update ai-hub to first optimize model for Workbench

Remove old examples
2026-06-09 14:55:26 -04:00
parent 6c9f30d290
commit f26e8256f0
12 changed files with 260 additions and 700 deletions
--- a/README.md
+++ b/README.md
@@ -177,27 +177,35 @@ The expected output artifact is SageMaker’s `model.tar.gz`, normally containin
 ```
 qc-cli ai-hub upload <calibration.npz|calibration-dir> <inputs.npz|inputs.npy>
 qc-cli ai-hub upload <calibration> <inputs> --from-step validate
-qc-cli ai-hub quantize <calibration.npz|calibration-dir> [--onnx-path PATH] [--model-s3-uri URI] [--from-job NAME]
+qc-cli ai-hub optimize [--onnx-path PATH] [--model-s3-uri URI] [--from-job NAME]
+qc-cli ai-hub quantize <calibration.npz|calibration-dir> [--model-id ID] [--onnx-path PATH] [--model-s3-uri URI] [--from-job NAME]
 qc-cli ai-hub compile [--model-id ID] [--onnx-path PATH] [--model-s3-uri URI] [--from-job NAME]
 qc-cli ai-hub validate <inputs.npz|inputs.npy> [--model-id ID] [--input-name NAME]
 qc-cli ai-hub profile [--model-id ID]
 qc-cli ai-hub download [--model-id ID] [--output PATH]
 ```

-`ai-hub upload` runs the four Workbench upload steps in order: quantize, compile, validate, and profile. Use `--from-step compile`, `--from-step validate`, or `--from-step profile` to resume from saved local state after a completed earlier step.
+`ai-hub upload` optimizes to ONNX, quantizes, validates, and profiles. When `aihub.target_runtime` is not `onnx`, it
+also compiles the quantized model to that deployment runtime. The initial ONNX optimization gives external models
+Workbench provenance and applies compiler optimization passes before quantization.

 Resume behavior:

 ```text
--from-step quantize  Run quantize, compile, validate, and profile.
--from-step compile   Skip quantize; compile the last quantized model unless an explicit source is passed.
--from-step validate  Skip quantize and compile; validate the last compiled model.
--from-step profile   Skip quantize, compile, and validate; profile the last compiled model.
+--from-step optimize  Run optimize, quantize, optional final compile, validate, and profile.
+--from-step quantize  Quantize the last optimized ONNX, then optionally compile, validate, and profile.
+--from-step compile   Skip optimize and quantize; finalize the last quantized model for the target runtime.
+--from-step validate  Skip optimize, quantize, and compile; validate the last compiled model.
+--from-step profile   Skip optimize, quantize, compile, and validate; profile the last compiled model.
 ```

 When a step runs in the current command, `upload` passes its returned model ID directly to the next step. When a step is skipped, the next step resolves the needed model ID from `.qc-cli.json`. This avoids re-running earlier AI Hub jobs when you only need to continue from a later step.

-`ai-hub compile` resolves model sources in this order: `--model-id`, explicit source options (`--onnx-path`, `--model-s3-uri`, `--from-job`), last quantized model from state, then the last training job from local state. `ai-hub download` is separate because downloading the optimized artifact is outside the four-step Workbench upload loop.
+`ai-hub optimize` compiles an external model with `--target_runtime onnx`. `ai-hub quantize` uses an explicit
+`--model-id`, the last optimized ONNX model, or an explicit/local model source in that order. `ai-hub compile` resolves
+model sources in this order: `--model-id`, explicit source options, last quantized model, then the last training job.
+For `target_runtime: onnx`, upload treats the quantized ONNX as the final model and skips a redundant second compile.
+`ai-hub download` remains separate because downloading is outside the Workbench processing loop.

 AI Hub authentication currently uses the local `qai-hub` SDK configuration. A planned follow-up is to support AWS Systems Manager Parameter Store `SecureString` for team-managed tokens, where `config.yaml` stores only a parameter name such as `/qc-cli/aihub/token`, AWS KMS encrypts the token at rest, and the CLI retrieves it at runtime with `ssm:GetParameter` plus `kms:Decrypt` permissions.

