modeld_v2: emu

bump opendbc
modeld_v2: artifact parity with upstream tg (#2093 )
2026-11-07 07:43:44 +08:00 · 2026-11-06 15:18:45 -08:00 · 2026-11-06 16:03:02 -05:00 · 2026-11-05 16:11:27 -05:00 · 2026-11-05 15:35:07 -05:00 · 2026-11-04 14:32:06 -08:00
23 changed files with 1062 additions and 500 deletions
@@ -34,6 +34,11 @@ on:
        required: false
        type: string
        default: 'qcom'
+      on_device:
+        description: 'Compile on the tici / chestnut runner instead of on a hosted runner with tinygrad''s qcom / amd mock'
+        required: false
+        type: boolean
+        default: false
      hf_repo:
        description: 'Hugging Face dataset repository (e.g. sunnypilot/sunnypilot_models_v1)'
        required: false
@@ -107,6 +112,11 @@ on:
        options:
          - qcom
          - chestnut
+      on_device:
+        description: 'Compile on the tici / chestnut runner instead of on a hosted runner with tinygrad''s qcom / amd mock'
+        required: false
+        type: boolean
+        default: false
      hf_repo:
        description: 'Hugging Face dataset repository'
        required: false
@@ -130,11 +140,45 @@ jobs:
      is_20hz: ${{ inputs.is_20hz }}
      artifact_suffix: ${{ inputs.artifact_suffix }}
      target_hardware: ${{ inputs.target_hardware }}
+      on_device: ${{ inputs.on_device || false }}
    secrets: inherit

-  publish_model:
+  mock_parity:
    if: ${{ !cancelled() && needs.build_model.result == 'success' }}
    needs: build_model
+    runs-on: ubuntu-24.04
+    permissions:
+      actions: read
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+      - uses: ./.github/workflows/setup-amd-mock
+        if: inputs.target_hardware == 'chestnut'
+      - run: python3 -m pip install -q numpy ${{ inputs.target_hardware != 'chestnut' && 'tinymesa==25.2.7.2' || '' }}
+      - name: Download artifact name file
+        uses: actions/download-artifact@v4
+        with:
+          name: artifact-name-${{ inputs.custom_name || inputs.upstream_branch }}
+          path: artifact_name
+      - name: Download the built model
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          ARTIFACT_NAME=$(cat artifact_name/artifact_name.txt)
+          id=$(gh api -X GET "repos/$GITHUB_REPOSITORY/actions/runs/$GITHUB_RUN_ID/artifacts" -f name="$ARTIFACT_NAME" --jq '.artifacts[0].id')
+          url=$(curl -sf -o /dev/null -w '%{redirect_url}' -H "Authorization: Bearer $GH_TOKEN" "https://api.github.com/repos/$GITHUB_REPOSITORY/actions/artifacts/$id/zip")
+          sudo apt-get install -y -qq aria2 > /dev/null
+          aria2c -q -x16 -s16 -k8M --max-tries=5 --file-allocation=none -o model.zip "$url" && unzip -q model.zip -d output && rm model.zip
+      - name: Replay its check frame on the mock and require the hardware's result bit for bit
+        env:
+          PYTHONPATH: ${{ github.workspace }}/tinygrad_repo:${{ github.workspace }}
+          DEV: ${{ inputs.target_hardware == 'chestnut' && 'MOCKUSB+AMD' || 'MOCK+QCOM' }}
+        run: python3 -m openpilot.sunnypilot.modeld_v2.mock_parity check output/${{ inputs.target_hardware == 'chestnut' && 'big_' || '' }}driving_tinygrad.pkl
+
+  publish_model:
+    if: ${{ !cancelled() && needs.build_model.result == 'success' && contains(fromJSON('["success", "skipped"]'), needs.mock_parity.result) }}
+    needs: [build_model, mock_parity]
    runs-on: ubuntu-latest
    permissions:
      actions: read
@@ -0,0 +1,33 @@
+name: Chestnut mock build parity
+description: >-
+  Convert one commaai precompiled chestnut model (LS2) on this runner under tinygrad's amd mock, as a hosted chestnut build does, and require the
+  exact bytes the chestnut runner built (the published artifact). Needs setup-amd-mock, numpy and zstandard; tinygrad_repo checked out.
+runs:
+  using: composite
+  steps:
+    - name: Rebuild LS2 on the mock and require the chestnut runner's artifact byte for byte
+      shell: bash
+      env:
+        REF: b92fc4828664f8767bce5e92f1630ed319cbe509  # commaai/openpilot: precompiled big_driving_tinygrad.pkl, pins tinygrad ec91b74b5b
+        PKL_OID: b4dead3ee091d66ab002a6b8e716c26f4f79e3c931e4be5f48e017ba92f6e960
+        PKL_SIZE: "767228907"
+        TG_SHA: ec91b74b5b6b0fc924b72d3f27689679ca2438b9
+        WANT: 680a8d629785f2b1d1a2b0e40d0dfbcd0f7291c21e20c51259f1d768a22ec217  # driving_light_speed_2_november_04_2026_tinygrad.pkl, recompiled46
+      run: |
+        set -euo pipefail
+        work="$RUNNER_TEMP/chestnut_parity" && rm -rf "$work" && mkdir -p "$work"
+        href=$(curl -fsSL --retry 3 -X POST -H "Accept: application/vnd.git-lfs+json" -H "Content-Type: application/vnd.git-lfs+json" \
+          https://huggingface.co/commaai/openpilot-lfs.git/info/lfs/objects/batch \
+          -d "{\"operation\":\"download\",\"transfer\":[\"basic\"],\"objects\":[{\"oid\":\"$PKL_OID\",\"size\":$PKL_SIZE}]}" | jq -r '.objects[0].actions.download.href')
+        curl -fsSL --retry 3 -o "$work/big_driving_tinygrad.pkl" "$href"
+        echo "$PKL_OID  $work/big_driving_tinygrad.pkl" | sha256sum -c -
+        tg="$work/tinygrad" && git init -q "$tg" && git -C "$tg" fetch -q --depth 1 https://github.com/tinygrad/tinygrad "$TG_SHA" && git -C "$tg" checkout -q FETCH_HEAD
+        rm -rf "$tg/test/mockgpu" && cp -r "$GITHUB_WORKSPACE/tinygrad_repo/test/mockgpu" "$tg/test/mockgpu"
+        LLVM_PATH="${LLVM_PATH:-$(ls /usr/lib/llvm-20/lib/libLLVM.so* | head -1)}" DEV=MOCKUSB+AMD CHECK_OOB=0 GMMU=0 PYTHONPATH="$tg:$GITHUB_WORKSPACE" \
+          python3 openpilot/sunnypilot/modeld_v2/convert_upstream_pkl.py "$work/big_driving_tinygrad.pkl" "$work/LS2.pkl"
+        got=$(sha256sum "$work/LS2.pkl" | cut -d' ' -f1)
+        if [ "$got" != "$WANT" ]; then
+          echo "::error::the mock's LS2 build ($got) is not the chestnut runner's ($WANT): this runner's LLVM or the mock no longer builds what the chestnut does"
+          exit 1
+        fi
+        echo "LS2 rebuilt on the mock: byte-identical to the chestnut runner's build"
@@ -0,0 +1,18 @@
+name: Set up tinygrad's amd mock
+description: LLVM 20 and ROCm comgr, as tinygrad_repo/.github/actions/setup-tinygrad installs them for the mock gpu
+runs:
+  using: composite
+  steps:
+    - name: Install LLVM 20 and ROCm comgr
+      shell: bash
+      run: |
+        . /etc/os-release
+        curl -fsSL https://apt.llvm.org/llvm-snapshot.gpg.key | sudo tee /etc/apt/trusted.gpg.d/apt.llvm.org.asc > /dev/null
+        echo "deb http://apt.llvm.org/$VERSION_CODENAME/ llvm-toolchain-$VERSION_CODENAME-20 main" | sudo tee /etc/apt/sources.list.d/llvm.list
+        sudo mkdir -p /etc/apt/keyrings
+        curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null
+        echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/7.1 $VERSION_CODENAME main" | sudo tee /etc/apt/sources.list.d/rocm.list
+        printf 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600\n' | sudo tee /etc/apt/preferences.d/rocm-pin-600
+        sudo apt-get -qq update && sudo apt-get -qq install -y --no-install-recommends libllvm20 clang-20 lld-20 comgr
+        printf '/opt/rocm/lib\n/opt/rocm/lib64\n' | sudo tee /etc/ld.so.conf.d/rocm.conf && sudo ldconfig
+        echo "/usr/lib/llvm-20/bin" >> "$GITHUB_PATH"
@@ -35,6 +35,11 @@ on:
        required: false
        type: string
        default: 'qcom'
+      on_device:
+        description: 'Compile on the tici / chestnut runner instead of on a hosted runner with tinygrad''s qcom / amd mock'
+        required: false
+        type: boolean
+        default: false
  workflow_dispatch:
    inputs:
      upstream_branch:
@@ -59,6 +64,11 @@ on:
          - qcom
          - chestnut
        default: 'qcom'
+      on_device:
+        description: 'Compile on the tici / chestnut runner instead of on a hosted runner with tinygrad''s qcom / amd mock'
+        required: false
+        type: boolean
+        default: false


