feat: add ROCm (AMD GPU) and Intel GPU support (v0.2.12)

New image variants: - :rocm — InsightFace via ROCmExecutionProvider, SigLIP via PyTorch ROCm 6.3 - :intel — InsightFace via OpenVINOExecutionProvider (onnxruntime-openvino); Intel GPU compute runtime auto-installed from Intel graphics repo; OPENVINO_DEVICE=GPU opts into Arc/iGPU inference (default: CPU) Also adds: - pyproject-rocm.toml + uv-rocm.lock, pyproject-intel.toml + uv-intel.lock - compose.yml device passthrough snippets for AMD and Intel - CI: build-rocm and build-intel jobs in docker-publish.yml; all four variants built and tagged in release.yml - README reworked: cleaner structure, GPU variant quick-start examples, OPENVINO_DEVICE env var documented - CHANGELOG entry and version bump to 0.2.12 Fix: IntPrompt in dict literal was eagerly evaluated in the no-embedding fallback path of _get_strategy_choice, prompting users for a custom count regardless of which strategy they picked. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-06-13 02:31:09 +00:00
parent a281b4b896
commit b2be19e259
16 changed files with 4550 additions and 183 deletions
@@ -199,3 +199,127 @@ jobs:
            -f visibility=public || true
        env:
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+  build-rocm:
+    name: Build ROCm / AMD GPU (amd64)
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+
+    steps:
+      - name: Free up disk space
+        run: |
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /opt/ghc
+          sudo rm -rf "/usr/local/share/boost"
+          sudo rm -rf "$AGENT_TOOLSDIRECTORY"
+          echo "Disk space freed."
+
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v4
+
+      - name: Log in to GHCR
+        uses: docker/login-action@v4
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Determine ROCm image tag
+        id: rocm-tag
+        run: |
+          if [ "${{ github.ref_name }}" = "dev" ]; then
+            echo "tag=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:dev-rocm" >> "$GITHUB_OUTPUT"
+          else
+            echo "tag=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:rocm" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Build and push ROCm image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ./Dockerfile
+          platforms: linux/amd64
+          build-args: VARIANT=rocm
+          cache-from: type=gha,scope=linux/amd64-rocm
+          cache-to: type=gha,mode=max,scope=linux/amd64-rocm
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          push: true
+          tags: ${{ steps.rocm-tag.outputs.tag }}
+
+      - name: Inspect ROCm image
+        run: |
+          docker buildx imagetools inspect ${{ steps.rocm-tag.outputs.tag }}
+
+      - name: Ensure package is public
+        run: |
+          gh api -X PATCH /user/packages/container/winnow \
+            -f visibility=public || true
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+  build-intel:
+    name: Build Intel GPU (amd64)
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      packages: write
+
+    steps:
+      - name: Free up disk space
+        run: |
+          sudo rm -rf /usr/share/dotnet
+          sudo rm -rf /opt/ghc
+          sudo rm -rf "/usr/local/share/boost"
+          sudo rm -rf "$AGENT_TOOLSDIRECTORY"
+          echo "Disk space freed."
+
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v4
+
+      - name: Log in to GHCR
+        uses: docker/login-action@v4
+        with:
+          registry: ${{ env.REGISTRY }}
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Determine Intel image tag
+        id: intel-tag
+        run: |
+          if [ "${{ github.ref_name }}" = "dev" ]; then
+            echo "tag=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:dev-intel" >> "$GITHUB_OUTPUT"
+          else
+            echo "tag=${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:intel" >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Build and push Intel image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ./Dockerfile
+          platforms: linux/amd64
+          build-args: VARIANT=intel
+          cache-from: type=gha,scope=linux/amd64-intel
+          cache-to: type=gha,mode=max,scope=linux/amd64-intel
+          github-token: ${{ secrets.GITHUB_TOKEN }}
+          push: true
+          tags: ${{ steps.intel-tag.outputs.tag }}
+
+      - name: Inspect Intel image
+        run: |
+          docker buildx imagetools inspect ${{ steps.intel-tag.outputs.tag }}
+
+      - name: Ensure package is public
+        run: |
+          gh api -X PATCH /user/packages/container/winnow \
+            -f visibility=public || true
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
@@ -118,15 +118,57 @@ jobs:
          username: ${{ github.actor }}
          password: ${{ secrets.GITHUB_TOKEN }}

-      - name: Build and push Docker image
+      - name: Build and push GPU image (latest)
        uses: docker/build-push-action@v6
        with:
          context: .
          file: ./Dockerfile
          platforms: linux/amd64,linux/arm64
          push: true
-          cache-from: type=gha
-          cache-to: type=gha,mode=max
+          cache-from: type=gha,scope=release-gpu
+          cache-to: type=gha,mode=max,scope=release-gpu
          tags: |
            ghcr.io/sudolulo/winnow:latest
            ghcr.io/sudolulo/winnow:${{ steps.tag.outputs.TAG }}
+
+      - name: Build and push CPU image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ./Dockerfile
+          platforms: linux/amd64,linux/arm64
+          build-args: VARIANT=cpu
+          push: true
+          cache-from: type=gha,scope=release-cpu
+          cache-to: type=gha,mode=max,scope=release-cpu
+          tags: |
+            ghcr.io/sudolulo/winnow:cpu
+            ghcr.io/sudolulo/winnow:${{ steps.tag.outputs.TAG }}-cpu
+
+      - name: Build and push ROCm image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ./Dockerfile
+          platforms: linux/amd64
+          build-args: VARIANT=rocm
+          push: true
+          cache-from: type=gha,scope=release-rocm
+          cache-to: type=gha,mode=max,scope=release-rocm
+          tags: |
+            ghcr.io/sudolulo/winnow:rocm
+            ghcr.io/sudolulo/winnow:${{ steps.tag.outputs.TAG }}-rocm
+
+      - name: Build and push Intel image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: ./Dockerfile
+          platforms: linux/amd64
+          build-args: VARIANT=intel
+          push: true
+          cache-from: type=gha,scope=release-intel
+          cache-to: type=gha,mode=max,scope=release-intel
+          tags: |
+            ghcr.io/sudolulo/winnow:intel
+            ghcr.io/sudolulo/winnow:${{ steps.tag.outputs.TAG }}-intel
@@ -7,6 +7,20 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0

 ## [Unreleased]

+## [0.2.12] - 2026-06-13
+
+### Added
+
+- **ROCm (AMD GPU) support**: new `:rocm` image tag. InsightFace runs via `ROCmExecutionProvider`; SigLIP runs via PyTorch ROCm 6.3 (ROCm builds expose `torch.cuda.is_available() == True`, so the existing CUDA path is reused automatically). Requires `/dev/kfd` and `/dev/dri` device passthrough plus `video` and `render` group membership — see `compose.yml` for the snippet.
+- **Intel GPU support**: new `:intel` image tag. InsightFace runs via `OpenVINOExecutionProvider` from `onnxruntime-openvino`. By default OpenVINO targets CPU (no device passthrough needed); set `OPENVINO_DEVICE=GPU` to target Intel Arc discrete or integrated graphics. Intel's GPU compute runtime (Level Zero + OpenCL ICD) is installed automatically from Intel's official graphics repo in the image — no manual package installation required. SigLIP uses CPU inference for now (Intel Extension for PyTorch has no Python 3.13 wheels yet; the `torch.xpu` path is wired and will activate automatically when they ship).
+- **`OPENVINO_DEVICE` env var**: controls the OpenVINO execution provider device for the `:intel` variant. `CPU` (default) requires no device passthrough. `GPU` targets Intel Arc discrete and integrated graphics via Level Zero.
+- **AMD and Intel device passthrough snippets in `compose.yml`**: documented as commented-out alternatives to the NVIDIA `deploy:` block.
+- **`:rocm` and `:intel` CI jobs**: `docker-publish.yml` now builds and pushes `:rocm` / `:dev-rocm` and `:intel` / `:dev-intel` alongside `:latest` and `:cpu`. `release.yml` builds all four variants on tag push.
+
+### Fixed
+
+- **Interactive custom-count prompt firing for all choices**: in the no-embedding fallback path of the strategy selector, `IntPrompt.ask` was inside a dict literal and evaluated eagerly — users selecting Standard (30) or Broad (100) were still prompted to enter a custom image count. Each choice is now handled in a dedicated branch.
+
 ## [0.2.11] - 2026-06-12

 ### Fixed
@@ -1,13 +1,19 @@
 # ── Base images ───────────────────────────────────────────────────────────────
-#   amd64 + gpu:  NVIDIA CUDA 13.3 + cuDNN (GPU acceleration when available)
-#   amd64 + cpu:  Ubuntu 22.04 (CPU-only, ~2 GB smaller image)
-#   arm64:        Ubuntu 24.04 (CPU-only; no CUDA wheels on ARM)
+#   amd64 + gpu:   NVIDIA CUDA 13.3 + cuDNN (GPU acceleration via NVIDIA Container Toolkit)
+#   amd64 + rocm:  Ubuntu 22.04 (AMD GPU via ROCm — pass /dev/kfd and /dev/dri)
+#   amd64 + intel: Ubuntu 22.04 (Intel Arc / iGPU via OpenVINO — pass /dev/dri)
+#   amd64 + cpu:   Ubuntu 22.04 (CPU-only, ~2 GB smaller image)
+#   arm64:         Ubuntu 24.04 (CPU-only; no CUDA/ROCm wheels on ARM)

 ARG VARIANT=gpu

 FROM --platform=$BUILDPLATFORM nvidia/cuda:13.3.0-cudnn-runtime-ubuntu22.04 AS base-amd64-gpu
+FROM ubuntu:22.04 AS base-amd64-rocm
+FROM ubuntu:22.04 AS base-amd64-intel
 FROM ubuntu:22.04 AS base-amd64-cpu
 FROM ubuntu:24.04 AS base-arm64-gpu
+FROM ubuntu:24.04 AS base-arm64-rocm
+FROM ubuntu:24.04 AS base-arm64-intel
 FROM ubuntu:24.04 AS base-arm64-cpu

 # ── Build stage ───────────────────────────────────────────────────────────────
@@ -35,11 +41,15 @@ RUN curl -LsSf https://astral.sh/uv/install.sh | sh \

 WORKDIR /app

-# For cpu variant, swap in the CPU-only pyproject and lockfile before syncing.
-COPY pyproject.toml uv.lock pyproject-cpu.toml uv-cpu.lock ./
+# Swap in the variant-specific pyproject and lockfile before syncing.
+COPY pyproject.toml uv.lock pyproject-cpu.toml uv-cpu.lock \
+     pyproject-rocm.toml uv-rocm.lock pyproject-intel.toml uv-intel.lock ./
 RUN if [ "$VARIANT" = "cpu" ]; then \
-        cp pyproject-cpu.toml pyproject.toml && \
-        cp uv-cpu.lock uv.lock; \
+        cp pyproject-cpu.toml pyproject.toml && cp uv-cpu.lock uv.lock; \
+    elif [ "$VARIANT" = "rocm" ]; then \
+        cp pyproject-rocm.toml pyproject.toml && cp uv-rocm.lock uv.lock; \
+    elif [ "$VARIANT" = "intel" ]; then \
+        cp pyproject-intel.toml pyproject.toml && cp uv-intel.lock uv.lock; \
    fi && \
    uv sync --frozen --no-dev \
    && uv cache clean
@@ -72,11 +82,31 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 COPY --from=build /app /app
 COPY --from=build /usr/local/bin/uv /usr/local/bin/uv