--- a/examples/ai-hub/README.md
+++ b/examples/ai-hub/README.md
@@ -1,79 +0,0 @@
-# Qualcomm AI Hub Example
-
-This example takes the ONNX model produced by the SageMaker training example and runs the Qualcomm AI Hub upload workflow:
-
-1. Quantize
-2. Compile
-3. Validate
-4. Profile
-5. Download the compiled artifact
-
-## Prerequisites
-
-Run the training example first and wait for it to complete:
-
-```bash
-examples/training/run_training.sh --wait
-```
-
-The `config.yaml` file must include AI Hub settings:
-
-```yaml
-aihub:
-  device:
-    name: Samsung Galaxy S25 (Family)
-  target_runtime: tflite
-  input_specs:
-    input: [[1, 3, 160, 160], float32]
-  output_dir: build/qai-hub
-```
-
-Finally, the user needs to authenticate with Qualcomm AI Hub using:
-
-```bash
-qai-hub configure --api_token
-```
-
-## Prepare Inputs
-
-AI Hub does not consume the raw JPG training images directly. It needs NumPy tensors that match the ONNX model input shape and preprocessing.
-
-To generate calibration and validation inputs:
-
-```bash
-python examples/ai-hub/prepare_inputs.py
-```
-
-This writes:
-
-```text
-examples/training/data/aihub_calibration/*.npy
-examples/training/data/inputs.npz
-```
-
-The script applies the same image preprocessing used by the training example:
-
- resize to `160x160`
- convert to channel-first `1x3x160x160`
- normalize with ImageNet mean and standard deviation
-
-## Upload Model to Qualcomm Workbench
-
-The model can be uploaded to Qualcomm Workbench using:
-
-```bash
-qc-cli ai-hub upload examples/training/data/aihub_calibration examples/training/data/inputs.npz
-```
-
-The first argument is the calibration path for the model and the second argument is the input file, both of which were created by the `prepare_inputs.py` script. For more details, add `--help` after the `upload` command.
-
-The `upload` command runs the following commands in order:
-1. `qc-cli ai-hub quantize`
-2. `qc-cli ai-hub compile`
-3. `qc-cli ai-hub validate`
-4. `qc-cli ai-hub profile`
-
-Finally the user can download the model from AI Workbench using the command
-```bash
-qc-cli ai-hub download
-```
--- a/examples/ai-hub/prepare_inputs.py
+++ b/examples/ai-hub/prepare_inputs.py
@@ -1,74 +0,0 @@
-#!/usr/bin/env python3
-"""Prepare Qualcomm AI Hub calibration and validation inputs for the training example."""
-
-from __future__ import annotations
-
-import argparse
-from pathlib import Path
-
-import numpy as np
-from PIL import Image
-
-IMAGE_EXTENSIONS = {".jpg", ".jpeg", ".png"}
-
-
-def parse_args() -> argparse.Namespace:
-    parser = argparse.ArgumentParser(description=__doc__)
-    parser.add_argument(
-        "--dataset-dir",
-        type=Path,
-        default=Path("examples/training/data/flower_photos_sagemaker"),
-        help="ImageFolder-style dataset used for training.",
-    )
-    parser.add_argument(
-        "--calibration-dir",
-        type=Path,
-        default=Path("examples/training/data/aihub_calibration"),
-        help="Directory where .npy calibration samples will be written.",
-    )
-    parser.add_argument(
-        "--input-file",
-        type=Path,
-        default=Path("examples/training/data/inputs.npz"),
-        help="Validation .npz input file for qc-cli ai-hub validate.",
-    )
-    parser.add_argument("--input-name", default="input", help="ONNX input name.")
-    parser.add_argument("--image-size", type=int, default=160, help="Square image size used by training.")
-    parser.add_argument("--samples", type=int, default=16, help="Number of calibration samples to write.")
-    return parser.parse_args()
-
-
-def preprocess_image(path: Path, image_size: int) -> np.ndarray:
-    image = Image.open(path).convert("RGB").resize((image_size, image_size), Image.Resampling.BILINEAR)
-    array = np.asarray(image, dtype=np.float32) / 255.0
-    array = np.transpose(array, (2, 0, 1))
-    mean = np.array([0.485, 0.456, 0.406], dtype=np.float32)[:, None, None]
-    std = np.array([0.229, 0.224, 0.225], dtype=np.float32)[:, None, None]
-    return ((array - mean) / std)[None, ...].astype("float32")
-
-
-def main() -> None:
-    args = parse_args()
-    images = sorted(p for p in args.dataset_dir.rglob("*") if p.suffix.lower() in IMAGE_EXTENSIONS)
-    if not images:
-        raise SystemExit(f"No images found under {args.dataset_dir}")
-    if args.samples < 1:
-        raise SystemExit("--samples must be at least 1")
-
-    args.calibration_dir.mkdir(parents=True, exist_ok=True)
-    args.input_file.parent.mkdir(parents=True, exist_ok=True)
-
-    sample_count = min(args.samples, len(images))
-    prepared = []
-    for index, image_path in enumerate(images[:sample_count]):
-        sample = preprocess_image(image_path, args.image_size)
-        np.save(args.calibration_dir / f"sample_{index:03d}.npy", sample)
-        prepared.append(sample)
-
-    np.savez(args.input_file, **{args.input_name: prepared[0]})
-    print(f"Wrote {sample_count} calibration samples to {args.calibration_dir}")
-    print(f"Wrote validation input to {args.input_file}")
-
-
-if __name__ == "__main__":
-    main()
--- a/examples/meter-detection/README.md
+++ b/examples/meter-detection/README.md
@@ -181,7 +181,7 @@ Add AI Hub settings to `config.yaml`. The input name and image size must match t
 aihub:
  device:
    name: Dragonwing IQ-9075 EVK
-  target_runtime: tflite
+  target_runtime: onnx
  input_specs:
    images: [[1, 3, 640, 640], float32]
  job_name: meter-detection
@@ -189,7 +189,7 @@ aihub:
  output_dir: build/qai-hub/meter-detection
 ```

-Use the same image size configured in `sagemaker.training.hyperparameters.imgsz`. For example, a smoke-test model trained with `imgsz: 320` requires `images: [[1, 3, 320, 320], float32]`.
+The ONNX graph is the source of truth. The export normally uses the same value as `sagemaker.training.hyperparameters.imgsz`, but changing `config.yaml` after training does not resize an existing model. For example, a model exported with `imgsz: 320` requires `images: [[1, 3, 320, 320], float32]`.

 ## 7. Prepare AI Hub Inputs

@@ -208,8 +208,6 @@ examples/meter-detection/data/inputs.npz

 The script applies the preprocessing expected by the exported YOLO model: aspect-ratio-preserving letterboxing, RGB channel order, channel-first layout, and pixel values normalized to `[0, 1]`.

-Set `--image-size` to the training `imgsz` value when it is not `640`.
-
 ## 8. Upload To Qualcomm AI Hub

 Use the SageMaker job name printed by `qc-cli train start`:
@@ -221,7 +219,12 @@ qc-cli ai-hub upload \
  --from-job qc-cli-YYYYMMDD-HHMMSS
 ```

-The command downloads the job's `model.tar.gz`, finds `model.onnx`, uploads it to AI Hub, and runs quantization, compilation, validation, and profiling. The uploaded source model uses the configured `aihub.model_name`.
+The command downloads the job's `model.tar.gz`, finds `model.onnx`, and runs the following AI Hub workflow:
+
+1. Compile the external ONNX to a Workbench-optimized ONNX model.
+2. Quantize the optimized ONNX model.
+3. Compile the quantized model when the configured deployment runtime is not `onnx`.
+4. Validate and profile the final model.

 The training example sanitizes the Ultralytics ONNX export before saving `model.onnx`. This removes graph input or output names, such as `output0`, that are duplicated in the ONNX `value_info` metadata and rejected by AI Hub.