 run-name: Build model [${{ inputs.custom_name || inputs.upstream_branch }}] from ref [${{ inputs.upstream_branch }}]
@@ -153,18 +163,48 @@ jobs:
          if-no-files-found: error

  build_model:
-    runs-on: [self-hosted, "${{ inputs.target_hardware == 'chestnut' && 'chestnut' || 'tici' }}"]
+    runs-on: ${{ inputs.on_device && fromJSON(inputs.target_hardware == 'chestnut' && '["self-hosted","chestnut"]' || '["self-hosted","tici"]') || 'ubuntu-24.04' }}
    needs: get_model
    env:
      MODEL_NAME: ${{ inputs.custom_name || inputs.upstream_branch }} (${{ needs.get_model.outputs.model_date }})
      REF: ${{ inputs.upstream_branch }}
+      MOCK_BUILD: ${{ !inputs.on_device }}
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 1
          submodules: recursive

+      - name: Set up the hosted mock build
+        if: env.MOCK_BUILD == 'true'
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+
+      - name: Install LLVM 20 and ROCm comgr for the amd mock
+        if: env.MOCK_BUILD == 'true' && inputs.target_hardware == 'chestnut'
+        uses: ./.github/workflows/setup-amd-mock
+
+      - name: Install the amd mock build dependencies
+        if: env.MOCK_BUILD == 'true' && inputs.target_hardware == 'chestnut'
+        run: |
+          python3 -m pip install -q numpy zstandard
+          echo "LLVM_PATH=$(ls /usr/lib/llvm-20/lib/libLLVM.so* | head -1)" >> "$GITHUB_ENV"  # the chestnut runner's LLVM, never a newer one
+
+      - name: Check this runner builds what the chestnut runner does
+        if: env.MOCK_BUILD == 'true' && inputs.target_hardware == 'chestnut'
+        uses: ./.github/workflows/chestnut-build-parity
+
+      - name: Install the qcom mock build dependencies and check the compiler
+        if: env.MOCK_BUILD == 'true' && inputs.target_hardware != 'chestnut'
+        run: |
+          sudo apt-get -qq update && sudo apt-get -qq install -y --no-install-recommends qemu-user-static aria2
+          python3 -m pip install -q numpy zstandard tinymesa==25.2.7.2
+          # every kernel an a630 has compiled must compile to the same bytes here before a model is built here
+          PYTHONPATH="$TINYGRAD_PATH" python3 "$TINYGRAD_PATH/test/device/qcom/compiler_parity.py" check "$TINYGRAD_PATH/test/mockgpu/qcom/a630_compiled.pkl.xz"
+
      - name: Set environment variables
+        if: env.MOCK_BUILD != 'true'
        id: set-env
        run: |
          # Set up common environment
@@ -177,6 +217,7 @@ jobs:
          printenv >> $GITHUB_ENV

      - name: Setup build environment
+        if: env.MOCK_BUILD != 'true'
        run: |
          mkdir -p "${BUILD_DIR}/"
          sudo find $BUILD_DIR/ -mindepth 1 -delete
@@ -198,21 +239,28 @@ jobs:

      - name: Build Model
        run: |
-          source /etc/profile
-          export UV_PROJECT_ENVIRONMENT=${HOME}/venv
-          export VIRTUAL_ENV=$UV_PROJECT_ENVIRONMENT
-          source ${UV_PROJECT_ENVIRONMENT}/bin/activate
+          if [ "$MOCK_BUILD" = "true" ]; then
+            TASKSET=""
+            export CHECK_OOB=0 GMMU=0  # as test/mockgpu runs the amd mock
+          else
+            source /etc/profile
+            export UV_PROJECT_ENVIRONMENT=${HOME}/venv
+            export VIRTUAL_ENV=$UV_PROJECT_ENVIRONMENT
+            source ${UV_PROJECT_ENVIRONMENT}/bin/activate
+            TASKSET="taskset -c 7"
+          fi
          export PYTHONPATH="${PYTHONPATH}:${{ env.TINYGRAD_PATH }}:${{ github.workspace }}"

          COMPILE_MODELD="${{ github.workspace }}/openpilot/sunnypilot/modeld_v2/compile_modeld.py"
          MODEL_SIZE=$(python3 -c "from openpilot.common.transformations.model import MEDMODEL_INPUT_SIZE as s; print(f'{s[0]}x{s[1]}')")
          CAMERA_RES=$(python3 -c "from openpilot.common.transformations.camera import _ar_ox_fisheye as a, _os_fisheye as o; print(f'{a.width}x{a.height} {o.width}x{o.height}')")

-          TG_FLAGS_QCOM="DEV=QCOM IMAGE=1 FLOAT16=1 NOLOCALS=1 OPENPILOT_HACKS=1"
+          MOCK=$([ "$MOCK_BUILD" = "true" ] && echo MOCK || true)
+          TG_FLAGS_QCOM="DEV=${MOCK}QCOM IMAGE=1 FLOAT16=1 NOLOCALS=1 OPENPILOT_HACKS=1"
          if [ "${{ inputs.target_hardware }}" == "chestnut" ]; then
            echo "CHESTNUT build"
            export CHESTNUT=1
-            TG_FLAGS="DEBUG=1 DEV=USB+AMD:LLVM FRAME_DEV=CPU FLOAT16=1 GMMU=0 TC_OPT=2 TC_MIN_GLOBALS=32"
+            TG_FLAGS="DEBUG=1 DEV=${MOCK}USB+AMD:LLVM FRAME_DEV=CPU FLOAT16=1 GMMU=0 TC_OPT=2 TC_MIN_GLOBALS=32"
            OUTPUT_PKL="${{ env.MODELS_DIR }}/big_driving_tinygrad.pkl"
          else
            echo "QCOM build"
@@ -227,14 +275,17 @@ jobs:
            echo "Converting commaai's precompiled model with tinygrad $TG_SHA"
            UPSTREAM_TG="${{ runner.temp }}/upstream_tinygrad"
            rm -rf "$UPSTREAM_TG" && git init -q "$UPSTREAM_TG" && git -C "$UPSTREAM_TG" fetch -q --depth 1 "$TG_URL" "$TG_SHA" && git -C "$UPSTREAM_TG" checkout -q FETCH_HEAD
+            if [ "$MOCK_BUILD" = "true" ]; then  # convert under commaai's tinygrad, on our mock: it is the one measured against a chestnut
+              rm -rf "$UPSTREAM_TG/test/mockgpu" && cp -r "$TINYGRAD_PATH/test/mockgpu" "$UPSTREAM_TG/test/mockgpu"
+            fi
            CONVERTED="${{ runner.temp }}/big_driving_tinygrad_plain.pkl"
-            env DEV=USB+AMD PYTHONPATH="$UPSTREAM_TG:${{ github.workspace }}" taskset -c 7 python3 openpilot/sunnypilot/modeld_v2/convert_upstream_pkl.py \
+            env DEV=${MOCK}USB+AMD PYTHONPATH="$UPSTREAM_TG:${{ github.workspace }}" $TASKSET python3 openpilot/sunnypilot/modeld_v2/convert_upstream_pkl.py \
              "${{ env.MODELS_DIR }}/big_driving_tinygrad.pkl" "$CONVERTED"
            mv "$CONVERTED" "$OUTPUT_PKL"
          elif NEW_ARCH_ONNX=$(find "${{ env.MODELS_DIR }}" -maxdepth 1 -name "${{ inputs.target_hardware == 'chestnut' && 'big_' || '' }}*driving*.onnx" -print -quit) &&
               [ -n "$NEW_ARCH_ONNX" ] && grep -qa new_img "$NEW_ARCH_ONNX"; then # the new architecture: one ONNX taking the warped frames as new_img
            echo "Compiling new architecture ONNX: $NEW_ARCH_ONNX"
-            env ${TG_FLAGS} taskset -c 7 python3 openpilot/sunnypilot/modeld_v2/compile_modeld_v2.py \
+            env ${TG_FLAGS} $TASKSET python3 openpilot/sunnypilot/modeld_v2/compile_modeld_v2.py \
              "$NEW_ARCH_ONNX" "$OUTPUT_PKL" --device-input "*" --benchmark-runs 1
          else
            echo "Detected onnx architecture. Routing to native compile_modeld.py."
@@ -284,7 +335,7 @@ jobs:

            if [ -n "$MODEL_TYPE" ]; then
              echo "Detected: $MODEL_TYPE -> $OUTPUT_PKL"
-              env ${TG_FLAGS} taskset -c 7 python3 "$COMPILE_MODELD" \
+              env ${TG_FLAGS} $TASKSET python3 "$COMPILE_MODELD" \
                --model-type $MODEL_TYPE \
                --model-size $MODEL_SIZE \
                --camera-resolutions $CAMERA_RES \
@@ -304,7 +355,7 @@ jobs:
            --include='/${{ inputs.target_hardware == 'chestnut' && 'big_' || '' }}driving_tinygrad.pkl' \
            --exclude='*' \
            --delete-excluded \
-            --chown=comma:comma \
+            $([ "$MOCK_BUILD" = "true" ] || echo --chown=comma:comma) \
            ${{ env.MODELS_DIR }}/ ${{ env.OUTPUT_DIR }}/

          python3 "${{ github.workspace }}/release/ci/model_generator.py" \
@@ -333,7 +384,7 @@ jobs:
          path: ${{ env.OUTPUT_DIR }}/artifact_name.txt

      - name: Re-enable powersave
-        if: always()
+        if: always() && env.MOCK_BUILD != 'true'
        run: |
          source ${UV_PROJECT_ENVIRONMENT}/bin/activate
          PYTHONPATH=$PYTHONPATH:${{ github.workspace }}/ ${{ github.workspace }}/scripts/manage-powersave.py --enable
@@ -28,24 +28,30 @@ jobs:
        env:
          DEV: CPU
        run: tools/op.sh test -v openpilot/sunnypilot/modeld_v2/tests/test_plain_artifact.py
-      - name: Install LLVM 20 and ROCm comgr for the amd mock, as tinygrad_repo/.github/actions/setup-tinygrad does
-        run: |
-          . /etc/os-release
-          curl -fsSL https://apt.llvm.org/llvm-snapshot.gpg.key | sudo tee /etc/apt/trusted.gpg.d/apt.llvm.org.asc > /dev/null
-          echo "deb http://apt.llvm.org/$VERSION_CODENAME/ llvm-toolchain-$VERSION_CODENAME-20 main" | sudo tee /etc/apt/sources.list.d/llvm.list
-          sudo mkdir -p /etc/apt/keyrings
-          curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor | sudo tee /etc/apt/keyrings/rocm.gpg > /dev/null
-          echo "deb [arch=amd64 signed-by=/etc/apt/keyrings/rocm.gpg] https://repo.radeon.com/rocm/apt/7.1 $VERSION_CODENAME main" | sudo tee /etc/apt/sources.list.d/rocm.list
-          printf 'Package: *\nPin: release o=repo.radeon.com\nPin-Priority: 600\n' | sudo tee /etc/apt/preferences.d/rocm-pin-600
-          sudo apt-get -qq update && sudo apt-get -qq install -y --no-install-recommends libllvm20 clang-20 lld-20 comgr
-          printf '/opt/rocm/lib\n/opt/rocm/lib64\n' | sudo tee /etc/ld.so.conf.d/rocm.conf && sudo ldconfig
-          echo "/usr/lib/llvm-20/bin" >> "$GITHUB_PATH"
+      - uses: ./.github/workflows/setup-amd-mock
      - name: Convert a pickled USB+AMD jit and replay it under tinygrad's amd mock
        env:
          DEV: MOCKUSB+AMD
          CHECK_OOB: 0
          GMMU: 0
        run: tools/op.sh test -v openpilot/sunnypilot/modeld_v2/tests/test_convert_upstream_pkl.py
+      - name: Resolve commaai's tinygrad pin, which their precompiled chestnut model is built and converted with
+        id: upstream-tg
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          read -r sha url < <(gh api repos/commaai/openpilot/contents/tinygrad_repo --jq '.sha + " " + .submodule_git_url')
+          echo "commaai/openpilot master pins $url@$sha, ours is $(git rev-parse HEAD:tinygrad_repo)"
+          dir="$RUNNER_TEMP/upstream_tinygrad"
+          git init -q "$dir" && git -C "$dir" fetch -q --depth 1 "$url" "$sha" && git -C "$dir" checkout -q FETCH_HEAD
+          echo "path=$dir" >> "$GITHUB_OUTPUT"
+      - name: Convert a pickled USB+AMD jit under commaai's tinygrad, as the chestnut model build does
+        env:
+          DEV: MOCKUSB+AMD
+          CHECK_OOB: 0
+          GMMU: 0
+          PYTHONPATH: ${{ steps.upstream-tg.outputs.path }}:${{ github.workspace }}
+        run: tools/op.sh test -v openpilot/sunnypilot/modeld_v2/tests/test_convert_upstream_pkl.py
      - name: Where tinygrad moved, for the helper or failure point named above
        if: failure() && github.event_name == 'pull_request'
        run: |
@@ -63,6 +69,8 @@ jobs:
    runs-on: ubuntu-latest
    outputs:
      models: ${{ steps.set-matrix.outputs.models }}
+      chestnut: ${{ steps.set-matrix.outputs.chestnut }}
+      qcom: ${{ steps.set-matrix.outputs.qcom }}
    steps:
      - uses: actions/checkout@v4
      - name: Fetch and Parse json
@@ -74,16 +82,22 @@ jobs:
          with open("openpilot/sunnypilot/models/fetcher.py", "r") as f:
            urls = re.findall(r"MODEL_URL(?:_CHESTNUT)?\s*=\s*[\"'"'"']([^\"'"'"']+)[\"'"'"']", f.read())

-          artifacts = []
+          artifacts, chestnut, qcom = [], [], []
          for url in urls:
            data = json.loads(urllib.request.urlopen(url).read())
            for bundle in data.get("bundles", []):
              for model in bundle.get("models", []):
                if "artifact" in model:
                  artifacts.append(model["artifact"])
+                  if "chestnut" in url:
+                    chestnut.append(model["artifact"])
+                  elif "tinygrad" in model["artifact"]["file_name"]:
+                    qcom.append(model["artifact"])

          with open(os.environ["GITHUB_OUTPUT"], "a") as f:
            f.write(f"models={json.dumps(artifacts)}\n")
+            f.write(f"chestnut={json.dumps(chestnut)}\n")
+            f.write(f"qcom={json.dumps(qcom)}\n")
          '