-# Register every nvidia pip-package lib/ directory with ldconfig so that
-# onnxruntime-gpu and torch can find libcudnn, libcublas, libcufft, etc.
-# without a hand-maintained LD_LIBRARY_PATH. Skipped silently on cpu builds.
-RUN find /app/.venv/lib/python3.*/site-packages/nvidia -type d -name "lib" \
-        2>/dev/null > /etc/ld.so.conf.d/nvidia-pip.conf && ldconfig || true
+# NVIDIA: register pip-installed nvidia lib/ dirs with ldconfig so onnxruntime-gpu
+# and torch can find libcudnn, libcublas, etc. Skipped silently on other variants.
+RUN if [ "$VARIANT" = "gpu" ]; then \
+        find /app/.venv/lib/python3.*/site-packages/nvidia -type d -name "lib" \
+            2>/dev/null > /etc/ld.so.conf.d/nvidia-pip.conf && ldconfig || true; \
+    fi
+# Intel: install GPU compute runtime so OpenVINO EP can target Intel Arc / iGPU.
+# onnxruntime-openvino bundles OpenVINO itself; only the userspace GPU driver
+# (OpenCL ICD + Level Zero) is needed from the OS.
+# libze-intel-gpu1 / intel-level-zero-gpu aren't in Ubuntu 22.04 main, so this
+# block adds Intel's official graphics repo first, then installs.
+RUN if [ "$VARIANT" = "intel" ]; then \
+        apt-get update \
+        && apt-get install -y --no-install-recommends curl gnupg \
+        && curl -fsSL https://repositories.intel.com/graphics/intel-graphics.key \
+           | gpg --dearmor > /usr/share/keyrings/intel-graphics.gpg \
+        && echo "deb [arch=amd64 signed-by=/usr/share/keyrings/intel-graphics.gpg] \
+https://repositories.intel.com/graphics/ubuntu jammy flex" \
+           > /etc/apt/sources.list.d/intel-graphics.list \
+        && apt-get update \
+        && apt-get install -y --no-install-recommends \
+               intel-opencl-icd intel-level-zero-gpu libze-intel-gpu1 \
+        && apt-get remove -y --autoremove curl gnupg \
+        && rm -rf /var/lib/apt/lists/*; \
+    fi

 RUN groupadd -g 568 apps && useradd -u 568 -g apps -m -s /bin/bash appuser \
    && mkdir -p /models/.insightface /models/huggingface \
@@ -4,28 +4,14 @@

 **Docs:** [Setup](https://github.com/sudolulo/winnow/wiki/Setup) · [Troubleshooting](https://github.com/sudolulo/winnow/wiki/Troubleshooting) · [FAQ](https://github.com/sudolulo/winnow/wiki/FAQ)

-`winnow` pulls photos of people and objects from your [Immich](https://immich.app) library, selects the most diverse and highest-quality subset using AI embeddings, and delivers them as training data for [Frigate](https://frigate.video)'s face recognition and object classification models.
+`winnow` pulls photos from your [Immich](https://immich.app) library, selects the most diverse and highest-quality subset using AI embeddings, and delivers them as training data for [Frigate](https://frigate.video)'s face recognition and object classification models.

-It runs fully headless in Docker, is configured entirely through environment variables, and can run on a schedule — no interactive prompts, no manual steps.
-
---
-
-## The Problem
-
-Frigate's face recognition model (ArcFace) and object classifier are only as good as the training data you give them. The instinct is to feed them as many photos as possible, but volume is not what matters — **diversity is**.
-
-If you upload 100 photos from the same week, the model learns the lighting in your living room and the jacket you wore that month. It struggles the moment anything changes. What you actually want is a spread: different years, different lighting conditions, different angles, different contexts.
-
-This is especially true for people who have never been to your property, or who visit rarely — family members, friends, anyone Frigate has never seen in person. Live detections alone will never build a reliable model for these people. Your photo library already has the data; winnow finds and delivers the right subset of it.
-
-Finding that spread manually across a library of thousands of photos is not practical. `winnow` does it automatically.
+Frigate's face recognition is only as good as its training data — and the key quality metric is **diversity**, not volume. A hundred photos from the same week teach the model one lighting condition. What you need is a spread: different years, different angles, different lighting, different contexts. Your photo library already has that data. winnow finds and delivers the right subset automatically.

 ---

 ## How It Works

-For each person (or object) you configure, the tool runs this pipeline:
-
 ```
 Immich library
      │
@@ -41,71 +27,39 @@ Immich library
      ▼
 4. Quality filter — reject:
   • Blurry images (Laplacian variance)
-   • Grayscale / infrared (channel similarity check)
+   • Grayscale / infrared (channel similarity)
   • Over- or underexposed
   • Low detection confidence
   • Face crops below minimum pixel size
      │
      ▼
-5. Compute embeddings for remaining candidates
+5. Compute embeddings
   • Faces   → InsightFace (ArcFace / Buffalo_L)
   • Objects → SigLIP (Vision Transformer)
      │
      ▼
 6. Diversity selection
-   • K-Medoids clustering to find natural groupings
-   • Farthest Point Sampling (FPS) to pick maximally spread representatives
-   • Hard example weighting — unusual angles, partial occlusions,
-     and low-confidence detections are biased toward selection
-   • Auto mode: keeps selecting until marginal diversity drops off
+   • K-Medoids clustering → one representative per natural group
+   • Farthest Point Sampling → fill remaining slots with maximally spread picks
+   • Hard example weighting — unusual angles and low-confidence detections
+     are biased toward selection, since those are where models tend to fail
+   • Auto mode: stops when the next candidate is too similar to what's already chosen
      │
      ▼
-7. Crop and export
-   • Face mode: aligned 112×112 crops (ArcFace standard input),
-     uploaded directly to Frigate's face training API
+7. Crop and deliver
+   • Face mode: aligned 112×112 crops uploaded directly to Frigate's face training API
   • Object mode: YOLO-detected crops saved to disk
 ```

-Uploaded asset IDs are recorded so the same image is never uploaded twice, even across runs weeks apart.
-
---
-
-## Note on Crop Quality
-
-winnow works well, but no automated pipeline is perfect. Occasionally a bad crop will slip through quality filtering — a partial face, someone in the background, a blurry frame. After a run it's worth a quick review in Frigate's face management UI to remove anything that doesn't belong.
-
-Issues and feedback welcome via [GitHub Issues](https://github.com/sudolulo/winnow/issues).
+Uploaded asset IDs are persisted so the same image is never uploaded twice, even across runs weeks apart.

 ---

 ## Modes

-### Face Mode (default)
+**Face mode** (default) — extracts face crops using Immich's bounding box metadata, applies EXIF orientation correction, and aligns them to ArcFace's standard 112×112 format using 5-point facial landmarks. Crops are uploaded directly to Frigate's face registration API.

-Extracts face crops using Immich's bounding box metadata, scales them to the source image resolution, applies EXIF orientation correction, then either aligns them to the standard ArcFace 112×112 format using 5-point facial landmarks or falls back to a margin-padded bounding box crop.
-
-Crops are uploaded directly to Frigate's face registration API (`POST /api/faces/{name}/register`). After each successful upload the asset ID is marked in the tracker so future runs skip it.
-
-### Object Mode
-
-Runs each full image through YOLOv9c to detect instances of a target class (dog, cat, car, etc.), then crops each detection and saves it to the output directory. Frigate has no API for uploading object training images, so the crops are saved for you to place into your Frigate data directory manually.
-
---
-
-## Diversity Selection in Detail
-
-The core of the tool is the embedding-based selection. Rather than picking images at random or evenly across time, it computes a vector embedding for each candidate image that encodes what the face or object actually looks like — the angle, lighting, expression, background context.
-
-It then:
-
-1. **Clusters** those embeddings using K-Medoids to find natural groups (e.g. "holiday photos", "outdoor summer shots", "indoor low light")
-2. **Selects one representative** from each cluster — the most central image in each group
-3. **Fills remaining slots** using Farthest Point Sampling, iteratively picking whichever image is most different from everything already selected
-4. **Weights toward hard examples** — images with unusual angles, partial occlusions, or borderline detection confidence are more likely to be picked, because those edge cases are where models fail
-
-In **Auto mode**, there is no fixed limit. The tool keeps selecting until the most-different remaining image is already close to something already in the set — at that point adding more would be redundant. This is capped at `MAX_AUTO_IMAGES` (default 80) as a safety limit.
-
-If the embedding model is unavailable, the tool falls back to **time spread**: evenly distributing picks across the date range of your photos.
+**Object mode** — runs each image through YOLOv9c to detect instances of a target class (dog, cat, car, etc.), crops each detection, and saves it to the output directory. Frigate has no API for uploading object training data; place the crops into your Frigate data directory manually.

 ---

@@ -113,22 +67,25 @@ If the embedding model is unavailable, the tool falls back to **time spread**: e

 ### Image Tags

-| Tag | Arch | GPU | Notes |
-| :-- | :-- | :-- | :-- |
-| `:latest` | amd64 + arm64 | CUDA 13.3 (amd64) | Requires NVIDIA Container Toolkit on amd64 |
-| `:cpu` | amd64 | None | ~2 GB smaller; use if you have no NVIDIA GPU |
+| Tag | Arch | Acceleration |
+| :-- | :-- | :-- |
+| `:latest` | amd64 + arm64 | NVIDIA CUDA 13.3 (amd64) · requires [NVIDIA Container Toolkit](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html) |
+| `:rocm` | amd64 | AMD ROCm · pass `/dev/kfd` + `/dev/dri` |
+| `:intel` | amd64 | Intel Arc / iGPU via OpenVINO · pass `/dev/dri`, set `OPENVINO_DEVICE=GPU` |
+| `:cpu` | amd64 + arm64 | CPU only · ~2 GB smaller · no GPU required |

 ### Quick Start

+**NVIDIA:**
 ```yaml
 services:
  winnow:
-    image: ghcr.io/sudolulo/winnow:latest   # or :cpu for CPU-only amd64
+    image: ghcr.io/sudolulo/winnow:latest
    environment:
      - IMMICH_URL=http://192.168.1.10:2283
      - API_KEY=your-immich-api-key
      - FRIGATE_URL=http://192.168.1.10:5000
-      - CRON_SCHEDULE=0 3 * * 0   # Every Sunday at 3 AM
+      - CRON_SCHEDULE=0 3 * * 0
    volumes:
      - /path/to/models:/models
      - /path/to/cache:/app/.if_cache
@@ -142,11 +99,31 @@ services:
              capabilities: [gpu]
 ```

-> **CPU users (`:cpu` tag):** remove the `deploy.resources` block — no NVIDIA runtime needed. Add `mem_limit: 2g` to the service to prevent an OOM restart loop on large libraries.
+**AMD (`:rocm`):** use `image: ghcr.io/sudolulo/winnow:rocm` and replace the `deploy:` block with:
+```yaml
+    devices:
+      - /dev/kfd
+      - /dev/dri
+    group_add:
+      - video
+      - render
+```

-See [compose.yml](compose.yml) for the full annotated example.
+**Intel (`:intel`):** use `image: ghcr.io/sudolulo/winnow:intel` and replace the `deploy:` block with:
+```yaml
+    devices:
+      - /dev/dri
+    group_add:
+      - render
+    environment:
+      - OPENVINO_DEVICE=GPU   # omit to run OpenVINO inference on CPU (default)
+```

-### Scheduling Behaviour
+**CPU (`:cpu`):** use `image: ghcr.io/sudolulo/winnow:cpu`, remove the `deploy:` block, and add `mem_limit: 2g` to prevent OOM on large libraries.
+
+See [compose.yml](compose.yml) for the full annotated example with all options.
+
+### Scheduling

 `CRON_SCHEDULE` controls container lifetime:

@@ -156,69 +133,68 @@ See [compose.yml](compose.yml) for the full annotated example.
 | *(empty string)* | Stay alive, run nothing — trigger manually with `docker exec -it winnow winnow` |
 | Cron expression | Run on startup, then repeat on schedule |

-In scheduled mode the process (and loaded models) stays resident between runs. In manual mode the container idles indefinitely with `sleep infinity` — useful when you want to trigger runs interactively on demand without pulling a new container each time.
-
-The first run after a fresh install downloads the embedding models (~1-2 GB). Subsequent runs use the cached models from the mounted volume and start immediately.
+In scheduled mode the process (and loaded models) stays resident between runs. The first run after a fresh install downloads the embedding models (~1–2 GB); subsequent runs use the cached models from the mounted volume.

