# Derive a per-job port so concurrent jobs don't collide.
+export MASTER_PORT=${MASTER_PORT:-$((20000 + SLURM_JOB_ID % 20000))}
+
+# Make sure the dataset is present before the profiled step.
+if [ ! -d ${PROFILER_TOP_DIR}/data/cifar-100-python ]; then
+    python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+        --data-path ${PROFILER_TOP_DIR}/data --download-only
+fi
+
+OUT_DIR=${SCRIPT_DIR}/single_process
+rm -rf ${OUT_DIR}
+cd ${SCRIPT_DIR}
+
+# Profile kernels with rocprofv3.
+echo
+echo "==================================================================="
+echo "rocprofv3 --stats --kernel-trace -- python3 train_cifar_100.py"
+echo "==================================================================="
+srun -n 1 --gpus=1 --cpus-per-task=8 \
+    rocprofv3 --stats --kernel-trace --output-format csv \
+        --output-directory ${OUT_DIR} --output-file kernels \
+        -- \
+        python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+            --batch-size 32 --max-steps 5 \
+            --data-path ${PROFILER_TOP_DIR}/data
+
+echo
+echo "==================================================================="
+echo "Profile complete."
+echo "Output directory: ${OUT_DIR}"
+echo "==================================================================="
diff --git a/MLExamples/PyTorch_Profiling/rocprofv3/slurm_single_process_traces.sh b/MLExamples/PyTorch_Profiling/rocprofv3/slurm_single_process_traces.sh
new file mode 100755
index 00000000..910890ab
--- /dev/null
+++ b/MLExamples/PyTorch_Profiling/rocprofv3/slurm_single_process_traces.sh
@@ -0,0 +1,66 @@
+#!/bin/bash
+#SBATCH --job-name=rpv3-single-traces
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=8
+#SBATCH --gpus=1
+#SBATCH --time=00:30:00
+#SBATCH --output=rpv3_single_process_traces_%j.out
+#SBATCH --error=rpv3_single_process_traces_%j.err
+
+# ---------------------------------------------------------------------------
+# Collect GPU timeline traces for the single-process CIFAR-100 workload with
+# rocprofv3 (--sys-trace). Sources ../setup_rocm.sh to activate the ROCm venv.
+# ---------------------------------------------------------------------------
+
+set -e
+
+# Resolve this script's dir; under sbatch prefer SLURM_SUBMIT_DIR. Submit from
+# MLExamples/PyTorch_Profiling/rocprofv3/.
+if [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    SCRIPT_DIR="$(dirname "$(readlink -fm "$0")")"
+fi
+PROFILER_TOP_DIR="$(dirname "${SCRIPT_DIR}")"
+echo "SCRIPT_DIR=${SCRIPT_DIR}"
+echo "PROFILER_TOP_DIR=${PROFILER_TOP_DIR}"
+
+# ../setup_rocm.sh activates the ROCm PyTorch venv and exports ROCm env vars.
+source ${PROFILER_TOP_DIR}/setup_rocm.sh
+rocprofv3 --version
+
+# Distributed bootstrap variables expected by train_cifar_100.py (single rank).
+export NPROCS=1
+export MASTER_ADDR=${MASTER_ADDR:-$(hostname)}
+# Derive a per-job port so concurrent jobs don't collide.
+export MASTER_PORT=${MASTER_PORT:-$((20000 + SLURM_JOB_ID % 20000))}
+
+# Make sure the dataset is present before the profiled step.
+if [ ! -d ${PROFILER_TOP_DIR}/data/cifar-100-python ]; then
+    python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+        --data-path ${PROFILER_TOP_DIR}/data --download-only
+fi
+
+OUT_DIR=${SCRIPT_DIR}/single_process
+rm -rf ${OUT_DIR}
+cd ${SCRIPT_DIR}
+
+# Collect GPU timeline traces with rocprofv3.
+echo
+echo "==================================================================="
+echo "rocprofv3 --sys-trace -- python3 train_cifar_100.py"
+echo "==================================================================="
+srun -n 1 --gpus=1 --cpus-per-task=8 \
+    rocprofv3 --sys-trace --output-format pftrace \
+        --output-directory ${OUT_DIR} --output-file traces \
+        -- \
+        python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+            --batch-size 32 --max-steps 5 \
+            --data-path ${PROFILER_TOP_DIR}/data
+
+echo
+echo "==================================================================="
+echo "Trace collection complete."
+echo "Output directory: ${OUT_DIR}"
+echo "==================================================================="
diff --git a/MLExamples/PyTorch_Profiling/roofline-extractor/slurm_single_process.sh b/MLExamples/PyTorch_Profiling/roofline-extractor/slurm_single_process.sh
new file mode 100755
index 00000000..b98a1691
--- /dev/null
+++ b/MLExamples/PyTorch_Profiling/roofline-extractor/slurm_single_process.sh
@@ -0,0 +1,80 @@
+#!/bin/bash
+#SBATCH --job-name=roofline-single
+#SBATCH --nodes=1
+#SBATCH --ntasks=1
+#SBATCH --cpus-per-task=8
+#SBATCH --gpus=1
+#SBATCH --time=02:00:00
+#SBATCH --output=roofline_single_process_%j.out
+#SBATCH --error=roofline_single_process_%j.err
+
+# ---------------------------------------------------------------------------
+# Collect roofline plots for the single-process CIFAR-100 workload with
+# AMD-HPC's rooflineExtractor (https://github.com/AMD-HPC/rooflineExtractor).