@@ -238,24 +241,6 @@ qc-cli ai-hub upload \
  --onnx-path build/qai-hub/meter-detection/model.aihub.onnx
 ```

-If the meter-detection job is still the last training job in `.qc-cli.json`, `--from-job` can be omitted. Keeping it explicit prevents accidentally uploading an artifact from a different training run.
-
-To resume after a completed step, use one of:
-
-```bash
-qc-cli ai-hub upload \
-  examples/meter-detection/data/aihub_calibration \
-  examples/meter-detection/data/inputs.npz \
-  --from-step compile
-```
-
-```bash
-qc-cli ai-hub upload \
-  examples/meter-detection/data/aihub_calibration \
-  examples/meter-detection/data/inputs.npz \
-  --from-step validate
-```
-
 Download the compiled artifact after the workflow completes:

 ```bash
--- a/examples/training/README.md
+++ b/examples/training/README.md
@@ -1,89 +0,0 @@
-# SageMaker Training Example
-
-This example downloads a small image-classification dataset, uploads it through `qc-cli`, and submits a live SageMaker training job.
-
-## Prerequisites
-
- AWS credentials configured for the profile in `config.yaml`
- Infrastructure already deployed with `qc-cli infra setup`
- `config.yaml` updated with:
-
-```yaml
-s3:
-  bucket: your-bucket-name
-
-sagemaker:
-  training:
-    image_uri: 763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:2.6-cpu-py312-ubuntu22.04-sagemaker-v1
-    instance_type: ml.m4.xlarge
-    instance_count: 1
-    source_dir: examples/training/source
-    entry_point: train.py
-    hyperparameters:
-      epochs: 1
-      batch-size: 32
-      learning-rate: 0.001
-      image-size: 160
-      validation-split: 0.2
-```
-
-## Training Hyperparameters
-
-Values under `sagemaker.training.hyperparameters` are passed to the training entry point as command-line arguments. For this example, they map to arguments defined in [source/train.py](source/train.py).
-
-Supported by this example:
-
-| Name | Type | Default | Description |
-|---|---:|---:|---|
-| `epochs` | int | `1` | Number of training epochs. |
-| `batch-size` | int | `32` | Images per training batch. |
-| `learning-rate` | float | `0.001` | Adam optimizer learning rate. |
-| `image-size` | int | `160` | Resize images to square `image-size x image-size`. |
-| `validation-split` | float | `0.2` | Fraction of data used for validation. |
-| `max-samples` | int | `0` | Optional cap for smoke tests; `0` means use all images. |
-| `seed` | int | `13` | Random seed for reproducible splitting. |
-| `num-workers` | int | `2` | DataLoader worker count. |
-
-Do not set `train-dir` or `model-dir` in normal SageMaker runs. SageMaker sets those automatically through `SM_CHANNEL_TRAIN` and `SM_MODEL_DIR`.
-
-## 1. Download The Dataset
-
-```bash
-bash examples/training/download_flower_photos.sh
-```
-
-This creates:
-
-```text
-examples/training/data/flower_photos_sagemaker/
-  daisy/
-  dandelion/
-  roses/
-  sunflowers/
-  tulips/
-```
-
-## 2. Run Training
-
-Run the training script and wait until it finishes:
-
-```bash
-bash examples/training/run_training.sh --config config.yaml --wait
-```
-
-Use a dataset that is already uploaded to `s3.data_prefix`:
-
-```bash
-bash examples/training/run_training.sh \
-  --config config.yaml \
-  --skip-upload \
-  --wait
-```
-
-## Notes
-
- The default dataset path is `examples/training/data/flower_photos_sagemaker`.
- Uploaded data uses the `s3.bucket` and `s3.data_prefix` values from `config.yaml`.
- Training artifacts are written under `s3://<bucket>/<model_prefix>/`.
- The SageMaker `model.tar.gz` contains `model.onnx`, `model.pt`, `class_to_idx.json`, and `metrics.json`.
- SageMaker packages `examples/training/source`, installs `requirements.txt`, and runs `train.py`.
--- a/examples/training/download_flower_photos.sh
+++ b/examples/training/download_flower_photos.sh
@@ -1,40 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-DATASET_URL="https://storage.googleapis.com/download.tensorflow.org/example_images/flower_photos.tgz"
-DEST_DIR="${1:-examples/training/data}"
-ARCHIVE_PATH="${DEST_DIR}/flower_photos.tgz"
-RAW_DATASET_DIR="${DEST_DIR}/flower_photos"
-DATASET_DIR="${DEST_DIR}/flower_photos_sagemaker"
-CLASS_NAMES=("daisy" "dandelion" "roses" "sunflowers" "tulips")
-
-mkdir -p "${DEST_DIR}"
-
-if [[ -d "${DATASET_DIR}" ]]; then
-  echo "Dataset already exists: ${DATASET_DIR}"
-  echo "Use this path with run_training.py:"
-  echo "  ${DATASET_DIR}"
-  exit 0
-fi
-
-echo "Downloading TensorFlow flower_photos dataset..."
-if command -v curl >/dev/null 2>&1; then
-  curl -L "${DATASET_URL}" -o "${ARCHIVE_PATH}"
-elif command -v wget >/dev/null 2>&1; then
-  wget -O "${ARCHIVE_PATH}" "${DATASET_URL}"
-else
-  echo "Either curl or wget is required." >&2
-  exit 1
-fi
-
-echo "Extracting dataset..."
-tar -xzf "${ARCHIVE_PATH}" -C "${DEST_DIR}"
-
-echo "Preparing SageMaker directory layout..."
-mkdir -p "${DATASET_DIR}"
-for class_name in "${CLASS_NAMES[@]}"; do
-  cp -R "${RAW_DATASET_DIR}/${class_name}" "${DATASET_DIR}/${class_name}"
-done
-
-echo "Dataset ready: ${DATASET_DIR}"
-find "${DATASET_DIR}" -mindepth 1 -maxdepth 1 -type d -print | sort
--- a/examples/training/run_training.sh