 ---

 ## Environment Variables

-### Mode & Strategy
-
-| Variable | Default | Description |
-| :--- | :--- | :--- |
-| `AUTO_MODE` | *(auto)* | Force non-interactive mode even in a terminal; auto-detected otherwise (no TTY = auto) |
-| `VERBOSE` | `false` | Set to `true` to enable DEBUG-level console output (the log file is always DEBUG) |
-| `TRAINING_MODE` | `face` | `face` — upload crops to Frigate API; `object` — save crops to disk |
-| `STRATEGY` | `auto` | `auto` (adaptive), `standard` (30 images), `broad` (100 images) |
-| `LIMIT` | *(unset)* | Exact image count — overrides `STRATEGY` |
-| `OBJECT_CLASS` | `dog` | Target class for object mode (any YOLO class: `dog`, `cat`, `car`, etc.) |
-
-### People Filtering
-
-| Variable | Default | Description |
-| :--- | :--- | :--- |
-| `ONLY_PEOPLE` | *(unset)* | Comma-separated whitelist — only these people are processed |
-| `SKIP_PEOPLE` | *(unset)* | Comma-separated list of people to skip |
-| `MIN_FACE_COUNT` | `0` | Skip people with fewer than N tagged assets in Immich |
-| `YEARS_FILTER` | `10` | Ignore images older than N years |
-
 ### Connection

 | Variable | Default | Description |
 | :--- | :--- | :--- |
 | `IMMICH_URL` | *(required)* | Full URL to your Immich instance |
 | `API_KEY` | *(required)* | Immich API key |
-| `FRIGATE_URL` | *(unset)* | Frigate URL — required for face upload; omit to skip upload |
+| `FRIGATE_URL` | *(unset)* | Frigate URL — required for face upload; omit to skip |
+
+### Mode & Strategy
+
+| Variable | Default | Description |
+| :--- | :--- | :--- |
+| `TRAINING_MODE` | `face` | `face` — upload crops to Frigate; `object` — save crops to disk |
+| `STRATEGY` | `auto` | `auto` (adaptive), `standard` (30 images), `broad` (100 images) |
+| `LIMIT` | *(unset)* | Exact image count — overrides `STRATEGY` |
+| `OBJECT_CLASS` | `dog` | Target class for object mode (any YOLO class: `dog`, `cat`, `car`, etc.) |
+| `AUTO_MODE` | *(auto)* | Force non-interactive mode in a terminal; auto-detected otherwise |
+| `VERBOSE` | `false` | Enable DEBUG-level console output (log file is always DEBUG) |
+
+### People Filtering
+
+| Variable | Default | Description |
+| :--- | :--- | :--- |
+| `ONLY_PEOPLE` | *(unset)* | Comma-separated whitelist — process only these people |
+| `SKIP_PEOPLE` | *(unset)* | Comma-separated list — skip these people |
+| `MIN_FACE_COUNT` | `0` | Skip people with fewer than N tagged assets in Immich |
+| `YEARS_FILTER` | `10` | Ignore images older than N years |

 ### Image Quality

 | Variable | Default | Description |
 | :--- | :--- | :--- |
 | `MIN_FACE_WIDTH` | `50` | Minimum face crop width in pixels |
-| `FACE_MARGIN` | `0.15` | Padding added around the bounding box crop (fraction of face size) |
+| `FACE_MARGIN` | `0.15` | Padding around bounding box crop (fraction of face size) |
 | `ENABLE_FACE_ALIGNMENT` | `true` | Align to ArcFace 112×112 format using facial landmarks |
 | `USE_FULL_RESOLUTION` | `true` | Download full-resolution originals rather than preview thumbnails |
 | `MIN_CONFIDENCE` | `0.7` | Minimum Immich face detection confidence |
 | `BLUR_THRESHOLD` | `100.0` | Laplacian variance threshold — lower accepts more blur |
-| `MAX_AUTO_IMAGES` | `80` | Maximum images in auto-diversity mode |
+| `MAX_AUTO_IMAGES` | `80` | Maximum images selected in auto mode |

-### Caching & Models
+### GPU & Models

 | Variable | Default | Description |
 | :--- | :--- | :--- |
-| `FORCE_CPU` | `false` | Disable GPU — fall back to CPU for embedding computation |
+| `FORCE_CPU` | `false` | Disable GPU — fall back to CPU for all inference |
+| `OPENVINO_DEVICE` | `CPU` | Intel variant only: set `GPU` to use Arc or iGPU; default runs on CPU |
 | `ENABLE_CACHE` | `true` | Cache computed embeddings to disk (speeds up re-runs on the same library) |
 | `CACHE_DIR` | `.if_cache` | Path for embedding cache and upload tracker files |
-| `HF_HOME` | *(system)* | HuggingFace model cache location (SigLIP) |
-| `INSIGHTFACE_HOME` | *(system)* | InsightFace model cache location (Buffalo_L) |
+| `HF_HOME` | *(system)* | HuggingFace model cache path (SigLIP) |
+| `INSIGHTFACE_HOME` | *(system)* | InsightFace model cache path (Buffalo_L) |

 ### Tracker Overrides *(one-shot — remove after use)*

 | Variable | Default | Description |
 | :--- | :--- | :--- |
-| `DRY_RUN` | `false` | Show what would be selected and uploaded without doing it |
+| `DRY_RUN` | `false` | Preview selection without downloading or uploading |
 | `RETRY_REJECTED` | `false` | Re-attempt assets previously rejected by Frigate |
 | `RESET_PERSON` | *(unset)* | Clear upload and rejection history for one person by name |

@@ -226,7 +202,7 @@ The first run after a fresh install downloads the embedding models (~1-2 GB). Su

 | Variable | Default | Description |
 | :--- | :--- | :--- |
-| `CRON_SCHEDULE` | *(unset)* | Unset = run once and exit; empty = stay alive for manual `docker exec`; cron expression = scheduled |
+| `CRON_SCHEDULE` | *(unset)* | Unset = run once and exit; empty = stay alive; cron expression = scheduled |