+# Its profile_app.py runs rocprofv3 several times to collect counters + a kernel
+# trace, then produces the per-kernel roofline analysis and an HTML plot.
+# Sources ../setup_rocm.sh to activate the ROCm venv.
+# ---------------------------------------------------------------------------
+
+set -e
+
+# Resolve this script's dir; under sbatch prefer SLURM_SUBMIT_DIR. Submit from
+# MLExamples/PyTorch_Profiling/roofline-extractor/.
+if [[ -n "${SLURM_SUBMIT_DIR}" ]]; then
+    SCRIPT_DIR="${SLURM_SUBMIT_DIR}"
+else
+    SCRIPT_DIR="$(dirname "$(readlink -fm "$0")")"
+fi
+PROFILER_TOP_DIR="$(dirname "${SCRIPT_DIR}")"
+echo "SCRIPT_DIR=${SCRIPT_DIR}"
+echo "PROFILER_TOP_DIR=${PROFILER_TOP_DIR}"
+
+# ../setup_rocm.sh activates the ROCm PyTorch venv and exports ROCm env vars.
+source ${PROFILER_TOP_DIR}/setup_rocm.sh
+rocprofv3 --version
+
+# Fetch rooflineExtractor and install its Python dependencies (into the venv).
+RE_DIR=${SCRIPT_DIR}/rooflineExtractor
+if [ ! -d ${RE_DIR} ]; then
+    git clone https://github.com/AMD-HPC/rooflineExtractor.git ${RE_DIR}
+fi
+python3 -m pip install -r ${RE_DIR}/requirements.txt
+
+# Distributed bootstrap variables expected by train_cifar_100.py (single rank).
+export NPROCS=1
+export MASTER_ADDR=${MASTER_ADDR:-$(hostname)}
+# Derive a per-job port so concurrent jobs don't collide.
+export MASTER_PORT=${MASTER_PORT:-$((20000 + SLURM_JOB_ID % 20000))}
+
+# Make sure the dataset is present before the profiled runs (profile_app.py runs
+# the application several times, so pre-downloading avoids repeated downloads).
+if [ ! -d ${PROFILER_TOP_DIR}/data/cifar-100-python ]; then
+    python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+        --data-path ${PROFILER_TOP_DIR}/data --download-only
+fi
+
+OUT_DIR=${SCRIPT_DIR}/output
+rm -rf ${OUT_DIR}
+cd ${SCRIPT_DIR}
+
+# Collect roofline data and generate plots. --arch MI300A selects the gfx942
+# counter set; profile_app.py runs rocprofv3 itself, so it is the target.
+echo
+echo "==================================================================="
+echo "profile_app.py --arch MI300A -- python3 train_cifar_100.py"
+echo "==================================================================="
+srun -n 1 --gpus=1 --cpus-per-task=8 \
+    python3 ${RE_DIR}/profile_app.py \
+        --arch MI300A \
+        -o ${OUT_DIR} \
+        -- \
+        python3 ${PROFILER_TOP_DIR}/train_cifar_100.py \
+            --batch-size 32 --max-steps 5 \
+            --data-path ${PROFILER_TOP_DIR}/data
+
+echo
+echo "==================================================================="
+echo "Roofline analysis complete."
+echo "Output directory (counters, traces, plots, analysis): ${OUT_DIR}"
+echo "Open the generated .html file for the interactive roofline plot."
+echo "==================================================================="
diff --git a/MLExamples/PyTorch_Profiling/setup_rocm.sh b/MLExamples/PyTorch_Profiling/setup_rocm.sh
new file mode 100644
index 00000000..cb2998f5
--- /dev/null
+++ b/MLExamples/PyTorch_Profiling/setup_rocm.sh
@@ -0,0 +1,16 @@
+#!/usr/bin/env bash
+# Source this to activate the ROCm venv and set ROCm env vars:
+#   source setup_rocm.sh
+VENV_BASE="${VENV_BASE:-$HOME}"
+VENV="${VENV_BASE}/venvs/rocm-pytorch-pip"
+source "${VENV}/bin/activate"
+# Derive site-packages from the active venv so this works regardless of the
+# venv's Python minor version (e.g. python3.12 vs python3.13).
+DEVEL="$(python3 -c 'import site; print(site.getsitepackages()[0])')/_rocm_sdk_devel"
+export ROCM_PATH="$DEVEL"
+export HIP_PATH="$DEVEL"
+export HIP_DEVICE_LIB_PATH="${DEVEL}/lib/llvm/amdgcn/bitcode"
+export PATH="${DEVEL}/bin:$PATH"
+export LD_LIBRARY_PATH="${DEVEL}/lib:${DEVEL}/lib/rocm_sysdeps/lib:$LD_LIBRARY_PATH"
+echo "ROCm venv active: $VENV"
+