  test-model:
@@ -120,3 +134,74 @@ jobs:
          export PYTHONPATH="$GITHUB_WORKSPACE/tinygrad_repo:$GITHUB_WORKSPACE:$PYTHONPATH"
          python3 -m pip uninstall -y tinygrad || uv pip uninstall -y tinygrad --system || true
          python3 -m pytest openpilot/sunnypilot/modeld_v2/tests/test_models.py
+
+  chestnut-mock-parity:
+    name: Mock parity ${{ matrix.artifact.file_name }}
+    needs: generate-matrix
+    runs-on: ubuntu-24.04
+    strategy:
+      fail-fast: false
+      matrix:
+        artifact: ${{ fromJson(needs.generate-matrix.outputs.chestnut) }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+      - uses: ./.github/workflows/setup-amd-mock
+      - run: python3 -m pip install -q numpy
+      - name: Download model
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+          MODEL_URL: ${{ matrix.artifact.download_uri.url }}
+          MODEL_SHA: ${{ matrix.artifact.download_uri.sha256 }}
+        shell: bash
+        run: |
+          AUTH=()
+          if [ -n "$HF_TOKEN" ]; then AUTH=(-H "Authorization: Bearer $HF_TOKEN"); fi
+          curl --retry 3 --retry-all-errors -s -S -f -L "${AUTH[@]}" -o "$RUNNER_TEMP/model.pkl" "$MODEL_URL"
+          if [ -n "$MODEL_SHA" ]; then echo "$MODEL_SHA  $RUNNER_TEMP/model.pkl" | sha256sum -c -; fi
+      - name: Replay its check frame on the amd mock and require the chestnut's result bit for bit
+        env:
+          PYTHONPATH: ${{ github.workspace }}/tinygrad_repo:${{ github.workspace }}
+        run: python3 -m openpilot.sunnypilot.modeld_v2.mock_parity check "$RUNNER_TEMP/model.pkl"
+
+  qcom-mock-parity:
+    name: QCOM mock parity ${{ matrix.artifact.file_name }}
+    needs: generate-matrix
+    runs-on: ubuntu-24.04
+    strategy:
+      fail-fast: false
+      matrix:
+        artifact: ${{ fromJson(needs.generate-matrix.outputs.qcom) }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+      - run: python3 -m pip install -q numpy tinymesa==25.2.7.2
+      - name: Download model
+        env:
+          HF_TOKEN: ${{ secrets.HF_TOKEN }}
+          MODEL_URL: ${{ matrix.artifact.download_uri.url }}
+          MODEL_SHA: ${{ matrix.artifact.download_uri.sha256 }}
+        shell: bash
+        run: |
+          AUTH=()
+          if [ -n "$HF_TOKEN" ]; then AUTH=(-H "Authorization: Bearer $HF_TOKEN"); fi
+          curl --retry 3 --retry-all-errors -s -S -f -L "${AUTH[@]}" -o "$RUNNER_TEMP/model.pkl" "$MODEL_URL"
+          if [ -n "$MODEL_SHA" ]; then echo "$MODEL_SHA  $RUNNER_TEMP/model.pkl" | sha256sum -c -; fi
+      - name: Replay its check frame on the qcom mock and require the comma device's result bit for bit
+        env:
+          DEV: MOCK+QCOM
+          PYTHONPATH: ${{ github.workspace }}/tinygrad_repo:${{ github.workspace }}
+        run: python3 -m openpilot.sunnypilot.modeld_v2.mock_parity check "$RUNNER_TEMP/model.pkl"
+
+  chestnut-mock-build-parity:
+    name: Chestnut build on the amd mock = chestnut runner build
+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+      - uses: ./.github/workflows/setup-amd-mock
+      - run: python3 -m pip install -q numpy zstandard
+      - uses: ./.github/workflows/chestnut-build-parity
@@ -149,6 +149,7 @@ inline static std::unordered_map<std::string, ParamKeyAttributes> keys = {
    {"Version", {PERSISTENT, STRING}},

    // --- sunnypilot params --- //
+    {"AccelBoostSP", {PERSISTENT | BACKUP, BOOL, "1"}},
    {"ApiCache_DriveStats", {PERSISTENT, JSON}},
    {"AutoLaneChangeBsmDelay", {PERSISTENT | BACKUP, BOOL, "0"}},
    {"AutoLaneChangeTimer", {PERSISTENT | BACKUP, INT, "0"}},
@@ -57,6 +57,12 @@ class CruiseLayout(Widget):
      description=tr("Use map data to estimate the appropriate speed to drive through turns ahead."),
      param="SmartCruiseControlMap")

+    self.accel_boost_toggle = toggle_item_sp(
+      title=tr("Acceleration Boost"),
+      description=tr("Adds a small push on top of sunnypilot's own acceleration, stronger in Aggressive and gentler in Relaxed. " +
+                     "Eases off in curves and when sunnypilot slows or stops. Never goes past your set speed, and lead-car following still comes first."),
+      param="AccelBoostSP")
+
    self.custom_acc_toggle = toggle_item_sp(
      title=tr("Custom ACC Speed Increments"),
      description="",
@@ -86,6 +92,7 @@ class CruiseLayout(Widget):
      self.icbm_toggle,
      self.scc_v_toggle,
      self.scc_m_toggle,
+      self.accel_boost_toggle,
      self.custom_acc_toggle,
      self.custom_acc_short_increment,
      self.custom_acc_long_increment,
@@ -136,6 +143,8 @@ class CruiseLayout(Widget):
          self.icbm_toggle.set_description(new_desc)
          self.icbm_toggle.show_description(True)

+      self.accel_boost_toggle.action_item.set_enabled(has_long)
+
      if has_long or has_icbm:
        self.custom_acc_toggle.action_item.set_enabled(((has_long and not ui_state.CP.pcmCruise) or has_icbm) and ui_state.is_offroad())
        self.scc_v_toggle.action_item.set_enabled(True)
@@ -151,6 +160,7 @@ class CruiseLayout(Widget):
    else:
      has_icbm = has_long = False
      self.icbm_toggle.action_item.set_enabled(False)
+      self.accel_boost_toggle.action_item.set_enabled(False)
      self.icbm_toggle.set_description(tr(ONROAD_ONLY_DESCRIPTION))

    show_custom_acc_desc = False
@@ -371,7 +371,7 @@ def _compile_warp_resolution_worker(cam_w, cam_h, model_w, model_h, benchmark_ru


 if __name__ == "__main__":
-  if 'USB' in os.getenv('DEV', '') or os.getenv('CHESTNUT'):
+  if ('USB' in os.getenv('DEV', '') or os.getenv('CHESTNUT')) and not os.getenv('DEV', '').startswith('MOCK'):
    from openpilot.system.hardware.chestnut.flash import link_up
    for _ in range(10):
      if link_up():
@@ -66,47 +66,75 @@ def _patched(buf, topo) -> tuple[bytearray, dict]:
  return blob, rows


-def _evaluate(u, env):
+def _packed_args(program) -> dict[int, int]:
+  from tinygrad.device import TinyELF
+  params = getattr(program.arg, "params", None) or ()
+  return {off: p.arg.slot for (off, _), p in zip(TinyELF.iter_sig(tuple((None, p.addrspace, p.dtype, ()) for p in params)), params, strict=True)}
+
+
+def _outer_slot(u, packed: dict[int, int]) -> int | None:
+  from tinygrad.uop.ops import Ops
+  if u.op is Ops.PARAM:
+    return None if packed else u.arg.slot
+  if u.op.name == "REFPARAM":
+    return u.arg.slot
+  if packed and u.op is Ops.LOAD and u.src[0].op is Ops.INDEX and _root(u.src[0].src[0]).op is Ops.PARAM:
+    index = _evaluate(u.src[0].src[1], {}, packed)
+    return None if index is None else packed.get(index * u.dtype.itemsize)
+  return None
+
+
+def _evaluate(u, env, packed: dict[int, int]):
  from tinygrad.uop.ops import GroupOp, Ops, exec_alu
  if u.op in (Ops.AFTER, Ops.CAST, Ops.BITCAST):
-    return _evaluate(u.src[0], env)
+    return _evaluate(u.src[0], env, packed)
  if u.op is Ops.CONST:
    return u.arg
-  if u.op in (Ops.PARAM, Ops.RANGE):
-    return env.get(u.arg.slot if u.op is Ops.PARAM else u)
+  if u.op is Ops.RANGE:
+    return env.get(u)
+  if (slot := _outer_slot(u, packed)) is not None:
+    return env.get(slot)
  if u.op is Ops.LOAD:
-    table, index = _root(u.src[0].src[0]), _evaluate(u.src[0].src[1], env)
+    table, index = _root(u.src[0].src[0]), _evaluate(u.src[0].src[1], env, packed)
    return struct.unpack_from(u.dtype.fmt, table.arg, index * u.dtype.itemsize)[0] if table.op is Ops.BINARY and index is not None else None
-  if u.op in GroupOp.ALU and None not in (values := [_evaluate(x, env) for x in u.src]):
+  if u.op in GroupOp.ALU and None not in (values := [_evaluate(x, env, packed) for x in u.src]):
    return exec_alu(u.op, u.dtype, values, *([False] if "truncate_output" in inspect.signature(exec_alu).parameters else []))
  return None