 ---

@@ -239,16 +215,16 @@ uv sync
 uv run winnow
 ```

-Requires Python 3.13+ and [uv](https://astral.sh/uv/). An NVIDIA GPU is strongly recommended — CPU mode works but embedding computation is slower (typically tens of seconds per person vs under a second on GPU).
+Requires Python 3.13+ and [uv](https://astral.sh/uv). An NVIDIA, AMD, or Intel GPU is recommended — CPU mode works but embedding computation is slower.

 ---

 ## Requirements

 - **Immich** v1.106+
- **Frigate** v0.16+ (face mode only — object mode has no Frigate API dependency)
- **NVIDIA GPU** recommended (CUDA 12.x)
- **Python 3.13+**
+- **Frigate** v0.16+ (face mode only — object mode has no Frigate dependency)
+- **GPU** recommended: NVIDIA (CUDA), AMD (ROCm), or Intel (Arc / iGPU via OpenVINO)
+- **Python** 3.13+

 ---

@@ -38,6 +38,7 @@ services:

      # ── Caching & Models ──────────────────────────────────────────────────
      # - FORCE_CPU=true              # Disable GPU, fall back to CPU
+      # - OPENVINO_DEVICE=GPU         # Intel variant only: use Arc/iGPU instead of CPU (default: CPU)
      # - ENABLE_CACHE=false          # Disable embedding cache (default: true)
      - CACHE_DIR=/app/.if_cache
      - HF_HOME=/models/huggingface
@@ -63,6 +64,9 @@ services:
      - /path/to/winnow/cache:/app/.if_cache
      - /path/to/winnow/output:/app/frigate_train
    restart: unless-stopped
+
+    # ── GPU device passthrough ─────────────────────────────────────────────────
+    # NVIDIA (default — requires nvidia-container-toolkit):
    deploy:
      resources:
        reservations:
@@ -70,3 +74,18 @@ services:
            - driver: nvidia
              count: all
              capabilities: [gpu]
+    #
+    # AMD ROCm — replace the deploy block above with:
+    #   devices:
+    #     - /dev/kfd
+    #     - /dev/dri
+    #   group_add:
+    #     - video
+    #     - render
+    #
+    # Intel Arc / iGPU — replace the deploy block above with:
+    #   devices:
+    #     - /dev/dri
+    #   group_add:
+    #     - render
+    #   Also set: OPENVINO_DEVICE=GPU in the environment section above.
@@ -1,6 +1,6 @@
 [project]
 name = "winnow"
-version = "0.2.11"
+version = "0.2.12"
 description = "Immich to Frigate training sets"
 license = "AGPL-3.0-or-later"
 requires-python = ">=3.13"
@@ -0,0 +1,103 @@
+[project]
+name = "winnow"
+version = "0.2.12"
+description = "Immich to Frigate training sets"
+license = "AGPL-3.0-or-later"
+requires-python = ">=3.13"
+authors = [{ name = "Holden Salomon", email = "holden@arch.fyi" }]
+keywords = ["immich", "frigate", "face-recognition", "training-data", "arcface", "insightface"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: GNU Affero General Public License v3 or later (AGPLv3+)",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Scientific/Engineering :: Image Recognition",
+]
+dependencies = [
+    "croniter>=5.0.2",
+    "insightface>=0.7.3",
+    "numpy>=2.2.6",
+    "onnxruntime-openvino>=1.20.0",
+    "opencv-python-headless>=4.12.0.88",
+    "pillow>=12.1.0",
+    "python-dotenv>=1.2.1",
+    "requests>=2.32.5",
+    "rich>=14.2.0",
+    "torch>=2.12.0",
+    "torchvision>=0.27.0",
+    "transformers>=4.57.6",
+    "ultralytics>=8.4.66",
+]
+
+[project.scripts]
+winnow = "winnow.cli:main"
+
+[project.urls]
+Repository = "https://github.com/sudolulo/winnow"
+
+[tool.uv]
+conflicts = [
+    [
+        { package = "onnxruntime" },
+        { package = "onnxruntime-gpu" },
+        { package = "onnxruntime-openvino" },
+    ],
+]
+required-environments = [
+    "sys_platform == 'linux' and platform_machine == 'x86_64'",
+]
+
+[tool.uv.sources]
+torch = [
+    { index = "pytorch-cpu", marker = "sys_platform == 'linux' and platform_machine == 'x86_64'" },
+]
+torchvision = [
+    { index = "pytorch-cpu", marker = "sys_platform == 'linux' and platform_machine == 'x86_64'" },
+]
+
+[[tool.uv.index]]
+name = "pytorch-cpu"
+url = "https://download.pytorch.org/whl/cpu"
+explicit = true
+
+[dependency-groups]
+dev = [
+    "pytest>=8.0",
+    "ruff>=0.15.17",
+]
+
+[tool.hatch.build.targets.wheel]
+packages = ["winnow"]
+
+[tool.ruff]
+line-length = 120
+target-version = "py313"
+
+[tool.ruff.lint]
+select = ["E", "F", "I"]
+
+[tool.deptry]
+pep621_dev_dependency_groups = ["dev"]
+
+[tool.deptry.package_module_name_map]
+pillow = "PIL"
+opencv-python-headless = "cv2"
+python-dotenv = "dotenv"
+insightface = "insightface"
+numpy = "numpy"
+onnxruntime-openvino = "onnxruntime"
+requests = "requests"
+rich = "rich"
+torch = "torch"
+transformers = "transformers"
+ultralytics = "ultralytics"
+
+[tool.deptry.per_rule_ignores]
+DEP002 = ["onnxruntime-openvino"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
@@ -0,0 +1,103 @@
+[project]
+name = "winnow"
+version = "0.2.12"
+description = "Immich to Frigate training sets"
+license = "AGPL-3.0-or-later"
+requires-python = ">=3.13"
+authors = [{ name = "Holden Salomon", email = "holden@arch.fyi" }]
+keywords = ["immich", "frigate", "face-recognition", "training-data", "arcface", "insightface"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: GNU Affero General Public License v3 or later (AGPLv3+)",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Scientific/Engineering :: Image Recognition",
+]
+dependencies = [
+    "croniter>=5.0.2",
+    "insightface>=0.7.3",
+    "numpy>=2.2.6",
+    "onnxruntime-rocm>=1.16.0",
+    "opencv-python-headless>=4.12.0.88",
+    "pillow>=12.1.0",
+    "python-dotenv>=1.2.1",
+    "requests>=2.32.5",
+    "rich>=14.2.0",
+    "torch>=2.5.0",
+    "torchvision>=0.20.0",
+    "transformers>=4.57.6",
+    "ultralytics>=8.4.66",
+]
+
+[project.scripts]
+winnow = "winnow.cli:main"
+
+[project.urls]
+Repository = "https://github.com/sudolulo/winnow"
+
+[tool.uv]
+index-strategy = "unsafe-best-match"
+conflicts = [
+    [
+        { package = "onnxruntime" },
+        { package = "onnxruntime-gpu" },
+        { package = "onnxruntime-rocm" },
+    ],
+]
+required-environments = [
+    "sys_platform == 'linux' and platform_machine == 'x86_64'",
+]
+
+[tool.uv.sources]
+torch = [
+    { index = "pytorch-rocm63", marker = "sys_platform == 'linux' and platform_machine == 'x86_64'" },
+]
+torchvision = [
+    { index = "pytorch-rocm63", marker = "sys_platform == 'linux' and platform_machine == 'x86_64'" },
+]
+
+[[tool.uv.index]]
+name = "pytorch-rocm63"
+url = "https://download.pytorch.org/whl/rocm6.3"
+
+[dependency-groups]
+dev = [
+    "pytest>=8.0",
+    "ruff>=0.15.17",
+]
+
+[tool.hatch.build.targets.wheel]
+packages = ["winnow"]
+
+[tool.ruff]
+line-length = 120
+target-version = "py313"
+
+[tool.ruff.lint]
+select = ["E", "F", "I"]
+
+[tool.deptry]
+pep621_dev_dependency_groups = ["dev"]
+
+[tool.deptry.package_module_name_map]
+pillow = "PIL"
+opencv-python-headless = "cv2"
+python-dotenv = "dotenv"
+insightface = "insightface"
+numpy = "numpy"
+onnxruntime-rocm = "onnxruntime"
+requests = "requests"
+rich = "rich"
+torch = "torch"
+transformers = "transformers"
+ultralytics = "ultralytics"
+
+[tool.deptry.per_rule_ignores]
+DEP002 = ["onnxruntime-rocm"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
@@ -1,6 +1,6 @@
 [project]
 name = "winnow"
-version = "0.2.11"
+version = "0.2.12"
 description = "Immich to Frigate training sets"
 license = "AGPL-3.0-or-later"
 requires-python = ">=3.13"
@@ -266,7 +266,7 @@ wheels = [

 [package.optional-dependencies]
 cublas = [
-    { name = "nvidia-cublas-cu12", marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
+    { name = "nvidia-cublas-cu12", version = "12.6.4.1", source = { registry = "https://pypi.org/simple" }, marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
 ]
 cudart = [
    { name = "nvidia-cuda-runtime-cu12", marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
@@ -293,7 +293,7 @@ nvjitlink = [
    { name = "nvidia-nvjitlink-cu12", marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
 ]
 nvrtc = [
-    { name = "nvidia-cuda-nvrtc-cu12", marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
+    { name = "nvidia-cuda-nvrtc-cu12", version = "12.6.85", source = { registry = "https://pypi.org/simple" }, marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
 ]
 nvtx = [
    { name = "nvidia-nvtx-cu12", marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
@@ -482,7 +482,7 @@ wheels = [

 [[package]]
 name = "huggingface-hub"
-version = "1.18.0"
+version = "1.19.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "click" },
@@ -496,9 +496,9 @@ dependencies = [
    { name = "typer" },
    { name = "typing-extensions" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/fb/d8/748ea0a47f0fa15227fe682f7a80826b4b7c096e4818044b8f56d6cb66d6/huggingface_hub-1.18.0.tar.gz", hash = "sha256:f0c5ecd1ef8c6a60f86f61ee278f2c1570ba9e279c9f54de9094210723b3613b", size = 812699, upload-time = "2026-06-05T09:26:33.401Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/88/27/629cfe58c582f92ded066c4a07d1a057ff617118ab7973200f770bd853cb/huggingface_hub-1.19.0.tar.gz", hash = "sha256:fd771622182d40977272a923953ee3b1b13538f9f8a7f5d78398f10af0f1c0bd", size = 824721, upload-time = "2026-06-11T12:33:18.665Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/0b/03/40a05316cb6616e5b7efd7773656441ab04b4b022c2199e79bb4622a92a3/huggingface_hub-1.18.0-py3-none-any.whl", hash = "sha256:729be4a976fb706dcc02d176bcda8a3f32bdf21a294e8f4b3dda6fbcbc9c1ab1", size = 684411, upload-time = "2026-06-05T09:26:31.48Z" },
+    { url = "https://files.pythonhosted.org/packages/b2/a5/558da89f66464d8d0229ff497e8b8666977de2d8cf48c28a2862ecf1250f/huggingface_hub-1.19.0-py3-none-any.whl", hash = "sha256:1dc72e1f6b4d6df6b30eb72e57d00514ef453d660f04af2b87f0e67267f31ee0", size = 693398, upload-time = "2026-06-11T12:33:16.695Z" },