+++ b/examples/training/run_training.sh
@@ -1,112 +0,0 @@
-#!/usr/bin/env bash
-set -euo pipefail
-
-CONFIG_PATH="config.yaml"
-DATASET_DIR="examples/training/data/flower_photos_sagemaker"
-WAIT=false
-SKIP_UPLOAD=false
-POLL_SECONDS=60
-
-usage() {
-  cat <<EOF
-Usage: $0 [options]
-
-Options:
-  --config PATH          Path to qc-cli config file. Default: config.yaml
-  --dataset-dir PATH     Dataset directory to upload. Default: ${DATASET_DIR}
-  --skip-upload          Train against data already uploaded to s3.data_prefix.
-  --wait                 Poll until training completes.
-  -h, --help             Show this help.
-EOF
-}
-
-while [[ $# -gt 0 ]]; do
-  case "$1" in
-    --config)
-      CONFIG_PATH="$2"
-      shift 2
-      ;;
-    --dataset-dir)
-      DATASET_DIR="$2"
-      shift 2
-      ;;
-    --skip-upload)
-      SKIP_UPLOAD=true
-      shift
-      ;;
-    --wait)
-      WAIT=true
-      shift
-      ;;
-    -h|--help)
-      usage
-      exit 0
-      ;;
-    *)
-      echo "Unknown option: $1" >&2
-      usage >&2
-      exit 1
-      ;;
-  esac
-done
-
-if [[ ! -f "${CONFIG_PATH}" ]]; then
-  echo "Config not found: ${CONFIG_PATH}" >&2
-  exit 1
-fi
-
-if [[ "${SKIP_UPLOAD}" == false && ! -d "${DATASET_DIR}" ]]; then
-  echo "Dataset not found: ${DATASET_DIR}" >&2
-  echo "Run: bash examples/training/download_flower_photos.sh" >&2
-  exit 1
-fi
-
-run() {
-  echo "+ $*"
-  "$@"
-}
-
-run uv run qc-cli infra status --config "${CONFIG_PATH}"
-
-if [[ "${SKIP_UPLOAD}" == false ]]; then
-  run uv run qc-cli upload "${DATASET_DIR}" --config "${CONFIG_PATH}"
-fi
-
-TRAIN_OUTPUT_FILE="$(mktemp)"
-trap 'rm -f "${TRAIN_OUTPUT_FILE}"' EXIT
-run uv run qc-cli train start --config "${CONFIG_PATH}" | tee "${TRAIN_OUTPUT_FILE}"
-
-JOB_NAME="$(grep -Eo 'qc-cli-[0-9]{8}-[0-9]{6}' "${TRAIN_OUTPUT_FILE}" | tail -n 1)"
-if [[ -z "${JOB_NAME}" ]]; then
-  echo "Could not find training job name in qc-cli output." >&2
-  exit 1
-fi
-
-echo "Submitted SageMaker training job: ${JOB_NAME}"
-
-if [[ "${WAIT}" == false ]]; then
-  run uv run qc-cli train status "${JOB_NAME}" --config "${CONFIG_PATH}"
-  exit 0
-fi
-
-while true; do
-  STATUS_OUTPUT="$(uv run qc-cli train status "${JOB_NAME}" --config "${CONFIG_PATH}")"
-  echo "${STATUS_OUTPUT}"
-
-  if printf '%s\n' "${STATUS_OUTPUT}" | grep -q 'Status:.*Completed'; then
-    echo "Training completed successfully."
-    exit 0
-  fi
-
-  if printf '%s\n' "${STATUS_OUTPUT}" | grep -q 'Status:.*Failed'; then
-    echo "Training failed." >&2
-    exit 1
-  fi
-
-  if printf '%s\n' "${STATUS_OUTPUT}" | grep -q 'Status:.*Stopped'; then
-    echo "Training stopped." >&2
-    exit 1
-  fi
-
-  sleep "${POLL_SECONDS}"
-done
--- a/examples/training/source/requirements.txt
+++ b/examples/training/source/requirements.txt
@@ -1 +0,0 @@
-onnx==1.21.0
--- a/examples/training/source/train.py
+++ b/examples/training/source/train.py
@@ -1,188 +0,0 @@
-#!/usr/bin/env python3
-"""SageMaker entry point for CPU image-classification training."""
-
-from __future__ import annotations
-
-import argparse
-import json
-import os
-import random
-from pathlib import Path
-
-import torch
-from torch import nn
-from torch.utils.data import DataLoader, Subset, random_split
-from torchvision import datasets, transforms
-
-
-class SmallImageClassifier(nn.Module):
-    def __init__(self, class_count: int) -> None:
-        super().__init__()
-        self.features = nn.Sequential(
-            nn.Conv2d(3, 16, kernel_size=3, padding=1),
-            nn.ReLU(inplace=True),
-            nn.MaxPool2d(2),
-            nn.Conv2d(16, 32, kernel_size=3, padding=1),
-            nn.ReLU(inplace=True),
-            nn.MaxPool2d(2),
-            nn.Conv2d(32, 64, kernel_size=3, padding=1),
-            nn.ReLU(inplace=True),
-            nn.MaxPool2d(2),
-            nn.AdaptiveAvgPool2d((1, 1)),
-        )
-        self.classifier = nn.Linear(64, class_count)
-
-    def forward(self, x: torch.Tensor) -> torch.Tensor:
-        x = self.features(x)
-        x = torch.flatten(x, 1)
-        return self.classifier(x)
-
-
-def parse_args() -> argparse.Namespace:
-    parser = argparse.ArgumentParser()
-    parser.add_argument("--epochs", type=int, default=1)
-    parser.add_argument("--batch-size", type=int, default=32)
-    parser.add_argument("--learning-rate", type=float, default=0.001)
-    parser.add_argument("--image-size", type=int, default=160)
-    parser.add_argument("--validation-split", type=float, default=0.2)
-    parser.add_argument("--max-samples", type=int, default=0)
-    parser.add_argument("--seed", type=int, default=13)
-    parser.add_argument("--num-workers", type=int, default=2)
-    parser.add_argument("--train-dir", default=os.environ.get("SM_CHANNEL_TRAIN", "/opt/ml/input/data/train"))
-    parser.add_argument("--model-dir", default=os.environ.get("SM_MODEL_DIR", "/opt/ml/model"))
-    return parser.parse_args()
-
-
-def build_datasets(args: argparse.Namespace) -> tuple[Subset, Subset, dict[str, int]]:
-    transform = transforms.Compose(
-        [
-            transforms.Resize((args.image_size, args.image_size)),