 def _input_pokes(submit, kernargs) -> dict[int, tuple[int, int]]:
  from tinygrad.uop.ops import Ops
-  outer = dict(enumerate(submit.src[1:]))
+  outer, packed = dict(enumerate(submit.src[1:])), _packed_args(submit.src[0])
  kernargs_params = {i for i, a in outer.items() if a.op is Ops.GETADDR and _root(a.src[0]) is kernargs}
  inputs = {i: _root(a.src[0]).arg.slot for i, a in outer.items() if a.op is Ops.GETADDR and _root(a.src[0]).op is Ops.PARAM}
  pokes = {}
  for call in (u for u in submit.src[0].src[0].toposort() if u.op is Ops.CALL and len(u.src) > 3):
    _, address, value, *_ = call.src[1:]
-    used = list({x.arg.slot for x in value.toposort() if x.op is Ops.PARAM} & inputs.keys())
-    if len(used) != 1 or not kernargs_params & {x.arg.slot for x in address.toposort() if x.op is Ops.PARAM}:
+    used = list({slot for x in value.toposort() if (slot := _outer_slot(x, packed)) is not None} & inputs.keys())
+    if len(used) != 1 or not kernargs_params & {_outer_slot(x, packed) for x in address.toposort()}:
      continue
    ranges = list(dict.fromkeys(x for x in (*address.toposort(), *value.toposort()) if x.op is Ops.RANGE))
    for trip in itertools.product(*[range(int(r.vmax) + 1) for r in ranges]):
      env = dict.fromkeys(outer, 0) | dict(zip(ranges, trip, strict=True))
-      if (offset := _evaluate(address, env)) % 8 == 0:
-        pokes[offset] = (inputs[used[0]], _evaluate(value, env))
+      if (offset := _evaluate(address, env, packed)) % 8 == 0:
+        pokes[offset] = (inputs[used[0]], _evaluate(value, env, packed))
  return pokes


-def _elf(image: bytes, rodata: int) -> bytes:
-  names, offset, data, headers = b"\0.text\0.rodata\0.shstrtab\0", 64, b"", []
-  for name, kind, address, content in [(0, 0, 0, b""), (1, 1, ALIGNMENT, image[:rodata]), (7, 1, ALIGNMENT + rodata, image[rodata:]), (15, 3, 0, names)]:
-    headers.append(struct.pack("<IIQQQQIIQQ", name, kind, 2 if kind == 1 else 0, address, offset, len(content), 0, 0, 1, 0))
+def _elf(image: bytes, rodata: int, entry: int) -> bytes:
+  spans = [(".text", 0, rodata), (".rodata", rodata, len(image))] if entry < rodata else \
+          [(".data", 0, rodata), (".rodata", rodata, entry), (".text", entry, len(image))]
+  sections, names = [(0, 0, 0, 0, b"")], b"\0"
+  for name, start, end in (span for span in spans if span[1] < span[2]):
+    sections.append((len(names), 1, 6 if name == ".text" else 2, ALIGNMENT + start, image[start:end]))
+    names += name.encode() + b"\0"
+  sections.append((len(names), 3, 0, 0, names + b".shstrtab\0"))
+  offset, data, headers = 64, b"", []
+  for name, kind, flags, address, content in sections:
+    headers.append(struct.pack("<IIQQQQIIQQ", name, kind, flags, address, offset, len(content), 0, 0, 1, 0))
    data, offset = data + content, offset + len(content)
-  return b"\x7fELF\x02\x01\x01" + bytes(9) + struct.pack("<HHIQQQIHHHHHH", 1, 0xe0, 1, 0, 0, offset, 0, 64, 0, 0, 64, 4, 3) + data + b"".join(headers)
+  header = struct.pack("<HHIQQQIHHHHHH", 1, 0xe0, 1, 0, 0, offset, 0, 64, 0, 0, 64, len(sections), len(sections) - 1)
+  return b"\x7fELF\x02\x01\x01" + bytes(9) + header + data + b"".join(headers)


 def convert(jit, kwargs: dict, frame: dict, target) -> tuple[dict, memoryview]:
@@ -155,7 +183,7 @@ def convert(jit, kwargs: dict, frame: dict, target) -> tuple[dict, memoryview]:
    if image_alloc not in programs:
      image = bytes(_patched(image_alloc, topo)[0])
      rodata = next(r for r in range(0, len(image) - 24, 64) if r + struct.unpack_from("<q", image, r + 16)[0] == entry)
-      programs[image_alloc] = blob.add(_elf(image, rodata))
+      programs[image_alloc] = blob.add(_elf(image, rodata, entry))
    start, size = _view(arguments.src[0])
    args = []
    for slot in range(start, start + unwrap(size), 8):
@@ -166,7 +194,8 @@ def convert(jit, kwargs: dict, frame: dict, target) -> tuple[dict, memoryview]:
        base, at, nbytes = _address(argument_words[slot])
        args.append({"buffer": buffers.setdefault(base, len(buffers)), "offset": at, "nbytes": nbytes or base.max_numel() * base.dtype.itemsize, "dtype": u8})
    outs, ins = kernel[6]
-    calls.append({"kind": "program", "name": re.sub(r"\x1b\[[0-9;]*m", "", kernel[1]), "lib": programs[image_alloc], "target": dataclasses.asdict(target),
+    device_target = dataclasses.asdict(dataclasses.replace(target, interface=target.interface.removeprefix("MOCK")))
+    calls.append({"kind": "program", "name": re.sub(r"\x1b\[[0-9;]*m", "", kernel[1]), "lib": programs[image_alloc], "target": device_target,
                  "signature": [[f"data{i}", i, u8, [a.get("nbytes", 1)]] for i, a in enumerate(args)], "global_size": grid, "local_size": local,
                  "globals": list(range(len(args))), "outs": list(outs), "ins": list(ins), "args": args})
    needs_init |= {args[i]["buffer"] for i in ins if "buffer" in args[i] and args[i]["buffer"] not in written}
@@ -264,8 +264,8 @@ def _program_uop(call: dict, lib: bytes):
    parameter = UOp.param(call["globals"][compact_slot], _dtype(dtype), math.prod(shape))
    image = (shape[0], shape[1]) if len(shape) == 3 and shape[2] == 4 else None
    parameters.append(parameter.replace(arg=_fields(parameter.arg, name=name, image=image)))
-  info = _fields(ProgramInfo, global_size=tuple(call["global_size"]), local_size=tuple(call["local_size"]), globals=tuple(call["globals"]),
-                 outs=tuple(call["outs"]), ins=tuple(call["ins"]), target=_fields(Target, **call["target"]))
+  info = _fields(ProgramInfo, global_size=tuple(call["global_size"]), local_size=tuple(call["local_size"]), params=tuple(parameters),
+                 globals=tuple(call["globals"]), outs=tuple(call["outs"]), ins=tuple(call["ins"]), target=_fields(Target, **call["target"]))
  return UOp(Ops.PROGRAM, src=(UOp(Ops.SINK, arg=KernelInfo(name=call["name"])), UOp(Ops.LINEAR, src=tuple(parameters)),
                               UOp(Ops.SOURCE, arg=""), UOp(Ops.BINARY, arg=lib)), arg=info)