 ]

 [[package]]
@@ -708,7 +708,7 @@ wheels = [

 [[package]]
 name = "matplotlib"
-version = "3.10.9"
+version = "3.11.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "contourpy" },
@@ -721,36 +721,36 @@ dependencies = [
    { name = "pyparsing" },
    { name = "python-dateutil" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/63/1b/4be5be87d43d327a0cf4de1a56e86f7f84c89312452406cf122efe2839e6/matplotlib-3.10.9.tar.gz", hash = "sha256:fd66508e8c6877d98e586654b608a0456db8d7e8a546eb1e2600efd957302358", size = 34811233, upload-time = "2026-04-24T00:14:13.539Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/1f/24/080c99d223d158d3a8902769269ab6da5b50f7a0e6e072513907e02b7a6c/matplotlib-3.11.0.tar.gz", hash = "sha256:68c0c7be01b30dcca3638934f7f591df73401235cbdbf0d1ab1c71e7db7f8b57", size = 33251176, upload-time = "2026-06-12T02:29:15.508Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/aa/d3/8d4f6afbecb49fc04e060a57c0fce39ea51cc163a6bd87303ccd698e4fa6/matplotlib-3.10.9-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:b580440f1ff81a0e34122051a3dfabb7e4b7f9e380629929bde0eff9af72165f", size = 8320331, upload-time = "2026-04-24T00:12:39.688Z" },
-    { url = "https://files.pythonhosted.org/packages/63/d9/9e14bc7564bf92d5ffa801ae5fac819ce74b925dfb55e3ebde61a3bbad3e/matplotlib-3.10.9-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:b1b745c489cd1a77a0dc1120a05dc87af9798faebc913601feb8c73d89bf2d1e", size = 8216461, upload-time = "2026-04-24T00:12:42.494Z" },
-    { url = "https://files.pythonhosted.org/packages/8a/17/4402d0d14ccf1dfc70932600b68097fbbf9c898a4871d2cbbe79c7801a32/matplotlib-3.10.9-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:8f3bcac1ca5ed000a6f4337d47ba67dfddf37ed6a46c15fd7f014997f7bf865f", size = 8790091, upload-time = "2026-04-24T00:12:44.789Z" },
-    { url = "https://files.pythonhosted.org/packages/3e/0b/322aeec06dd9b91411f92028b37d447342770a24392aa4813e317064dad5/matplotlib-3.10.9-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:7a8d66a55def891c33147ba3ba9bfcabf0b526a43764c818acbb4525e5ed0838", size = 9605027, upload-time = "2026-04-24T00:12:47.583Z" },
-    { url = "https://files.pythonhosted.org/packages/74/88/5f13482f55e7b00bcfc09838b093c2456e1379978d2a146844aae05350ad/matplotlib-3.10.9-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:d843374407c4017a6403b59c6c81606773d136f3259d5b6da3131bc814542cc2", size = 9671269, upload-time = "2026-04-24T00:12:50.878Z" },
-    { url = "https://files.pythonhosted.org/packages/c5/e0/0840fd2f93da988ec660b8ad1984abe9f25d2aed22a5e394ff1c68c88307/matplotlib-3.10.9-cp313-cp313-win_amd64.whl", hash = "sha256:f4399f64b3e94cd500195490972ae1ee81170df1636fa15364d157d5bdd7b921", size = 8217588, upload-time = "2026-04-24T00:12:53.784Z" },
-    { url = "https://files.pythonhosted.org/packages/47/b9/d706d06dd605c49b9f83a2aed8c13e3e5db70697d7a80b7e3d7915de6b17/matplotlib-3.10.9-cp313-cp313-win_arm64.whl", hash = "sha256:ba7b3b8ef09eab7df0e86e9ae086faa433efbfbdb46afcb3aa16aabf779469a8", size = 8136913, upload-time = "2026-04-24T00:12:56.501Z" },
-    { url = "https://files.pythonhosted.org/packages/9b/45/6e32d96978264c8ca8c4b1010adb955a1a49cfaf314e212bbc8908f04a61/matplotlib-3.10.9-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:09218df8a93712bd6ea133e83a153c755448cf7868316c531cffcc43f69d1cc9", size = 8368019, upload-time = "2026-04-24T00:12:58.896Z" },
-    { url = "https://files.pythonhosted.org/packages/86/0a/c8e3d3bba245f0f7fc424937f8ff7ef77291a36af3edb97ccd78aa93d84f/matplotlib-3.10.9-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:82368699727bfb7b0182e1aa13082e3c08e092fa1a25d3e1fd92405bff96f6d4", size = 8264645, upload-time = "2026-04-24T00:13:01.406Z" },
-    { url = "https://files.pythonhosted.org/packages/3d/aa/5bf5a14fe4fed73a4209a155606f8096ff797aad89c6c35179026571133e/matplotlib-3.10.9-cp313-cp313t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:3225f4e1edcb8c86c884ddf79ebe20ecd0a67d30188f279897554ccd8fded4dc", size = 8802194, upload-time = "2026-04-24T00:13:03.702Z" },
-    { url = "https://files.pythonhosted.org/packages/dd/5e/b4be852d6bba6fd15893fadf91ff26ae49cb91aac789e95dde9d342e664f/matplotlib-3.10.9-cp313-cp313t-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:de2445a0c6690d21b7eb6ce071cebad6d40a2e9bdf10d039074a96ba19797b99", size = 9622684, upload-time = "2026-04-24T00:13:06.647Z" },
-    { url = "https://files.pythonhosted.org/packages/4c/3d/ed428c971139112ef730f62770654d609467346d09d4b62617e1afd68a5a/matplotlib-3.10.9-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:b2b9516251cb89ff618d757daec0e2ed1bf21248013844a853d87ef85ab3081d", size = 9680790, upload-time = "2026-04-24T00:13:10.009Z" },
-    { url = "https://files.pythonhosted.org/packages/e7/09/052e884aaf2b985c63cb79f715f1d5b6a3eaa7de78f6a52b9dbc077d5b53/matplotlib-3.10.9-cp313-cp313t-win_amd64.whl", hash = "sha256:e9fae004b941b23ff2edcf1567a857ed77bafc8086ffa258190462328434faf8", size = 8287571, upload-time = "2026-04-24T00:13:13.087Z" },
-    { url = "https://files.pythonhosted.org/packages/f4/38/ae27288e788c35a4250491422f3db7750366fc8c97d6f36fbdecfc1f5518/matplotlib-3.10.9-cp313-cp313t-win_arm64.whl", hash = "sha256:6b63d9c7c769b88ab81e10dc86e4e0607cf56817b9f9e6cf24b2a5f1693b8e38", size = 8188292, upload-time = "2026-04-24T00:13:15.546Z" },
-    { url = "https://files.pythonhosted.org/packages/d6/e6/3bd8afd04949f02eabc1c17115ea5255e19cacd4d06fc5abdde4eeb0052c/matplotlib-3.10.9-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:172db52c9e683f5d12eaf57f0f54834190e12581fe1cc2a19595a8f5acb4e77d", size = 8321276, upload-time = "2026-04-24T00:13:18.318Z" },
-    { url = "https://files.pythonhosted.org/packages/41/86/86231232fff41c9f8e4a1a7d7a597d349a02527109c3af7d618366122139/matplotlib-3.10.9-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:97e35e8d39ccc85859095e01a53847432ba9a53ddf7986f7a54a11b73d0e143f", size = 8218218, upload-time = "2026-04-24T00:13:20.974Z" },
-    { url = "https://files.pythonhosted.org/packages/85/8f/becc9722cafc64f5d2eb0b7c1bf5f585271c618a45dbd8fabeb021f898b6/matplotlib-3.10.9-cp314-cp314-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:aba1615dabe83188e19d4f75a253c6a08423e04c1425e64039f800050a69de6b", size = 9608145, upload-time = "2026-04-24T00:13:23.228Z" },
-    { url = "https://files.pythonhosted.org/packages/32/5d/f7e914f7d9325abff4057cee62c0fa70263683189f774473cbfb534cd13b/matplotlib-3.10.9-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:34cf8167e023ad956c15f36302911d5406bd99a9862c1a8499ea6f7c0e015dc2", size = 9885085, upload-time = "2026-04-24T00:13:25.849Z" },
-    { url = "https://files.pythonhosted.org/packages/a5/fd/fa69f2221534e80cc5772ac2b7d222011a2acafc2ec7216d5dd174c864ae/matplotlib-3.10.9-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:59476c6d29d612b8e9bb6ce8c5b631be6ba8f9e3a2421f22a02b192c7dd28716", size = 9672358, upload-time = "2026-04-24T00:13:28.906Z" },
-    { url = "https://files.pythonhosted.org/packages/ab/1a/5a4f747a8b271cbb024946d2dd3c913ab5032ba430626f8c3528ada96b4b/matplotlib-3.10.9-cp314-cp314-win_amd64.whl", hash = "sha256:336b9acc64d309063126edcdaca00db9373af3c476bb94388fe9c5a53ad13e6f", size = 8349970, upload-time = "2026-04-24T00:13:31.904Z" },
-    { url = "https://files.pythonhosted.org/packages/64/dc/95d60ecaefe30680a154b52ea96ab4b0dab547f1fd6aa12f5fb655e89cae/matplotlib-3.10.9-cp314-cp314-win_arm64.whl", hash = "sha256:2dc9477819ffd78ad12a20df1d9d6a6bd4fec6aaa9072681465fddca052f1456", size = 8272785, upload-time = "2026-04-24T00:13:34.511Z" },
-    { url = "https://files.pythonhosted.org/packages/70/a0/005d68bc8b8418300ce6591f18586910a8526806e2ab663933d9f20a41e9/matplotlib-3.10.9-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:da4e09638420548f31c354032a6250e473c68e5a4e96899b4844cf39ddea23fe", size = 8367999, upload-time = "2026-04-24T00:13:36.962Z" },
-    { url = "https://files.pythonhosted.org/packages/22/05/1236cc9290be70b2498af20ca348add76e3fffe7f67b477db5133a84f3ea/matplotlib-3.10.9-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:345f6f68ecc8da0ca56fad2ea08fde1a115eda530079eca185d50a7bc3e146c6", size = 8264543, upload-time = "2026-04-24T00:13:39.851Z" },
-    { url = "https://files.pythonhosted.org/packages/cd/c2/071f5a5ff6c5bd63aaaf2f45c811d9bf2ced94bde188d9e1a519e21d0cba/matplotlib-3.10.9-cp314-cp314t-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:4edcfbd8565339aa62f1cd4012f7180926fdbe71850f7b0d3c379c175cd6b66c", size = 9622800, upload-time = "2026-04-24T00:13:42.296Z" },
-    { url = "https://files.pythonhosted.org/packages/95/57/da7d1f10a85624b9e7db68e069dd94e58dc41dbf9463c5921632ecbe3661/matplotlib-3.10.9-cp314-cp314t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:6be157fe17fc37cb95ac1d7374cf717ce9259616edec911a78d9d26dae8522d4", size = 9888561, upload-time = "2026-04-24T00:13:45.026Z" },
-    { url = "https://files.pythonhosted.org/packages/67/b2/ef8d6bb59b0edb6c16c968b70f548aa13b54348972def5aa6ac85df67145/matplotlib-3.10.9-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:4e42042d54db34fda4e95a7bd3e5789c2a995d2dad3eb8850232ee534092fbbf", size = 9680884, upload-time = "2026-04-24T00:13:48.066Z" },
-    { url = "https://files.pythonhosted.org/packages/61/1c/d21bfeb9931881ebe96bcfcff27c7ae4b160ae0ec291a714c42641a56d75/matplotlib-3.10.9-cp314-cp314t-win_amd64.whl", hash = "sha256:c27df8b3848f32a83d1767566595e43cfaa4460380974da06f4279a7ec143c39", size = 8432333, upload-time = "2026-04-24T00:13:51.008Z" },
-    { url = "https://files.pythonhosted.org/packages/78/23/92493c3e6e1b635ccfff146f7b99e674808787915420373ac399283764c2/matplotlib-3.10.9-cp314-cp314t-win_arm64.whl", hash = "sha256:a49f1eadc84ca85fd72fa4e89e70e61bf86452df6f971af04b12c60761a0772c", size = 8324785, upload-time = "2026-04-24T00:13:53.633Z" },