-            transforms.ToTensor(),
-            transforms.Normalize(mean=(0.485, 0.456, 0.406), std=(0.229, 0.224, 0.225)),
-        ]
-    )
-    dataset = datasets.ImageFolder(args.train_dir, transform=transform)
-    if len(dataset.classes) < 2:
-        raise ValueError(f"Expected at least two classes in {args.train_dir}. Found: {dataset.classes}")
-
-    if args.max_samples > 0 and args.max_samples < len(dataset):
-        indices = list(range(len(dataset)))
-        random.Random(args.seed).shuffle(indices)
-        dataset = Subset(dataset, indices[: args.max_samples])
-
-    validation_size = max(1, int(len(dataset) * args.validation_split))
-    train_size = len(dataset) - validation_size
-    if train_size < 1:
-        raise ValueError("Not enough images to create a train/validation split.")
-
-    generator = torch.Generator().manual_seed(args.seed)
-    train_dataset, validation_dataset = random_split(dataset, [train_size, validation_size], generator=generator)
-    return train_dataset, validation_dataset, getattr(dataset, "dataset", dataset).class_to_idx
-
-
-def run_epoch(
-    model: nn.Module,
-    data_loader: DataLoader,
-    criterion: nn.Module,
-    optimizer: torch.optim.Optimizer | None,
-    device: torch.device,
-) -> tuple[float, float]:
-    training = optimizer is not None
-    model.train(training)
-
-    total_loss = 0.0
-    total_correct = 0
-    total_examples = 0
-
-    for images, labels in data_loader:
-        images = images.to(device)
-        labels = labels.to(device)
-
-        with torch.set_grad_enabled(training):
-            logits = model(images)
-            loss = criterion(logits, labels)
-
-            if training:
-                optimizer.zero_grad()
-                loss.backward()
-                optimizer.step()
-
-        total_loss += loss.item() * images.size(0)
-        total_correct += (logits.argmax(dim=1) == labels).sum().item()
-        total_examples += images.size(0)
-
-    return total_loss / total_examples, total_correct / total_examples
-
-
-def export_onnx(model: nn.Module, model_dir: Path, image_size: int) -> None:
-    model.eval()
-    dummy_input = torch.randn(1, 3, image_size, image_size)
-    torch.onnx.export(
-        model,
-        dummy_input,
-        model_dir / "model.onnx",
-        export_params=True,
-        opset_version=17,
-        do_constant_folding=True,
-        input_names=["input"],
-        output_names=["logits"],
-    )
-
-
-def main() -> None:
-    args = parse_args()
-    random.seed(args.seed)
-    torch.manual_seed(args.seed)
-
-    train_dataset, validation_dataset, class_to_idx = build_datasets(args)
-    train_loader = DataLoader(
-        train_dataset,
-        batch_size=args.batch_size,
-        shuffle=True,
-        num_workers=args.num_workers,
-    )
-    validation_loader = DataLoader(
-        validation_dataset,
-        batch_size=args.batch_size,
-        shuffle=False,
-        num_workers=args.num_workers,
-    )
-
-    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    model = SmallImageClassifier(class_count=len(class_to_idx)).to(device)
-    criterion = nn.CrossEntropyLoss()
-    optimizer = torch.optim.Adam(model.parameters(), lr=args.learning_rate)
-
-    print(f"Training on {device}. Classes: {sorted(class_to_idx)}")
-    metrics = []
-    for epoch in range(1, args.epochs + 1):
-        train_loss, train_accuracy = run_epoch(model, train_loader, criterion, optimizer, device)
-        validation_loss, validation_accuracy = run_epoch(model, validation_loader, criterion, None, device)
-        epoch_metrics = {
-            "epoch": epoch,
-            "train_loss": train_loss,
-            "train_accuracy": train_accuracy,
-            "validation_loss": validation_loss,
-            "validation_accuracy": validation_accuracy,
-        }
-        metrics.append(epoch_metrics)
-        print(json.dumps(epoch_metrics, sort_keys=True))
-
-    model_dir = Path(args.model_dir)
-    model_dir.mkdir(parents=True, exist_ok=True)
-    torch.save(
-        {
-            "model_state_dict": model.cpu().state_dict(),
-            "class_to_idx": class_to_idx,
-            "image_size": args.image_size,
-        },
-        model_dir / "model.pt",
-    )
-    export_onnx(model, model_dir, args.image_size)
-    (model_dir / "class_to_idx.json").write_text(json.dumps(class_to_idx, indent=2), encoding="utf-8")
-    (model_dir / "metrics.json").write_text(json.dumps(metrics, indent=2), encoding="utf-8")
-    print(f"Saved model artifacts to {model_dir}")
-
-
-if __name__ == "__main__":
-    main()
--- a/src/commands/ai_hub.py
+++ b/src/commands/ai_hub.py
@@ -1,4 +1,5 @@
 from collections.abc import Mapping, Sequence
+from dataclasses import dataclass
 from datetime import datetime
 from enum import StrEnum
 from pathlib import Path
@@ -12,9 +13,9 @@ from src import state as state_ops
 from src.commands.utils import CONFIG_OPT, CONSOLE, load_cfg
 from src.config import Config
 from src.qualcomm import aihub_jobs
-from src.qualcomm.artifacts import resolve_onnx
+from src.qualcomm.artifacts import ResolvedOnnx, resolve_onnx

-app = typer.Typer(help="Quantize, compile, validate, profile, and download models with Qualcomm Workbench")
+app = typer.Typer(help="Optimize, quantize, compile, validate, profile, and download models with Qualcomm Workbench")

 _RUNTIME_EXTENSIONS = {
    "tflite": "tflite",
@@ -24,12 +25,19 @@ _RUNTIME_EXTENSIONS = {


 class UploadStep(StrEnum):
+    optimize = "optimize"
    quantize = "quantize"
    compile = "compile"
    validate = "validate"
    profile = "profile"


+@dataclass(frozen=True)
+class ResolvedModelSource:
+    model: str | Path
+    model_artifact: str | None = None
+
+
 def _input_specs(cfg: Config) -> dict[str, tuple[tuple[int, ...], str]]:
    specs = {name: (tuple(shape), dtype) for name, (shape, dtype) in cfg.aihub.input_specs.items()}
    if not specs:
@@ -101,6 +109,57 @@ def _model_id_or_state(config_path: str, model_id: str | None, *, quantized: boo
    return resolved


+def _resolve_model_source(
+    cfg: Config,
+    config_path: str,
+    *,
+    model_id: str | None = None,
+    previous_model_id: str | None = None,
+    from_job: str | None = None,
+    model_s3_uri: str | None = None,
+    onnx_path: str | None = None,
+) -> ResolvedModelSource:
+    if model_id:
+        return ResolvedModelSource(model_id)
+
+    has_explicit_source = bool(from_job or model_s3_uri or onnx_path)
+    if previous_model_id and not has_explicit_source:
+        return ResolvedModelSource(previous_model_id)
+
+    resolved = _resolve_onnx_source(
+        cfg,
+        config_path,
+        from_job=from_job,
+        model_s3_uri=model_s3_uri,
+        onnx_path=onnx_path,
+    )
+    return ResolvedModelSource(resolved.onnx_path, resolved.model_artifact)
+
+
+def _resolve_onnx_source(
+    cfg: Config,
+    config_path: str,
+    *,
+    from_job: str | None = None,
+    model_s3_uri: str | None = None,
+    onnx_path: str | None = None,
+) -> ResolvedOnnx:
+    st = state_ops.store(config_path)
+    last_training_job = st.get_last_training_job()
+    saved_model_artifact = None
+    if not from_job and not model_s3_uri and not onnx_path and not last_training_job:
+        saved_model_artifact = st.get_last_model_artifact()
+
+    return resolve_onnx(
+        cfg=cfg,
+        output_dir=cfg.aihub.output_dir,
+        from_job=from_job,
+        model_s3_uri=model_s3_uri or saved_model_artifact,
+        onnx_path=onnx_path,
+        last_training_job=last_training_job,
+    )
+
+
 def _device_selector(device: Device) -> str:
    parts: list[str] = []
    if device.name:
@@ -132,20 +191,23 @@ def _quantize_step(
    cfg: Config,
    config_path: str,
    calibration_path: Path,
-    from_job: str | None,
-    model_s3_uri: str | None,
-    onnx_path: str | None,
+    *,
+    model_id: str | None = None,
+    from_job: str | None = None,
+    model_s3_uri: str | None = None,
+    onnx_path: str | None = None,
 ) -> str:
    st = state_ops.store(config_path)
    specs = _input_specs(cfg)
    try:
-        resolved = resolve_onnx(
-            cfg=cfg,
-            output_dir=cfg.aihub.output_dir,
+        source = _resolve_model_source(
+            cfg,
+            config_path,
+            model_id=model_id,
+            previous_model_id=st.get_last_optimized_model_id(),
            from_job=from_job,
-            model_s3_uri=model_s3_uri or st.get_last_model_artifact(),
+            model_s3_uri=model_s3_uri,
            onnx_path=onnx_path,
-            last_training_job=st.get_last_training_job(),
        )
        calibration_data = _load_calibration(calibration_path, specs)
    except (FileNotFoundError, ValueError) as e:
@@ -153,73 +215,117 @@ def _quantize_step(
        raise typer.Exit(1)

    try:
+        hub_model = (
+            hub.upload_model(str(source.model), name=cfg.aihub.model_name)
+            if isinstance(source.model, Path)
+            else hub.get_model(source.model)
+        )
        result = aihub_jobs.submit_quantize_job(
-            resolved.onnx_path,
+            hub_model,
            calibration_data,
            cfg.aihub.quantize_options,
            job_name=_job_name(cfg, "quantize"),
-            model_name=cfg.aihub.model_name,
        )
    except Exception as e:
        CONSOLE.print(f"[red]AI Hub quantize failed: {e}[/red]")
        raise typer.Exit(1)

-    st.update(
-        last_model_artifact=resolved.model_artifact,
-        last_quantize_job_id=result["job_id"],
-        last_quantized_model_id=result["model_id"],
-    )
+    updates: dict[str, Any] = {
+        "last_quantize_job_id": result["job_id"],
+        "last_quantized_model_id": result["model_id"],
+    }
+    if source.model_artifact:
+        updates["last_model_artifact"] = source.model_artifact
+    st.update(**updates)
    CONSOLE.print(f"[green]✓[/green] Quantize job: [bold]{result['job_id']}[/bold]")
    CONSOLE.print(f"[green]✓[/green] Quantized model: [bold]{result['model_id']}[/bold]")
    return str(result["model_id"])


+def _optimize_step(
+    cfg: Config,
+    config_path: str,
+    from_job: str | None,
+    model_s3_uri: str | None,
+    onnx_path: str | None,
+) -> str:
+    st = state_ops.store(config_path)
+    _validate_device(cfg)
+    specs = _input_specs(cfg)
+    try:
+        source = _resolve_onnx_source(
+            cfg,
+            config_path,
+            from_job=from_job,
+            model_s3_uri=model_s3_uri,
+            onnx_path=onnx_path,
+        )
+    except (FileNotFoundError, ValueError) as e:
+        CONSOLE.print(f"[red]{e}[/red]")
+        raise typer.Exit(1)
+
+    try:
+        hub_model = hub.upload_model(str(source.onnx_path), name=cfg.aihub.model_name)
+        result = aihub_jobs.submit_compile_job(
+            model=hub_model,
+            device=cfg.aihub.device,
+            input_specs=specs,
+            target_runtime="onnx",
+            job_name=_job_name(cfg, "optimize"),
+        )
+    except Exception as e:
+        CONSOLE.print(f"[red]AI Hub ONNX optimization failed: {e}[/red]")
+        raise typer.Exit(1)
+
+    st.update(
+        last_model_artifact=source.model_artifact,
+        last_optimize_job_id=result["job_id"],
+        last_optimized_model_id=result["model_id"],
+    )
+    CONSOLE.print(f"[green]✓[/green] ONNX optimization job: [bold]{result['job_id']}[/bold]")
+    CONSOLE.print(f"[green]✓[/green] Optimized ONNX model: [bold]{result['model_id']}[/bold]")
+    return str(result["model_id"])
+
+
 def _compile_step(
    cfg: Config,
    config_path: str,
-    model_id: str | None,
-    from_job: str | None,
-    model_s3_uri: str | None,
-    onnx_path: str | None,
    *,
-    prefer_quantized: bool,
+    model_id: str | None = None,
+    from_job: str | None = None,
+    model_s3_uri: str | None = None,
+    onnx_path: str | None = None,
 ) -> str:
    st = state_ops.store(config_path)
    _validate_device(cfg)
    specs = _input_specs(cfg)
-
-    model: Any
-    model_artifact: str | None = None
-    has_explicit_source = bool(from_job or model_s3_uri or onnx_path)
-    if model_id:
-        model = model_id
-    elif prefer_quantized and not has_explicit_source and st.get_last_quantized_model_id():
-        model = st.get_last_quantized_model_id()
-    else:
-        try:
-            resolved = resolve_onnx(
-                cfg=cfg,
-                output_dir=cfg.aihub.output_dir,
-                from_job=from_job,
-                model_s3_uri=model_s3_uri,
-                onnx_path=onnx_path,
-                last_training_job=st.get_last_training_job(),
-            )
-        except (FileNotFoundError, ValueError) as e:
-            CONSOLE.print(f"[red]{e}[/red]")
-            raise typer.Exit(1)
-        model = resolved.onnx_path
-        model_artifact = resolved.model_artifact
+    try:
+        source = _resolve_model_source(
+            cfg,
+            config_path,
+            model_id=model_id,
+            previous_model_id=st.get_last_quantized_model_id(),
+            from_job=from_job,
+            model_s3_uri=model_s3_uri,
+            onnx_path=onnx_path,
+        )
+    except (FileNotFoundError, ValueError) as e:
+        CONSOLE.print(f"[red]{e}[/red]")
+        raise typer.Exit(1)

    try:
+        hub_model = (
+            hub.upload_model(str(source.model), name=cfg.aihub.model_name)
+            if isinstance(source.model, Path)
+            else hub.get_model(source.model)
+        )
        result = aihub_jobs.submit_compile_job(
-            model=model,
+            model=hub_model,
            device=cfg.aihub.device,
            input_specs=specs,
            target_runtime=cfg.aihub.target_runtime,
            options=cfg.aihub.compile_options,
            job_name=_job_name(cfg, "compile"),
-            model_name=cfg.aihub.model_name if isinstance(model, Path) else None,
        )
    except Exception as e:
        CONSOLE.print(f"[red]AI Hub compile failed: {e}[/red]")
@@ -229,8 +335,8 @@ def _compile_step(
        "last_compile_job_id": result["job_id"],
        "last_compiled_model_id": result["model_id"],
    }
-    if model_artifact:
-        updates["last_model_artifact"] = model_artifact
+    if source.model_artifact:
+        updates["last_model_artifact"] = source.model_artifact
    st.update(**updates)
    CONSOLE.print(f"[green]✓[/green] Compile job: [bold]{result['job_id']}[/bold]")
    CONSOLE.print(f"[green]✓[/green] Compiled model: [bold]{result['model_id']}[/bold]")
@@ -256,8 +362,9 @@ def _validate_step(
    run = datetime.now().strftime("%Y%m%d-%H%M%S")
    out_dir = Path(cfg.aihub.output_dir) / run / "validation"
    try:
+        hub_model = hub.get_model(resolved_model_id)
        result = aihub_jobs.submit_inference_job(
-            resolved_model_id,
+            hub_model,
            cfg.aihub.device,
            inputs,
            out_dir,
@@ -281,8 +388,9 @@ def _profile_step(cfg: Config, config_path: str, model_id: str | None) -> str:
    _validate_device(cfg)
    resolved_model_id = _model_id_or_state(config_path, model_id)
    try:
+        hub_model = hub.get_model(resolved_model_id)
        result = aihub_jobs.submit_profile_job(
-            resolved_model_id,
+            hub_model,
            cfg.aihub.device,
            cfg.aihub.profile_options,
            job_name=_job_name(cfg, "profile"),
@@ -295,9 +403,24 @@ def _profile_step(cfg: Config, config_path: str, model_id: str | None) -> str:
    return str(result["job_id"])


+@app.command()
+def optimize(
+    from_job: str | None = typer.Option(None, "--from-job", help="Training job name whose model artifact should optimize"),
+    model_s3_uri: str | None = typer.Option(None, "--model-s3-uri", help="S3 URI of model.tar.gz to optimize"),
+    onnx_path: str | None = typer.Option(
+        None, "--onnx-path", help="Local ONNX path or ONNX path inside extracted artifact"
+    ),
+    config: str = CONFIG_OPT,
+) -> None:
+    """Optimize an external model into a Workbench-produced ONNX model."""
+    cfg = load_cfg(config)
+    _optimize_step(cfg, config, from_job, model_s3_uri, onnx_path)
+
+
@app.command()
 def quantize(
    calibration_path: Path = typer.Argument(..., help="Calibration .npz file or directory of .npy samples"),
+    model_id: str | None = typer.Option(None, "--model-id", help="AI Hub optimized ONNX model ID"),
    from_job: str | None = typer.Option(None, "--from-job", help="Training job name whose model artifact should quantize"),
    model_s3_uri: str | None = typer.Option(None, "--model-s3-uri", help="S3 URI of model.tar.gz to quantize"),
    onnx_path: str | None = typer.Option(
@@ -307,7 +430,15 @@ def quantize(
 ) -> None:
    """Quantize an ONNX model to INT8."""
    cfg = load_cfg(config)
-    _quantize_step(cfg, config, calibration_path, from_job, model_s3_uri, onnx_path)
+    _quantize_step(
+        cfg,
+        config,
+        calibration_path,
+        model_id=model_id,
+        from_job=from_job,
+        model_s3_uri=model_s3_uri,
+        onnx_path=onnx_path,
+    )


@app.command()
@@ -322,7 +453,14 @@ def compile(
 ) -> None:
    """Compile a model for the configured Qualcomm AI Hub target."""
    cfg = load_cfg(config)
-    _compile_step(cfg, config, model_id, from_job, model_s3_uri, onnx_path, prefer_quantized=True)
+    _compile_step(
+        cfg,
+        config,
+        model_id=model_id,
+        from_job=from_job,
+        model_s3_uri=model_s3_uri,
+        onnx_path=onnx_path,
+    )


@app.command()
@@ -351,7 +489,7 @@ def profile(
 def upload(
    calibration_path: Path = typer.Argument(..., help="Calibration .npz file or directory of .npy samples"),
    input_file: Path = typer.Argument(..., help="Validation .npz or .npy inputs to run on device"),
-    from_step: UploadStep = typer.Option(UploadStep.quantize, "--from-step", help="Resume from this Workbench step"),
+    from_step: UploadStep = typer.Option(UploadStep.optimize, "--from-step", help="Resume from this Workbench step"),
    from_job: str | None = typer.Option(None, "--from-job", help="Training job name whose model artifact should upload"),
    model_s3_uri: str | None = typer.Option(None, "--model-s3-uri", help="S3 URI of model.tar.gz to upload"),
    onnx_path: str | None = typer.Option(
@@ -360,25 +498,48 @@ def upload(
    input_name: str | None = typer.Option(None, "--input-name", help="Input name for .npy validation files"),
    config: str = CONFIG_OPT,
 ) -> None:
-    """Run the four Workbench upload steps: quantize, compile, validate, and profile."""
+    """Optimize, quantize, optionally compile, validate, and profile a model."""
    cfg = load_cfg(config)
-    steps = [UploadStep.quantize, UploadStep.compile, UploadStep.validate, UploadStep.profile]
+    steps = [UploadStep.optimize, UploadStep.quantize, UploadStep.compile, UploadStep.validate, UploadStep.profile]
    selected = steps[steps.index(from_step) :]

+    optimized_model_id: str | None = None
    quantized_model_id: str | None = None
    compiled_model_id: str | None = None
+    if UploadStep.optimize in selected:
+        optimized_model_id = _optimize_step(cfg, config, from_job, model_s3_uri, onnx_path)
    if UploadStep.quantize in selected:
-        quantized_model_id = _quantize_step(cfg, config, calibration_path, from_job, model_s3_uri, onnx_path)
-    if UploadStep.compile in selected:
-        compiled_model_id = _compile_step(
+        if UploadStep.optimize not in selected:
+            optimized_model_id = state_ops.store(config).get_last_optimized_model_id()
+            if not optimized_model_id:
+                CONSOLE.print(
+                    "[red]No optimized ONNX model found. Resume from --from-step optimize or run "
+                    "'qc-cli ai-hub optimize' first.[/red]"
+                )
+                raise typer.Exit(1)
+        quantized_model_id = _quantize_step(
            cfg,
            config,
-            model_id=quantized_model_id,
-            from_job=from_job,
-            model_s3_uri=model_s3_uri,
-            onnx_path=onnx_path,
-            prefer_quantized=True,
+            calibration_path,
+            model_id=optimized_model_id,
        )
+    if UploadStep.compile in selected:
+        if cfg.aihub.target_runtime == "onnx":
+            compiled_model_id = quantized_model_id or state_ops.store(config).get_last_quantized_model_id()
+            if not compiled_model_id:
+                CONSOLE.print(
+                    "[red]No quantized ONNX model found. Resume from --from-step quantize or run "
+                    "'qc-cli ai-hub quantize' first.[/red]"
+                )
+                raise typer.Exit(1)
+            state_ops.store(config).update(last_compiled_model_id=compiled_model_id)
+            CONSOLE.print("[green]✓[/green] Target runtime is ONNX; skipping final compile.")
+        else:
+            compiled_model_id = _compile_step(
+                cfg,
+                config,
+                model_id=quantized_model_id,
+            )
    if UploadStep.validate in selected:
        _validate_step(cfg, config, input_file, compiled_model_id, input_name)
    if UploadStep.profile in selected:
--- a/src/qualcomm/aihub_jobs.py
+++ b/src/qualcomm/aihub_jobs.py
@@ -28,30 +28,19 @@ def _dataset_entries(inputs: dict[str, Any]) -> dict[str, list[Any]]:


 def submit_compile_job(
-    model: Any,
+    model: Model,
    device: Device,
    input_specs: dict[str, tuple[tuple[int, ...], str]],
    target_runtime: str,
    options: str | None = None,
    job_name: str | None = None,
-    model_name: str | None = None,
 ) -> ModelJobResult:
    compile_options = f"--target_runtime {target_runtime}"
    if options:
        compile_options = f"{compile_options} {options}"

-    model_arg = model
-    if isinstance(model, Path):
-        model_arg = str(model)
-    elif isinstance(model, str):
-        candidate = Path(model)
-        model_arg = model if candidate.exists() or candidate.suffix else hub.get_model(model)
-
-    if model_name and isinstance(model_arg, str) and Path(model_arg).exists():
-        model_arg = hub.upload_model(model_arg, name=model_name)
-
    job = hub.submit_compile_job(
-        model=model_arg,
+        model=model,
        device=device,
        name=job_name,
        input_specs=input_specs,
@@ -64,14 +53,14 @@ def submit_compile_job(


 def submit_inference_job(
-    model_id: str,
+    model: Model,
    device: Device,
    inputs: dict[str, Any],
    output_dir: str | Path,
    job_name: str | None = None,
 ) -> InferenceJobResult:
    job = hub.submit_inference_job(
-        model=hub.get_model(model_id),
+        model=model,
        device=device,
        inputs=_dataset_entries(inputs),
        name=job_name,
@@ -83,13 +72,13 @@ def submit_inference_job(


 def submit_profile_job(
-    model_id: str,
+    model: Model,
    device: Device,
    options: str | None = None,
    job_name: str | None = None,
 ) -> ProfileJobResult:
    job = hub.submit_profile_job(
-        model=hub.get_model(model_id),
+        model=model,
        device=device,
        name=job_name,
        options=options or "",
@@ -98,17 +87,13 @@ def submit_profile_job(


 def submit_quantize_job(
-    model: str | Path,
+    model: Model,
    calibration_data: dict[str, Any],
    options: str | None = None,
    job_name: str | None = None,
-    model_name: str | None = None,
 ) -> ModelJobResult:
-    model_arg = str(model)
-    if model_name and Path(model_arg).exists():
-        model_arg = hub.upload_model(model_arg, name=model_name)
    job = hub.submit_quantize_job(
-        model=model_arg,
+        model=model,
        calibration_data=_dataset_entries(calibration_data),
        weights_dtype=QuantizeDtype.INT8,
        activations_dtype=QuantizeDtype.INT8,
--- a/src/state.py
+++ b/src/state.py
@@ -37,6 +37,10 @@ class CliStateStore:
        value = self.get("last_model_artifact")
        return str(value) if value else None

+    def get_last_optimized_model_id(self) -> str | None:
+        value = self.get("last_optimized_model_id")
+        return str(value) if value else None
+
    def get_last_quantized_model_id(self) -> str | None:
        value = self.get("last_quantized_model_id")
        return str(value) if value else None