@@ -0,0 +1,152 @@
+import json
+import os
+import shutil
+import struct
+import sys
+import tempfile
+import time
+
+os.environ.setdefault("DEV", "MOCKUSB+AMD")
+QCOM = "QCOM" in os.environ["DEV"]
+ON_MOCK = os.environ["DEV"].startswith("MOCK")
+ARCH, HARDWARE = ("a630", "comma 3x/4 (a630)") if QCOM else ("gfx1200", "chestnut")
+os.environ.setdefault("CHECK_OOB", "0")
+os.environ.setdefault("GMMU", "0")
+if sys.platform == "darwin" and os.environ["DEV"].startswith("MOCKUSB"):  # the mock's VRAM is a memfd, which macOS lacks: an unlinked file is the same
+  from tinygrad.runtime.autogen import libc
+
+  def _memfd_create(name, flags):
+    fd, path = tempfile.mkstemp()
+    os.unlink(path)
+    return fd
+  libc.memfd_create = _memfd_create
+
+import numpy as np
+
+from openpilot.sunnypilot.modeld_v2.helpers import ALIGNMENT, MAGIC, PlainJit, _device, read_parts
+
+
+class ParityError(RuntimeError):
+  pass
+
+
+def _emulator():
+  try:
+    if QCOM:
+      from test.mockgpu.qcom import emu
+    else:
+      from test.mockgpu.amd import emu
+  except ImportError as e:
+    raise ParityError(f"tinygrad's {'qcom' if QCOM else 'amd'} mock is not importable: put tinygrad_repo on PYTHONPATH") from e
+  if not hasattr(emu, "hw_unverified"):
+    raise ParityError(f"this tinygrad's emulator has no {ARCH} rounding model, its check frames would not match a {HARDWARE}")
+  return emu
+
+
+def unverified(header: dict, blob: memoryview) -> dict[str, int]:
+  emu, out = _emulator(), {}
+  for call in header["calls"]:
+    if call["kind"] != "program":
+      continue
+    if (arch := call["target"].get("arch", "")).split(",")[0] != ARCH:  # IMAGE builds append ",IMAGE_PITCH_ALIGNMENT=64"
+      out[f"arch={arch}"] = out.get(f"arch={arch}", 0) + 1
+      continue
+    for sig, n in emu.hw_unverified(bytes(blob[slice(*call["lib"])])).items():
+      out[sig] = out.get(sig, 0) + n
+  return out
+
+
+def _array(blob: memoryview, stored: dict) -> np.ndarray:
+  return np.frombuffer(blob[slice(*stored["data"])], dtype=np.dtype(stored["dtype"])).reshape(stored["shape"]).copy()
+
+
+def replay(header: dict, blob: memoryview, shared: dict) -> list[np.ndarray]:
+  from tinygrad import Tensor
+  check = header["check"]
+  jit = PlainJit(header, blob, None, verify=False, shared=shared)
+  kwargs: dict = {}
+  for path, stored, device in check["inputs"]:
+    tensor = Tensor(_array(blob, stored), device=_device(device, None)).realize()
+    kwargs[path[0]] = tensor if len(path) == 1 else {**kwargs.get(path[0], {}), path[1]: tensor}
+  result = jit(**kwargs)
+  outputs = kwargs[check["output_container"]].values() if check["output_container"] else result
+  return [t.numpy() for t in (outputs if isinstance(outputs, (list, tuple)) or check["output_container"] else [outputs])]
+
+
+def _parts_with_offsets(path: str) -> tuple[dict, dict]:
+  parts, _ = read_parts(path)
+  with open(path, "rb") as f:
+    head = f.read(len(MAGIC) + 8)
+    start = len(MAGIC) + 8 + struct.unpack_from("<Q", head, len(MAGIC))[0]
+    f.seek(len(MAGIC) + 8)
+    entries = json.loads(f.read(start - len(MAGIC) - 8))["parts"]
+  base = start + -start % ALIGNMENT
+  offsets = {tuple(e["key"]) if isinstance(e["key"], list) else e["key"]: base + e["blob"][0] for e in entries}
+  return parts, offsets
+
+
+def _gate(parts: dict) -> None:
+  missing: dict[str, int] = {}
+  for header, blob in parts.values():
+    if "check" not in header:
+      raise ParityError("artifact has no check frame")
+    for sig, n in unverified(header, blob).items():
+      missing[sig] = missing.get(sig, 0) + n
+  if missing:
+    raise ParityError(f"kernels use instructions the emulator is not verified against a {HARDWARE} for: " +
+                      ", ".join(f"{s} ({n}x)" for s, n in sorted(missing.items(), key=lambda kv: -kv[1])))
+
+
+def check(path: str) -> None:
+  parts, _ = read_parts(path)
+  if ON_MOCK:
+    _gate(parts)
+  shared: dict = {}
+  for key, (header, blob) in parts.items():
+    got = replay(header, blob, shared)
+    for i, (g, stored) in enumerate(zip(got, header["check"]["outputs"], strict=True)):
+      want = _array(blob, stored)
+      if g.shape != want.shape or not np.array_equal(g.view(np.uint8), want.view(np.uint8)):
+        n = int(np.count_nonzero(g != want)) if g.shape == want.shape else -1
+        raise ParityError(f"part {key} output {i}: the {'mock' if ON_MOCK else 'device'}'s check frame differs from the stored one in {n} of {want.size} values")
+
+
+def record(path: str, out: str) -> None:
+  parts, offsets = _parts_with_offsets(path)
+  _gate(parts)
+  shared, writes = {}, []
+  for key, (header, blob) in parts.items():
+    for g, stored in zip(replay(header, blob, shared), header["check"]["outputs"], strict=True):
+      want = np.dtype(stored["dtype"])
+      if tuple(g.shape) != tuple(stored["shape"]) or g.dtype != want:
+        raise ParityError(f"part {key}: replayed output {g.dtype}{g.shape} does not fit the stored {want}{tuple(stored['shape'])}")
+      writes.append((offsets[key] + stored["data"][0], np.ascontiguousarray(g).tobytes()))
+  shutil.copyfile(path, out)
+  with open(out, "r+b") as f:
+    for offset, data in writes:
+      f.seek(offset)
+      f.write(data)
+
+
+def main(argv: list[str]) -> int:
+  if len(argv) < 2 or argv[0] not in ("check", "record") or (argv[0] == "record" and len(argv) != 3):
+    print("usage: mock_parity check ARTIFACT... | record ARTIFACT OUT", file=sys.stderr)
+    return 2
+  try:
+    if argv[0] == "record":
+      st = time.perf_counter()
+      record(argv[1], argv[2])
+      print(f"{argv[2]}: check frame recorded on the mock in {time.perf_counter() - st:.0f}s")
+      return 0
+    for path in argv[1:]:
+      st = time.perf_counter()
+      check(path)
+      print(f"{os.path.basename(path)}: bit-identical to its {HARDWARE} check frame on the {'mock' if ON_MOCK else 'device'} ({time.perf_counter() - st:.0f}s)")
+    return 0
+  except ParityError as e:
+    print(f"PARITY FAIL: {e}", file=sys.stderr)
+    return 1
+
+
+if __name__ == "__main__":
+  sys.exit(main(sys.argv[1:]))
@@ -242,7 +242,7 @@ class ModelState(ModelStateBase):
      desired_curvature = get_curvature_from_output(model_output, curvature_plan, v_ego, lat_action_t, self.mlsim)
    else:
      desired_accel = model_output['action'][0, 1]
-      desired_curvature = model_output['action'][0, 0] / (max(1.0, v_ego))**2
+      desired_curvature = model_output['action'][0, 0] / (max(2.5, v_ego))**2
    if blendful is not None:
      desired_curvature += blendful(desired_curvature)

@@ -80,7 +80,8 @@ def serialize_jit(jit, kernel_linear, check: dict | None = None) -> tuple[dict,
    if ast.op is Ops.PROGRAM:
      elf, info = ast.to_elf(), ast.arg
      global_size, local_size = info.launch_dims({})
-      calls.append({"kind": "program", "name": elf.name, "lib": blob.add(bytes(elf.lib)), "target": dataclasses.asdict(elf.target),
+      device_target = dataclasses.asdict(dataclasses.replace(elf.target, interface=elf.target.interface.removeprefix("MOCK")))
+      calls.append({"kind": "program", "name": elf.name, "lib": blob.add(bytes(elf.lib)), "target": device_target,
                    "signature": kernel_signature(ast, elf),
                    "global_size": [int(x) for x in global_size], "local_size": [int(x) for x in local_size],
                    "globals": list(info.globals), "outs": list(info.outs), "ins": list(info.ins), "args": arguments})
@@ -6,6 +6,7 @@ See the LICENSE.md file in the root directory for more details.
 """
 import os
 import re
+import struct
 import sys
 import tempfile
 import unittest
@@ -25,8 +26,8 @@ if sys.platform == "darwin" and os.environ.get("DEV", "").startswith("MOCKUSB"):

 from openpilot.common.test import OpenpilotTestCase
 from openpilot.sunnypilot.modeld_v2 import convert_upstream_pkl
-from openpilot.sunnypilot.modeld_v2.convert_upstream_pkl import convert
-from openpilot.sunnypilot.modeld_v2.helpers import PlainJit, load_model
+from openpilot.sunnypilot.modeld_v2.convert_upstream_pkl import _elf, convert
+from openpilot.sunnypilot.modeld_v2.helpers import ALIGNMENT, PlainJit, load_model
 from openpilot.sunnypilot.modeld_v2.plain_artifact_compiler import capture_frame