+    { url = "https://files.pythonhosted.org/packages/55/41/aa47f156b061d14c98b906f76c428507397708ec63ff94f410ae1752b426/matplotlib-3.11.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:6ce3b839b34ae1f430b4616893a2945a2999debaa7e94e7e29a2a8bbf286f7b5", size = 9450532, upload-time = "2026-06-12T02:28:06.769Z" },
+    { url = "https://files.pythonhosted.org/packages/8c/4f/5a9eb0375e81413953febf8af7b012a6b6357f53438a15c4f5ad86c6bbb5/matplotlib-3.11.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:373db8f91214e8ccaf35ac833cc1dd59dd961e148bbd55dd027141591dde1313", size = 9279760, upload-time = "2026-06-12T02:28:09.152Z" },
+    { url = "https://files.pythonhosted.org/packages/a4/c0/1117d53077e3ac3152503a84e9cf7a5c239576805ee71276e80c2aaa7471/matplotlib-3.11.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:be152b7570324dc8d01574cc9474dd2d803237acf528bcbb5b211fa347461a09", size = 10031623, upload-time = "2026-06-12T02:28:11.26Z" },
+    { url = "https://files.pythonhosted.org/packages/92/7e/e937138daffad65b71bf831a377809dcbc830fb4f31a31e067dc1faa2575/matplotlib-3.11.0-cp313-cp313-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:126f256df600652d7e4b394cf3164ff75210a00038f287c95a012a6f58d0e83f", size = 10839372, upload-time = "2026-06-12T02:28:14.102Z" },
+    { url = "https://files.pythonhosted.org/packages/1d/c2/438ecc197ffb8023b6b9922915542f2172f5fd45b76703b0b4fc47322243/matplotlib-3.11.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:03acfeddf87b0dddb11b081ef7740ad445a3ca8bcb6b8e3011b08f2cf802b75c", size = 10924099, upload-time = "2026-06-12T02:28:16.383Z" },
+    { url = "https://files.pythonhosted.org/packages/40/2e/395883da416f378b3ed2c9f3e843ac477eae1ce731b671b79adaa6f0bacd/matplotlib-3.11.0-cp313-cp313-win_amd64.whl", hash = "sha256:ab3722f04f3ff34c23b5012c5873d2894174e06c3822fcdac3610965a5ac7d06", size = 9329727, upload-time = "2026-06-12T02:28:18.581Z" },
+    { url = "https://files.pythonhosted.org/packages/61/82/2c388956abf8bf392dfb5b8917c502f1082df6a941b781ab8c8e5ba2474b/matplotlib-3.11.0-cp313-cp313-win_arm64.whl", hash = "sha256:c945824670fb8915b4ac879e5e61f3c58e0913022f70a0de4c082b17372f8771", size = 9003506, upload-time = "2026-06-12T02:28:20.474Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/c1/34454baa44da7975ada82e9aea37105ec47059514dc967d3be14426ba8dc/matplotlib-3.11.0-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:3489c3dc487669b4a980bc3068f87856de7a1564248d3f6c629efb2a58b03f24", size = 9499838, upload-time = "2026-06-12T02:28:22.713Z" },
+    { url = "https://files.pythonhosted.org/packages/b1/c3/98fe79a398cf232219f090163a7fa7e6766e9f2e0ad26df54d6f8934d8ee/matplotlib-3.11.0-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:6a98f5476ce784a50ce09998f4ae1e6a9f25043cef8a480c98949902eda74620", size = 9332298, upload-time = "2026-06-12T02:28:24.796Z" },
+    { url = "https://files.pythonhosted.org/packages/95/e4/b4b7c33151e74e5c802f3cde1ba807ebfc38401e329b44e215a5888dd76d/matplotlib-3.11.0-cp313-cp313t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:565af866fd63e4bd3f987d580afe27c44c2552a3b3305f4ecbb85133601ea6f3", size = 10045491, upload-time = "2026-06-12T02:28:27.141Z" },
+    { url = "https://files.pythonhosted.org/packages/71/28/394548efd68354110c1a1be11fe6b6e559e06d1a23da35908a0e316c55a9/matplotlib-3.11.0-cp313-cp313t-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:e6b3e64dea5062c570f04358e2711859f3531b459f29516274fbad889079e4f3", size = 10857059, upload-time = "2026-06-12T02:28:29.222Z" },
+    { url = "https://files.pythonhosted.org/packages/c8/44/e7922e6e2a4d63bdfbc9dc4a53e3850ab438d46cf42e6779bb15ec92c948/matplotlib-3.11.0-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:942b37c5db1899610bd1543ce8e13e4ecff9a4633e7f63bb6aa9205d2644ebd1", size = 10939576, upload-time = "2026-06-12T02:28:31.66Z" },
+    { url = "https://files.pythonhosted.org/packages/3d/be/b1ca96003a441d619b727fee21d671fdff7a5ce2f1bb797b2521aa2f679a/matplotlib-3.11.0-cp313-cp313t-win_amd64.whl", hash = "sha256:c08e649a6313e1291e713623b97a38e5bb4aa580b2a100a94a3309bc6b9c8eb3", size = 9379519, upload-time = "2026-06-12T02:28:33.888Z" },
+    { url = "https://files.pythonhosted.org/packages/e3/72/4bf3b91821c34596dd6a7bdac5836d94f744144c8208939ef49d8ec43f7e/matplotlib-3.11.0-cp313-cp313t-win_arm64.whl", hash = "sha256:2746cd2c113742ff6ce37a864c5ac5fd7aa644568f445e66166e457ac78e40e0", size = 9055456, upload-time = "2026-06-12T02:28:35.878Z" },
+    { url = "https://files.pythonhosted.org/packages/57/52/a94102ac99eb78e2fe9b826674f9ef9ee23327110ea6ab4776c1b4eb6209/matplotlib-3.11.0-cp314-cp314-macosx_10_15_x86_64.whl", hash = "sha256:3338e3e3de128cf50d0d2fb92a122815daf9c755bd882a474343c05f8fd7ec79", size = 9452137, upload-time = "2026-06-12T02:28:37.93Z" },
+    { url = "https://files.pythonhosted.org/packages/7c/03/b8cdb625a21f710dfa11bbca1f48fb4057d2c0286975f8b415bf80942c99/matplotlib-3.11.0-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:25c2e5455efd8d99f41fb79871a31feb7d301569642e332ec58d72cfe9282bc3", size = 9281514, upload-time = "2026-06-12T02:28:40.028Z" },
+    { url = "https://files.pythonhosted.org/packages/b7/2d/4e1240ea82ee197dfb3851e71f71c87eeeb975f1753b56a0588e4e80739a/matplotlib-3.11.0-cp314-cp314-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:d9695457a467ff86d23f35037a43deb6f1134dd6d3e2ac8ce1e2087cff09ffb9", size = 10843005, upload-time = "2026-06-12T02:28:42.39Z" },
+    { url = "https://files.pythonhosted.org/packages/29/dc/6377ecfaa5fef79430f74a1a16638b4e2aa30d4692bae2c19f9d76fe3b01/matplotlib-3.11.0-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:19c16c61dea63b3582918503e6b294193961261d9daa806d4ae2151f1ad05430", size = 11127459, upload-time = "2026-06-12T02:28:44.483Z" },
+    { url = "https://files.pythonhosted.org/packages/6f/41/795c405aa7560443a3b01309424cde4a1113b85c90b8a63417444a749617/matplotlib-3.11.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:2d72ea8b7924f3cb955e61518d21e43b3df1e6c8a793b480a0c1214f185d30ba", size = 10925160, upload-time = "2026-06-12T02:28:46.564Z" },
+    { url = "https://files.pythonhosted.org/packages/1a/f7/3a9e6389a7cfaeff76c56e40c2dabcb13110e21e82f837228c834ebe748c/matplotlib-3.11.0-cp314-cp314-win_amd64.whl", hash = "sha256:1c02da0a629dfa9debf52725ea06866b74c1fb70a895bae05e4493d34074f9f2", size = 9485186, upload-time = "2026-06-12T02:28:49.344Z" },
+    { url = "https://files.pythonhosted.org/packages/8b/c0/396478ee7cf2091d182db8b4a8695f6a37f1ddb978989cf9dbb84cd5c123/matplotlib-3.11.0-cp314-cp314-win_arm64.whl", hash = "sha256:aa55d73b3117d4b07f959cd9eb6f69b375d8df3414139c479388e551aa5d999d", size = 9160349, upload-time = "2026-06-12T02:28:51.382Z" },
+    { url = "https://files.pythonhosted.org/packages/c5/6f/1c3bd51bb2b34eaacdcf3c3d859dbb357f952fc8020c617dc118ad7c9e38/matplotlib-3.11.0-cp314-cp314t-macosx_10_15_x86_64.whl", hash = "sha256:a9d8c6e7cd2f0ddf11d8d92e520dd1d9d2abb0cf6ac8831e338666c81e905847", size = 9500921, upload-time = "2026-06-12T02:28:53.443Z" },
+    { url = "https://files.pythonhosted.org/packages/e0/0d/4d861d0121840cb1a3fd4a10deb211efd6fccd481ed23e553f31f4f4da4a/matplotlib-3.11.0-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:be050fcf32f729eda99f7f75a80bf67612ce16ab9ac1c23a387dcaede95cb70e", size = 9332190, upload-time = "2026-06-12T02:28:55.623Z" },
+    { url = "https://files.pythonhosted.org/packages/4b/cb/22f6bc35711a0b5639a784e74e653e77c86210bd4304449dd399a482f74e/matplotlib-3.11.0-cp314-cp314t-manylinux_2_26_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:dfabef0230d0697aa0d717385194dd41162e00207a68bf4abf94c2bf4c27dca0", size = 10854181, upload-time = "2026-06-12T02:28:57.856Z" },
+    { url = "https://files.pythonhosted.org/packages/3f/7e/9a9eaca731a2939589da520f0ebe8fd8753d0f51fca98c7d20af6dbe261a/matplotlib-3.11.0-cp314-cp314t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:1644db30e759199443493ac5e5caec24fdb775a8f6123021f85ba47c4133c3cb", size = 11137715, upload-time = "2026-06-12T02:29:00.555Z" },
+    { url = "https://files.pythonhosted.org/packages/ef/f9/9b030b6088354acb0296871bb624b25befc1c42509d3c6cd17420c83a5b8/matplotlib-3.11.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:15b0d160079cb10699a0e98b5989c70677b2df7cacdc62af67c30f2facec46d9", size = 10939427, upload-time = "2026-06-12T02:29:02.527Z" },
+    { url = "https://files.pythonhosted.org/packages/59/94/6b273eaee4ee250863567d100865da61a5c1527fa67f527b7ed22e0dd29c/matplotlib-3.11.0-cp314-cp314t-win_amd64.whl", hash = "sha256:446307e6b04b57b1f1239e228a1ec2af0d589a1008cebc3dfa3f5441d095cfb6", size = 9535809, upload-time = "2026-06-12T02:29:04.994Z" },
+    { url = "https://files.pythonhosted.org/packages/60/95/1d36bddf2b7e2692c1540e78a6e5bc88bc1496b137e3e35a611f91b65ac3/matplotlib-3.11.0-cp314-cp314t-win_arm64.whl", hash = "sha256:652fb5696271d4c50f196d22a5ff4f8e4444c74f847423570d7dc0aa2bbd0159", size = 9209226, upload-time = "2026-06-12T02:29:07.033Z" },
 ]

 [[package]]
@@ -878,12 +878,39 @@ wheels = [
 name = "nvidia-cublas-cu12"
 version = "12.6.4.1"
 source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "platform_machine == 'x86_64' and sys_platform == 'linux'",
+]
 wheels = [
    { url = "https://files.pythonhosted.org/packages/af/eb/ff4b8c503fa1f1796679dce648854d58751982426e4e4b37d6fce49d259c/nvidia_cublas_cu12-12.6.4.1-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:08ed2686e9875d01b58e3cb379c6896df8e76c75e0d4a7f7dace3d7b6d9ef8eb", size = 393138322, upload-time = "2024-11-20T17:40:25.65Z" },