 IMG, DESIRE, STATE, OUT = (6, 8, 8), (8,), (4, 16), (1, 32)
@@ -119,3 +120,18 @@ class TestConvertUpstreamPkl(OpenpilotTestCase):
    bundle = load_model(out, cache=False, verify=True)
    assert bundle["metadata"] == {"name": "tiny"} and bundle["input_specs"] == self.input_specs and bundle["output_specs"] == self.output_specs
    self.assert_matches_eager(bundle["run"], 6)
+
+
+class TestKernelElf(OpenpilotTestCase):
+  def test_sections_follow_the_code(self):
+    from tinygrad.runtime.support.elf import elf_loader
+    code = bytes(range(1, 129))
+    for rodata, entry in ((len(code), 0), (0, 256)):
+      image = bytearray(code + bytes(64) if entry < rodata else bytes(256) + code)
+      struct.pack_into("<q", image, rodata + 16, entry - rodata)
+      loaded, sections, _ = elf_loader(_elf(bytes(image), rodata, entry))
+      headers = {section.name: section.header for section in sections}
+      assert bytes(loaded[ALIGNMENT:]) == bytes(image), "the uploaded program differs from the pickle's"
+      assert headers[".rodata"].sh_addr == ALIGNMENT + rodata, "the loader reads the kernel descriptor at .rodata"
+      text = headers[".text"]
+      assert text.sh_addr <= ALIGNMENT + entry < text.sh_addr + text.sh_size, "the amd mock sizes the code from the entry to the end of .text"
@@ -0,0 +1,67 @@
+"""
+Copyright (c) 2021-, sunnypilot, and a number of other contributors.
+
+This file is part of sunnypilot and is licensed under the MIT License.
+See the LICENSE.md file in the root directory for more details.
+"""
+import numpy as np
+
+from openpilot.cereal import log
+from openpilot.common.params import Params
+from openpilot.common.realtime import DT_MDL
+from openpilot.sunnypilot import PARAMS_UPDATE_PERIOD
+
+ACCEL_BOOST_MIN_SPEED = 1.0
+ACCEL_BOOST_MAX = (0.4, 0.25, 0.1)  # aggressive, standard, relaxed, m/s^2
+ACCEL_BOOST_BUILD, ACCEL_BOOST_RELEASE = 0.2, 1.0  # m/s^3
+BRAKE_PROB = (0.1, 0.3)  # the model's brake press in 4 s: it wants to slow
+GAS_PROB = (0.6, 0.8)  # predicts a human gas press at this hypothesis at about 23%
+CURVE_LAT_ACCEL = (1.5, 2.5)
+
+
+def get_boost_max(personality=log.LongitudinalPersonality.standard):
+  if personality == log.LongitudinalPersonality.relaxed:
+    return ACCEL_BOOST_MAX[2]
+  elif personality == log.LongitudinalPersonality.standard:
+    return ACCEL_BOOST_MAX[1]
+  elif personality == log.LongitudinalPersonality.aggressive:
+    return ACCEL_BOOST_MAX[0]
+  else:
+    raise NotImplementedError("Longitudinal personality not supported")
+
+
+class AccelBoostSP:
+  def __init__(self):
+    self.params = Params()
+    self.frame = -1
+    self.enabled = False
+    self.boost = 0.0
+    self.total_boost = 0.0
+
+  def update_params(self):
+    self.frame += 1
+    if self.frame % int(PARAMS_UPDATE_PERIOD / DT_MDL) != 0:
+      return False
+    enabled = self.params.get_bool("AccelBoostSP")
+    if enabled == self.enabled:
+      return False
+    self.enabled = enabled
+    self.boost = self.total_boost = 0.0
+    return True
+
+  def update(self, sm, output_a_target_e2e, output_a_target_mpc, a_cruise):
+    model = sm['modelV2']
+    brake_probs = model.meta.disengagePredictions.brakePressProbs
+    gas_probs = model.meta.disengagePredictions.gasPressProbs
+    if not sm['selfdriveState'].enabled or len(brake_probs) < 2 or len(gas_probs) < 2:
+      self.boost = self.total_boost = 0.0
+      return output_a_target_e2e
+
+    want = max(float(np.interp(brake_probs[1], BRAKE_PROB, [1.0, 0.0])), float(np.interp(gas_probs[1], GAS_PROB, [0.0, 1.0])))
+    want = 0.0 if model.action.shouldStop else want
+    want *= float(np.interp(abs(model.action.desiredCurvature) * sm['carState'].vEgo ** 2, CURVE_LAT_ACCEL, [1.0, 0.0]))
+    target = want * get_boost_max(sm['selfdriveState'].personality)
+    self.boost = float(np.clip(target, self.boost - ACCEL_BOOST_RELEASE * DT_MDL, self.boost + ACCEL_BOOST_BUILD * DT_MDL))
+    speed_scale = np.interp(sm['carState'].vEgo, [ACCEL_BOOST_MIN_SPEED, 2 * ACCEL_BOOST_MIN_SPEED], [0.0, 1.0])
+    self.total_boost = self.boost * speed_scale * np.interp(output_a_target_e2e, [-0.7, -0.2, 5.0], [0.0, 1.0, 1.0], right=0.0)
+    return output_a_target_e2e + self.total_boost
@@ -9,6 +9,8 @@ from openpilot.cereal import messaging, custom
 from opendbc.car import structs
 from openpilot.common.constants import CV
 from openpilot.selfdrive.car.cruise import V_CRUISE_MAX
+from openpilot.selfdrive.controls.lib.accel_boost import AccelBoost
+from openpilot.sunnypilot.selfdrive.controls.lib.accel_boost import AccelBoostSP
 from openpilot.sunnypilot.selfdrive.controls.lib.e2e_alerts_helper import E2EAlertsHelper
 from openpilot.sunnypilot.selfdrive.controls.lib.smart_cruise_control.smart_cruise_control import SmartCruiseControl
 from openpilot.sunnypilot.selfdrive.controls.lib.speed_limit.speed_limit_assist import SpeedLimitAssist
@@ -29,6 +31,8 @@ class LongitudinalPlannerSP:
    self.generation = int(model_bundle.generation) if (model_bundle := get_active_bundle()) else None
    self.source = LongitudinalPlanSource.cruise
    self.e2e_alerts_helper = E2EAlertsHelper()
+    self.accel_boost_sp = AccelBoostSP()
+    self.openpilot_longitudinal = CP.openpilotLongitudinalControl

    self.output_v_target = 0.
    self.output_a_target = 0.
@@ -64,6 +68,9 @@ class LongitudinalPlannerSP:
    return self.output_v_target, self.output_a_target

  def update(self, sm: messaging.SubMaster) -> None:
+    if self.accel_boost_sp.update_params():
+      self.accel_boost = self.accel_boost_sp if self.accel_boost_sp.enabled and self.openpilot_longitudinal else AccelBoost()
+
    self.events_sp.clear()
    self.e2e_alerts_helper.update(sm, self.events_sp)

@@ -0,0 +1,81 @@
+"""
+Copyright (c) 2021-, sunnypilot, and a number of other contributors.
+
+This file is part of sunnypilot and is licensed under the MIT License.
+See the LICENSE.md file in the root directory for more details.
+"""
+from types import SimpleNamespace
+
+from openpilot.common.parameterized import parameterized
+from openpilot.common.params import Params
+from openpilot.common.realtime import DT_MDL
+from openpilot.common.test import OpenpilotTestCase
+
+from openpilot.sunnypilot import PARAMS_UPDATE_PERIOD
+from openpilot.sunnypilot.selfdrive.controls.lib.accel_boost import AccelBoostSP, ACCEL_BOOST_MAX
+
+
+def make_sm(brake_prob=0.05, gas_prob=0.3, lat_accel=0.0, v_ego=30.0, personality=0, enabled=True, should_stop=False):
+  meta = SimpleNamespace(disengagePredictions=SimpleNamespace(brakePressProbs=[0.0, brake_prob, 0.0, 0.0, 0.0, 0.0],
+                                                              gasPressProbs=[0.0, gas_prob, 0.0, 0.0, 0.0, 0.0]))
+  action = SimpleNamespace(shouldStop=should_stop, desiredCurvature=lat_accel / v_ego ** 2)
+  return {'modelV2': SimpleNamespace(meta=meta, action=action), 'carState': SimpleNamespace(vEgo=v_ego),
+          'selfdriveState': SimpleNamespace(enabled=enabled, personality=personality)}
+
+
+def settle(sm, frames=100):
+  boost = AccelBoostSP()
+  for _ in range(frames):
+    boost.update(sm, 0.0, 9.0, 9.0)
+  return boost.total_boost
+
+
+class TestAccelBoostSP(OpenpilotTestCase):
+  @parameterized.expand([(0,), (1,), (2,)])
+  def test_personality_max(self, personality):
+    assert abs(settle(make_sm(personality=personality)) - ACCEL_BOOST_MAX[personality]) < 1e-6
+
+  def test_brake_gate_closes(self):
+    assert settle(make_sm(brake_prob=0.5)) == 0.0
+
+  def test_gas_reopens(self):
+    assert abs(settle(make_sm(brake_prob=0.5, gas_prob=0.9)) - ACCEL_BOOST_MAX[0]) < 1e-6
+
+  def test_should_stop(self):
+    assert settle(make_sm(gas_prob=0.9, should_stop=True)) == 0.0
+
+  @parameterized.expand([(1.5, 1.0), (2.0, 0.5), (2.5, 0.0), (-2.0, 0.5)])
+  def test_curve_fade(self, lat_accel, scale):
+    assert abs(settle(make_sm(lat_accel=lat_accel)) - scale * ACCEL_BOOST_MAX[0]) < 1e-6
+
+  def test_low_speed(self):
+    assert settle(make_sm(v_ego=0.5)) == 0.0
+
+  def test_disabled_resets(self):
+    boost = AccelBoostSP()
+    for _ in range(100):
+      boost.update(make_sm(), 0.0, 9.0, 9.0)
+    assert boost.update(make_sm(enabled=False), 0.0, 9.0, 9.0) == 0.0
+    assert boost.boost == 0.0 and boost.total_boost == 0.0
+
+  def test_build_and_release_rates(self):
+    boost = AccelBoostSP()
+    boost.update(make_sm(), 0.0, 9.0, 9.0)
+    assert abs(boost.total_boost - 0.01) < 1e-6
+    for _ in range(100):
+      boost.update(make_sm(), 0.0, 9.0, 9.0)
+    boost.update(make_sm(brake_prob=0.5), 0.0, 9.0, 9.0)
+    assert abs(boost.total_boost - (ACCEL_BOOST_MAX[0] - 0.05)) < 1e-6
+
+  def test_param_switch(self):
+    params = Params()
+    boost = AccelBoostSP()
+    params.put_bool("AccelBoostSP", True, block=True)
+    assert boost.update_params() and boost.enabled
+    for _ in range(100):
+      boost.update(make_sm(), 0.0, 9.0, 9.0)
+    params.put_bool("AccelBoostSP", False, block=True)
+    for _ in range(int(PARAMS_UPDATE_PERIOD / DT_MDL) - 1):
+      assert not boost.update_params()
+    assert boost.update_params() and not boost.enabled
+    assert boost.boost == 0.0 and boost.total_boost == 0.0
@@ -620,6 +620,19 @@
                }
              ]
            },
+            {
+              "key": "AccelBoostSP",
+              "widget": "toggle",
+              "title": "Acceleration Boost",
+              "description": "Adds a small push on top of sunnypilot's own acceleration, stronger in Aggressive and gentler in Relaxed. Eases off in curves and when sunnypilot slows or stops. Never goes past your set speed, and lead-car following still comes first.",
+              "enablement": [
+                {
+                  "type": "capability",
+                  "field": "has_longitudinal_control",
+                  "equals": true
+                }
+              ]
+            },
            {
              "key": "IntelligentCruiseButtonManagement",
              "widget": "toggle",
@@ -30,6 +30,12 @@ sections:
      label: Relaxed
    enablement:
    - $ref: '#/macros/longitudinal'
+  - key: AccelBoostSP
+    widget: toggle
+    title: Acceleration Boost
+    description: "Adds a small push on top of sunnypilot's own acceleration, stronger in Aggressive and gentler in Relaxed. Eases off in curves and when sunnypilot slows or stops. Never goes past your set speed, and lead-car following still comes first."
+    enablement:
+    - $ref: '#/macros/longitudinal'
  - key: IntelligentCruiseButtonManagement
    widget: toggle
    title: Intelligent Cruise Button Management (ICBM) (Alpha)
@@ -13,26 +13,41 @@ from pathlib import Path
 from datetime import datetime, UTC


+_VERSION_REGEX = re.compile(r'^(?:\d+[a-zA-Z]+|\d+[vVbB]\d+|[vVbB]\d+|\d{4})$')
+_DECIMAL_VERSION_REGEX = re.compile(r'^([vVbB]?)(\d+(?:\.\d+)+)$')
+
+
+def _version(word: str) -> str | None:
+  match = _DECIMAL_VERSION_REGEX.match(word.strip('()[]{}.,;:'))
+  if match:
+    return match.group(1).upper() + match.group(2)
+  stripped = re.sub(r'[^a-zA-Z0-9]', '', word)
+  return stripped.upper() if _VERSION_REGEX.match(stripped) else None
+
+
+def _truncate(short_name: str) -> str:
+  return short_name[:8].rstrip('.')
+
+
 def create_short_name(full_name: str) -> str:
-  # Remove parentheses and extract alphanumeric words
+  # Remove parentheses and extract alphanumeric words, keeping decimals for versioning
  clean_name = re.sub(r'\([^)]*\)', '', full_name)
-  words = [re.sub(r'[^a-zA-Z0-9]', '', word) for word in clean_name.split() if re.sub(r'[^a-zA-Z0-9]', '', word)]
+  words = [re.sub(r'[^a-zA-Z0-9.]', '', word) for word in clean_name.split()]
+  words = [word for word in words if word]

  if len(words) == 1:
-    return words[0][:8].upper()
+    return _truncate((_version(words[0]) or words[0]).upper())

-  # Handle special case: Name + Version (e.g., "Word A1" -> "WordA1")
-  if len(words) == 2 and re.match(r'^[A-Za-z]\d+$', words[1]):
-    return (words[0] + words[1])[:8].upper()
+  # Handle special case: Name + Version (e.g., "Word A1" -> "WordA1", "Model v1.1" -> "ModelV1.1")
+  if len(words) == 2 and re.match(r'^[A-Za-z]?\d+$', words[1]):
+    return _truncate((words[0] + (_version(words[1]) or words[1])).upper())

  result = ""
  for word in words:
    # Version or number patterns
-    if (re.match(r'^\d+[a-zA-Z]+$', word) or
-        re.match(r'^\d+[vVbB]\d+$', word) or
-        re.match(r'^[vVbB]\d+$', word) or
-        re.match(r'^\d{4}$', word)):
-      result += word.upper()
+    version = _version(word)
+    if version:
+      result += version
    # All uppercase abbreviations (2-3 letters)
    elif re.match(r'^[A-Z]{2,3}$', word):
      result += word
@@ -45,7 +60,7 @@ def create_short_name(full_name: str) -> str:
      result += word
    else:
      result += word[0].upper()
-  return result[:8]
+  return _truncate(result)


 def create_pkl_name(full_name: str) -> str:
Author	SHA1	Message	Date
discountchubbs	8bea2d48c1	modeld_v2: emu	2026-11-06 15:18:45 -08:00
Jason Wen	3d8f211d4b	bump opendbc	2026-11-06 16:03:02 -05:00
James Vecellio-Grant	02ef6e4ce0	modeld_v2: artifact parity with upstream tg (#2093 ) * modeld_v2: artifact parity with upstream tg * m * m * Update helpers.py * ew * Update convert_upstream_pkl.py * Update test_convert_upstream_pkl.py --------- Co-authored-by: Jason Wen <haibin.wen3@gmail.com>	2026-11-05 16:11:27 -05:00
Nayan	a1a28e467e	[models] Support major.minor versioning (#2092 ) support major.minor versioning, becase why not	2026-11-05 15:35:07 -05:00
James Vecellio-Grant	932ff06d93	modeld_v2: fix action curvature (#2091 )	2026-11-04 14:32:06 -08:00
James Vecellio-Grant	8cf6ff9a33	controls: predictive accel boost (#2089 ) * controls: predictive accel boost * ew * lower entry threshold for fade, pos still builds. brake_probs still triggers at same hyp * ui: add MICI accel boost side-panel bar * jason and kumar fought for hours to make this default so here we go * controls: accel boost review fixes * [MICI] ui: move accel boost ball swap out of stock * ui: drop restating accel boost comments * ui: drop accel boost onroad indicator --------- Co-authored-by: rav4kumar <meetkumardesai@gmail.com> Co-authored-by: Jason Wen <haibin.wen3@gmail.com>	2026-11-03 23:16:52 -05:00