    { url = "https://files.pythonhosted.org/packages/97/0d/f1f0cadbf69d5b9ef2e4f744c9466cb0a850741d08350736dfdb4aa89569/nvidia_cublas_cu12-12.6.4.1-py3-none-manylinux_2_27_aarch64.whl", hash = "sha256:235f728d6e2a409eddf1df58d5b0921cf80cfa9e72b9f2775ccb7b4a87984668", size = 390794615, upload-time = "2024-11-20T17:39:52.715Z" },
    { url = "https://files.pythonhosted.org/packages/84/f7/985e9bdbe3e0ac9298fcc8cfa51a392862a46a0ffaccbbd56939b62a9c83/nvidia_cublas_cu12-12.6.4.1-py3-none-win_amd64.whl", hash = "sha256:9e4fa264f4d8a4eb0cdbd34beadc029f453b3bafae02401e999cf3d5a5af75f8", size = 434535301, upload-time = "2024-11-20T17:50:41.681Z" },
 ]

+[[package]]
+name = "nvidia-cublas-cu12"
+version = "12.9.2.10"
+source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "platform_machine != 'aarch64' and platform_machine != 's390x' and platform_machine != 'x86_64' and sys_platform == 'linux'",
+    "platform_machine == 's390x' and sys_platform == 'linux'",
+    "platform_machine != 's390x' and sys_platform == 'win32'",
+    "platform_machine == 's390x' and sys_platform == 'win32'",
+    "platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32'",
+    "platform_machine == 's390x' and sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32'",
+    "platform_machine != 's390x' and sys_platform == 'darwin'",
+    "platform_machine == 's390x' and sys_platform == 'darwin'",
+]
+dependencies = [
+    { name = "nvidia-cuda-nvrtc-cu12", version = "12.9.86", source = { registry = "https://pypi.org/simple" }, marker = "platform_machine != 'x86_64' or sys_platform != 'linux' or (extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f7/a2/c96163a0fff1839c0c9548bbdeae7b853b867009e33b9b9264adc238b1cf/nvidia_cublas_cu12-12.9.2.10-py3-none-manylinux_2_27_aarch64.whl", hash = "sha256:5572131a59c3eebeeb1c4c8144f772d49372c20124916e072a0e3fc30df421d5", size = 575012079, upload-time = "2026-04-08T18:51:47.303Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/c0/0a517bfe63ccd3b92eb254d264e28fca3c7cab75d07daea315250fb1bf73/nvidia_cublas_cu12-12.9.2.10-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:e4f53a8ca8c5d6e8c492d0d0a3d565ecb59a751b19cfdaa4f6da0ab2104c1702", size = 581240110, upload-time = "2026-04-08T18:52:31.532Z" },
+    { url = "https://files.pythonhosted.org/packages/20/e2/fc9a0e985249d873150276d5afb02e39a66817fedbf1a385724393e505ed/nvidia_cublas_cu12-12.9.2.10-py3-none-win_amd64.whl", hash = "sha256:623f43027d40d44ceadf0043f002bd25cf353e8f13ce90b9a87057019f560661", size = 553162896, upload-time = "2026-04-08T18:53:10.035Z" },
+]
+
 [[package]]
 name = "nvidia-cuda-cupti"
 version = "13.0.85"
@@ -920,12 +947,36 @@ wheels = [
 name = "nvidia-cuda-nvrtc-cu12"
 version = "12.6.85"
 source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "platform_machine == 'x86_64' and sys_platform == 'linux'",
+]
 wheels = [
    { url = "https://files.pythonhosted.org/packages/f5/31/ffb400c5ae99daf09687aa6c42831c5d824f71c4851363ed2a4a1ac52bab/nvidia_cuda_nvrtc_cu12-12.6.85-py3-none-manylinux2010_x86_64.manylinux_2_12_x86_64.whl", hash = "sha256:800927308ccc5dd6246d3f61f7fcef2ed7ec4e59e199090d360d3293f78bd5a2", size = 23649944, upload-time = "2024-11-20T17:38:06.511Z" },
    { url = "https://files.pythonhosted.org/packages/48/ab/476146f59ff5ef5bd6e62c187097d859ea78b5752d19c6c3f9be5f90dafc/nvidia_cuda_nvrtc_cu12-12.6.85-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:3f3134f50963882373063901657554f230bedf6039d30b09f6be55c64c993a37", size = 23162872, upload-time = "2024-11-20T17:37:42.967Z" },
    { url = "https://files.pythonhosted.org/packages/0c/f7/472414aee887d626373d0b2140a59ac4308e3eaed815060e5410fc83305a/nvidia_cuda_nvrtc_cu12-12.6.85-py3-none-win_amd64.whl", hash = "sha256:a419e2c95e75b88b602f8bb66f82a6c5651e8475a509841c958486b1b71510bf", size = 39026436, upload-time = "2024-11-20T17:49:13.633Z" },
 ]

+[[package]]
+name = "nvidia-cuda-nvrtc-cu12"
+version = "12.9.86"
+source = { registry = "https://pypi.org/simple" }
+resolution-markers = [
+    "platform_machine == 'aarch64' and sys_platform == 'linux'",
+    "platform_machine != 'aarch64' and platform_machine != 's390x' and platform_machine != 'x86_64' and sys_platform == 'linux'",
+    "platform_machine == 's390x' and sys_platform == 'linux'",
+    "platform_machine != 's390x' and sys_platform == 'win32'",
+    "platform_machine == 's390x' and sys_platform == 'win32'",
+    "platform_machine != 's390x' and sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32'",
+    "platform_machine == 's390x' and sys_platform != 'darwin' and sys_platform != 'linux' and sys_platform != 'win32'",
+    "platform_machine != 's390x' and sys_platform == 'darwin'",
+    "platform_machine == 's390x' and sys_platform == 'darwin'",
+]
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/b8/85/e4af82cc9202023862090bfca4ea827d533329e925c758f0cde964cb54b7/nvidia_cuda_nvrtc_cu12-12.9.86-py3-none-manylinux2010_x86_64.manylinux_2_12_x86_64.whl", hash = "sha256:210cf05005a447e29214e9ce50851e83fc5f4358df8b453155d5e1918094dcb4", size = 89568129, upload-time = "2025-06-05T20:02:41.973Z" },
+    { url = "https://files.pythonhosted.org/packages/64/eb/c2295044b8f3b3b08860e2f6a912b702fc92568a167259df5dddb78f325e/nvidia_cuda_nvrtc_cu12-12.9.86-py3-none-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:096d4de6bda726415dfaf3198d4f5c522b8e70139c97feef5cd2ca6d4cd9cead", size = 44528905, upload-time = "2025-06-05T20:02:29.754Z" },
+    { url = "https://files.pythonhosted.org/packages/52/de/823919be3b9d0ccbf1f784035423c5f18f4267fb0123558d58b813c6ec86/nvidia_cuda_nvrtc_cu12-12.9.86-py3-none-win_amd64.whl", hash = "sha256:72972ebdcf504d69462d3bcd67e7b81edd25d0fb85a2c46d3ea3517666636349", size = 76408187, upload-time = "2025-06-05T20:12:27.819Z" },
+]
+
 [[package]]
 name = "nvidia-cuda-runtime"
 version = "13.0.96"
@@ -956,7 +1007,7 @@ resolution-markers = [
    "platform_machine == 'x86_64' and sys_platform == 'linux'",
 ]
 dependencies = [
-    { name = "nvidia-cublas-cu12", marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
+    { name = "nvidia-cublas-cu12", version = "12.6.4.1", source = { registry = "https://pypi.org/simple" }, marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
 ]
 wheels = [
    { url = "https://files.pythonhosted.org/packages/fa/41/e79269ce215c857c935fd86bcfe91a451a584dfc27f1e068f568b9ad1ab7/nvidia_cudnn_cu12-9.10.2.21-py3-none-manylinux_2_27_aarch64.whl", hash = "sha256:c9132cc3f8958447b4910a1720036d9eff5928cc3179b0a51fb6d167c6cc87d8", size = 705026878, upload-time = "2025-06-06T21:52:51.348Z" },
@@ -980,7 +1031,7 @@ resolution-markers = [
    "platform_machine == 's390x' and sys_platform == 'darwin'",
 ]
 dependencies = [
-    { name = "nvidia-cublas-cu12", marker = "platform_machine != 'x86_64' or sys_platform != 'linux' or (extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
+    { name = "nvidia-cublas-cu12", version = "12.9.2.10", source = { registry = "https://pypi.org/simple" }, marker = "platform_machine != 'x86_64' or sys_platform != 'linux' or (extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
 ]
 wheels = [
    { url = "https://files.pythonhosted.org/packages/10/30/ecca1c8194c8077c4b57a3d96b56d96f15852551b01b919bae6429d92218/nvidia_cudnn_cu12-9.23.1.3-py3-none-manylinux_2_27_aarch64.whl", hash = "sha256:6dbc18f05aab2a323a4ffd43d985410608f7db7db9a8596e189cddbd3e527441", size = 778220760, upload-time = "2026-06-09T19:38:19.281Z" },
@@ -1089,7 +1140,7 @@ name = "nvidia-cusolver-cu12"
 version = "11.7.1.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "nvidia-cublas-cu12", marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
+    { name = "nvidia-cublas-cu12", version = "12.6.4.1", source = { registry = "https://pypi.org/simple" }, marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
    { name = "nvidia-cusparse-cu12", marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
    { name = "nvidia-nvjitlink-cu12", marker = "(platform_machine == 'x86_64' and sys_platform == 'linux') or (platform_machine != 'x86_64' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu') or (sys_platform != 'linux' and extra == 'project-11-onnxruntime' and extra == 'project-15-onnxruntime-gpu')" },
 ]
@@ -1454,17 +1505,17 @@ wheels = [

 [[package]]
 name = "protobuf"
-version = "7.35.0"
+version = "7.35.1"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/60/fd/5b1491d9e4b586d621c54f4c36b888714164b6875f8d6afa3f9072906a51/protobuf-7.35.0.tar.gz", hash = "sha256:a2efd84605f41e559f1881b0912b44099d0a2ac9bf46b3474823f10fb393b0e6", size = 458677, upload-time = "2026-05-19T23:02:29.197Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/da/01/9ef0afd7999eb9badb3a768b4aedd78c86d4c65cfaf1958ab276199e76b4/protobuf-7.35.1.tar.gz", hash = "sha256:ce115a26fe0c39a2c29973d914d327e516a6455464489fe3cd1e51a1b354f81a", size = 458717, upload-time = "2026-06-11T21:55:40.257Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/83/ee/93d06e358a4aa32280b00e722d3ea0a1f25fc3cc5778d80581c9cca2c10e/protobuf-7.35.0-cp310-abi3-macosx_10_9_universal2.whl", hash = "sha256:66be6c513931c794fa92c080ffee41671390da3d79da219cf9c0c0907f035dda", size = 433225, upload-time = "2026-05-19T23:02:19.884Z" },
-    { url = "https://files.pythonhosted.org/packages/8b/39/1c76c2da93f3c507e958e0aecee2391cc44d4625de6c728bbc555195b5a8/protobuf-7.35.0-cp310-abi3-manylinux2014_aarch64.whl", hash = "sha256:fcbe42a4ac09d3ec9c987ddfcd956afd0b15f1ff613bd8371bde9405ffd5c8e5", size = 328847, upload-time = "2026-05-19T23:02:22.3Z" },
-    { url = "https://files.pythonhosted.org/packages/91/1a/39f7ce90a238c1a987a4d81ec26379e02ca0aff367de68e4a1fa474215b9/protobuf-7.35.0-cp310-abi3-manylinux2014_s390x.whl", hash = "sha256:4cbf5cc286130e06a6c9bbefac442431173906dfcc979712183d4adcc01b37ee", size = 344030, upload-time = "2026-05-19T23:02:23.591Z" },
-    { url = "https://files.pythonhosted.org/packages/70/5b/6baf9008817964454055ff3fe65f1de0b5f1e26c80c82f7fb108b7cd4ea3/protobuf-7.35.0-cp310-abi3-manylinux2014_x86_64.whl", hash = "sha256:6c0f98f10c8a05ea30f8993dfef2de093d27b490fdae78bb60c8343795d55011", size = 327130, upload-time = "2026-05-19T23:02:24.637Z" },
-    { url = "https://files.pythonhosted.org/packages/8e/e5/e46adb0badc388bfb84877a5f9f026aff63f60e611016cf64dbe77e05446/protobuf-7.35.0-cp310-abi3-win32.whl", hash = "sha256:4c4617b83ade0e279d1d2bfe04025a1adb87f9ed657de038620dc0ff959357f6", size = 428946, upload-time = "2026-05-19T23:02:25.741Z" },
-    { url = "https://files.pythonhosted.org/packages/a7/ab/547fbd9e16d879dd13c167478f8ae0a83a428008ca07a5e06acdc23ad473/protobuf-7.35.0-cp310-abi3-win_amd64.whl", hash = "sha256:f05bcadf9a2a6b8dda047007075135fb7d08c73d9177aabc067e1be46881a201", size = 439996, upload-time = "2026-05-19T23:02:26.808Z" },
-    { url = "https://files.pythonhosted.org/packages/b8/ef/50433d346c56657a70d27f156c7b349ac59a068b01de4eb796e747eecc43/protobuf-7.35.0-py3-none-any.whl", hash = "sha256:c13f325cf242bad135c350629eeb5d54b24228eb472fb3e2e9ebbd4c5dc20ca0", size = 171659, upload-time = "2026-05-19T23:02:27.842Z" },
+    { url = "https://files.pythonhosted.org/packages/10/03/8aeeb7458d22546bf64b5250ca1daeb5ff757d900e8e4a7476c6f0db843e/protobuf-7.35.1-cp310-abi3-macosx_10_9_universal2.whl", hash = "sha256:24f857477359a85c0c235261b8ba905fd51b2562f4a64ca1df5473f29850cbf6", size = 433226, upload-time = "2026-06-11T21:55:31.719Z" },
+    { url = "https://files.pythonhosted.org/packages/37/4b/dfb89eb0e652a1ff073c39a59fb5e3a83cfe9b57a2c83fa6d78270101767/protobuf-7.35.1-cp310-abi3-manylinux2014_aarch64.whl", hash = "sha256:11d6b0ec246892d85215b0a13ca6e0233cf5284b68f0ac02646427f4ff88a799", size = 328847, upload-time = "2026-06-11T21:55:34.035Z" },
+    { url = "https://files.pythonhosted.org/packages/0f/58/dc12f2cd484951524af6e3382c785869b9b3fb5e52ee95ae23add53ee8f9/protobuf-7.35.1-cp310-abi3-manylinux2014_s390x.whl", hash = "sha256:b73f9489a4b8b1c9cb1f8ed951c736392592edb24b9d6819f36d2e10b171d5b4", size = 344030, upload-time = "2026-06-11T21:55:34.941Z" },
+    { url = "https://files.pythonhosted.org/packages/e4/be/5b3cfe508bfab6761414ff944e3366eb13be4fd71efcd69450f89ba39f43/protobuf-7.35.1-cp310-abi3-manylinux2014_x86_64.whl", hash = "sha256:74758715c53d7158fb76caf4f0cfdacc5329a4b1bb994f865d6cf302d413a1c4", size = 327130, upload-time = "2026-06-11T21:55:35.921Z" },
+    { url = "https://files.pythonhosted.org/packages/d8/bc/6d6c7ba8709c85f8f2c390b2b118d6fb08a783676a572271851bf45a7d22/protobuf-7.35.1-cp310-abi3-win32.whl", hash = "sha256:353652e4efd0bca5b5fc2656abf8307ef351f0cf938c9eba09f0e09c20a25c30", size = 428945, upload-time = "2026-06-11T21:55:37.034Z" },
+    { url = "https://files.pythonhosted.org/packages/0a/19/8d0cb6f20a1ef7b18f1c8986ad5783f22f84cce39c6ce9a6e645ea55192e/protobuf-7.35.1-cp310-abi3-win_amd64.whl", hash = "sha256:230a75ddfc2de4806e56696ce9640c1cdfdb6543b7cfce98d42a4c0a0e7bdb87", size = 439996, upload-time = "2026-06-11T21:55:38.123Z" },
+    { url = "https://files.pythonhosted.org/packages/19/c7/5f7c636ec43e0c545e28d1f1db71990108306f7bdcb89f069ba97e428e7f/protobuf-7.35.1-py3-none-any.whl", hash = "sha256:4bc97768d8fe4ad6743c8a19403e314511ed9f6d13205b687e52421c023ac1b9", size = 171659, upload-time = "2026-06-11T21:55:39.155Z" },
 ]

 [[package]]
@@ -2183,7 +2234,7 @@ wheels = [

 [[package]]
 name = "transformers"
-version = "5.11.0"
+version = "5.12.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "huggingface-hub" },
@@ -2196,9 +2247,9 @@ dependencies = [
    { name = "tqdm" },
    { name = "typer" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/7d/4a/2ee05f9a06bb2dd34cac951548553738a826762fb7adfa7ce256daeaeac8/transformers-5.11.0.tar.gz", hash = "sha256:1dbecade5b8a09bdf3e9e8fdfc9d312cb9eccf5a201080dc894b373f0f3eb5f4", size = 8874363, upload-time = "2026-06-10T16:31:52.441Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/0c/f9/4552e2ba55db1c943aea0d4c09a32e9cbe5445b9eabe9856900de503dc8f/transformers-5.12.0.tar.gz", hash = "sha256:f0cf42ae1464c2eb41e7e0e66d7fd4b66145f48af17093b4cc0b2e9781faa7f4", size = 8923020, upload-time = "2026-06-12T14:39:20.43Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/9b/40/30fdefda9b8aff8d36054dbb37d8e1464cc1130c29623403c3c2147bdee1/transformers-5.11.0-py3-none-any.whl", hash = "sha256:06d0a34eab529955f8e704b530663c7d6bd6f9c7a9e47a05cddb4a44e7b6566b", size = 11094597, upload-time = "2026-06-10T16:31:49.08Z" },
+    { url = "https://files.pythonhosted.org/packages/ff/1f/d385913c38e900d23b728a4188fee625f2fccb306aeebc59be9d91404a5a/transformers-5.12.0-py3-none-any.whl", hash = "sha256:500be9eb644ede81c3103eee7687fc36d05dd75d1c76686c3820b26396fe7c7c", size = 11150246, upload-time = "2026-06-12T14:39:17.009Z" },
 ]

 [[package]]
@@ -2297,7 +2348,7 @@ wheels = [

 [[package]]
 name = "winnow"
-version = "0.2.11"
+version = "0.2.12"
 source = { editable = "." }
 dependencies = [
    { name = "croniter" },
@@ -85,15 +85,16 @@ def get_insightface_app():
        return _insightface_app
    _insightface_loaded = True

-    # Preload CUDA/cuDNN DLLs BEFORE any ORT InferenceSession is created
-    _preload_cuda_libs()
-
    ctx_id = -1
    insightface_home = os.environ.get("INSIGHTFACE_HOME", os.path.expanduser("~/.insightface"))
    try:
        import onnxruntime as ort
        from insightface.app import FaceAnalysis

+        # Preload CUDA/cuDNN DLLs before any ORT InferenceSession is created.
+        # Silently no-ops on ROCm/Intel builds where preload_dlls() is absent.
+        _preload_cuda_libs()
+
        # Disk cache check — lets the user know whether a download is coming
        buffalo_path = Path(insightface_home) / "models" / "buffalo_l"
        if buffalo_path.exists() and any(buffalo_path.iterdir()):
@@ -110,18 +111,35 @@ def get_insightface_app():
            "ROCmExecutionProvider",
            "MPSExecutionProvider",
            "CoreMLExecutionProvider",
+            "OpenVINOExecutionProvider",
        }
        has_gpu_provider = bool(gpu_providers & set(providers))
        ctx_id = -1 if _is_force_cpu() else (0 if has_gpu_provider else -1)

+        # For OpenVINO EP, inject device_type from env var (default CPU; set GPU for Intel Arc/iGPU)
+        has_openvino = "OpenVINOExecutionProvider" in providers
+        if has_openvino:
+            openvino_device = os.getenv("OPENVINO_DEVICE", "CPU")
+            providers = [
+                ("OpenVINOExecutionProvider", {"device_type": openvino_device})
+                if p == "OpenVINOExecutionProvider" else p
+                for p in providers
+            ]
+            logger.debug(f"OpenVINO EP: device_type={openvino_device}")
+
        if not has_gpu_provider and not _is_force_cpu():
            logger.warning(
                "No GPU execution provider found — running InsightFace on CPU. "
-                "If you have an NVIDIA GPU, ensure the NVIDIA Container Toolkit is "
-                "installed and the container has GPU access (deploy.resources in compose)."
+                "Ensure the container has GPU access and the correct variant image is used "
+                "(gpu for NVIDIA, rocm for AMD, intel for Intel Arc/iGPU)."
            )

-        device_str = "GPU" if ctx_id >= 0 else "CPU"
+        if ctx_id < 0:
+            device_str = "CPU"
+        elif has_openvino:
+            device_str = f"OpenVINO ({os.getenv('OPENVINO_DEVICE', 'CPU')})"
+        else:
+            device_str = "GPU"
        logger.info(f"InsightFace Buffalo_L: loading into memory on {device_str}...")

        t0 = time.time()
@@ -223,11 +241,14 @@ def get_siglip_model():

        _siglip_model.eval()

-        # Move to GPU if available
+        # Move to GPU if available (ROCm builds expose torch.cuda.is_available() == True)
        if not _is_force_cpu():
            if torch.cuda.is_available():
                _siglip_model = _siglip_model.cuda()
                device_name = "CUDA GPU"
+            elif hasattr(torch, "xpu") and torch.xpu.is_available():
+                _siglip_model = _siglip_model.to("xpu")
+                device_name = "Intel XPU"
            elif hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
                _siglip_model = _siglip_model.to("mps")
                device_name = "Apple MPS"
@@ -144,7 +144,14 @@ def process_object_mode(
    try:
        model = get_yolo_model()
        target_class = config.get("object_class", "dog")
-        device = "cpu" if os.getenv("FORCE_CPU", "").lower() in ("true", "1", "yes") else None
+        import torch
+
+        if os.getenv("FORCE_CPU", "").lower() in ("true", "1", "yes"):
+            device = "cpu"
+        elif hasattr(torch, "xpu") and torch.xpu.is_available():
+            device = "xpu"
+        else:
+            device = None  # YOLO auto-selects (CUDA/ROCm/CPU)

        results = model(img, verbose=False, device=device)

@@ -58,8 +58,11 @@ def _get_strategy_choice(has_embedding: bool, entity_type: str) -> tuple[int | s
    rprint("  [bold]4.[/bold] Skip")

    choice = Prompt.ask("Choice", choices=["1", "2", "3", "4"], default="1")
-    limits = {"1": 30, "2": 100, "3": IntPrompt.ask("Enter number of images", default=30)}
-    return limits.get(choice, 0), "time" if choice != "4" else "skip"
+    if choice == "4":
+        return 0, "skip"
+    if choice == "3":
+        return IntPrompt.ask("Enter number of images", default=30), "time"
+    return {"1": 30, "2": 100}.get(choice, 30), "time"


 def _resolve_strategy(strategy: str, has_embedding: bool) -> tuple[int | str, str]: