openvino: Update to OV 2026.2.1, self-contained release packages, operator improvements

.devops/openvino.Dockerfile

@@ -1,12 +1,12 @@
-ARG OPENVINO_VERSION_MAJOR=2026.2
-ARG OPENVINO_VERSION_FULL=2026.2.0.21903.52ddc073857
+ARG OPENVINO_VERSION_MAJOR=2026.2.1
+ARG OPENVINO_VERSION_FULL=2026.2.1.21919.ede283a88e3
 ARG UBUNTU_VERSION=24.04
 
 # Intel GPU driver versions. https://github.com/intel/compute-runtime/releases
-ARG IGC_VERSION=v2.34.4
-ARG IGC_VERSION_FULL=2_2.34.4+21428
-ARG COMPUTE_RUNTIME_VERSION=26.18.38308.1
-ARG COMPUTE_RUNTIME_VERSION_FULL=26.18.38308.1-0
+ARG IGC_VERSION=v2.36.3
+ARG IGC_VERSION_FULL=2_2.36.3+21719
+ARG COMPUTE_RUNTIME_VERSION=26.22.38646.4
+ARG COMPUTE_RUNTIME_VERSION_FULL=26.22.38646.4-0
 ARG IGDGMM_VERSION=22.10.0
 
 # Intel NPU driver versions. https://github.com/intel/linux-npu-driver/releases

.github/workflows/build-cache.yml

@@ -68,8 +68,8 @@ jobs:
 
     env:
       # Sync versions in build.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
-      OPENVINO_VERSION_MAJOR: "2026.2"
-      OPENVINO_VERSION_FULL: "2026.2.0.21903.52ddc073857"
+      OPENVINO_VERSION_MAJOR: "2026.2.1"
+      OPENVINO_VERSION_FULL: "2026.2.1.21919.ede283a88e3"
 
     steps:
       - name: Clone
@@ -96,8 +96,8 @@ jobs:
 
     env:
       # Sync versions in build.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
-      OPENVINO_VERSION_MAJOR: "2026.2"
-      OPENVINO_VERSION_FULL: "2026.2.0.21903.52ddc073857"
+      OPENVINO_VERSION_MAJOR: "2026.2.1"
+      OPENVINO_VERSION_FULL: "2026.2.1.21919.ede283a88e3"
 
     steps:
       - name: Clone

.github/workflows/build-openvino.yml

@@ -39,8 +39,8 @@ jobs:
 
     env:
       # Sync versions in build-openvino.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
-      OPENVINO_VERSION_MAJOR: "2026.2"
-      OPENVINO_VERSION_FULL: "2026.2.0.21903.52ddc073857"
+      OPENVINO_VERSION_MAJOR: "2026.2.1"
+      OPENVINO_VERSION_FULL: "2026.2.1.21919.ede283a88e3"
 
     steps:
       - name: Clone
@@ -96,8 +96,8 @@ jobs:
 
     env:
       # Sync versions in build-openvino.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
-      OPENVINO_VERSION_MAJOR: "2026.2"
-      OPENVINO_VERSION_FULL: "2026.2.0.21903.52ddc073857"
+      OPENVINO_VERSION_MAJOR: "2026.2.1"
+      OPENVINO_VERSION_FULL: "2026.2.1.21919.ede283a88e3"
 
     steps:
       - name: Clone

.github/workflows/build-self-hosted.yml

@@ -266,8 +266,8 @@ jobs:
 
     env:
       # Sync versions in build.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
-      OPENVINO_VERSION_MAJOR: "2026.2"
-      OPENVINO_VERSION_FULL: "2026.2.0.21903.52ddc073857"
+      OPENVINO_VERSION_MAJOR: "2026.2.1"
+      OPENVINO_VERSION_FULL: "2026.2.1.21919.ede283a88e3"
 
     steps:
       - name: Clone

.github/workflows/release.yml

@@ -446,8 +446,8 @@ jobs:
 
     env:
       # Sync versions in build-openvino.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
-      OPENVINO_VERSION_MAJOR: "2026.2"
-      OPENVINO_VERSION_FULL: "2026.2.0.21903.52ddc073857"
+      OPENVINO_VERSION_MAJOR: "2026.2.1"
+      OPENVINO_VERSION_FULL: "2026.2.1.21919.ede283a88e3"
 
     steps:
       - name: Set OpenVINO version output
@@ -506,8 +506,11 @@ jobs:
           cmake -B build/ReleaseOV -G Ninja \
             -DCMAKE_BUILD_TYPE=Release \
             -DGGML_OPENVINO=ON \
-            -DHF_UI_VERSION=${{ needs.get-version.outputs.ui_version }}
-          cmake --build build/ReleaseOV --config Release -j $(nproc)
+            -DCMAKE_INSTALL_RPATH='$ORIGIN' \
+            -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \
+            -DHF_UI_VERSION=${{ needs.get-version.outputs.ui_version }} \
+            ${{ env.CMAKE_ARGS }}
+          cmake --build build/ReleaseOV --config Release --parallel
 
       - name: ccache-clear
         uses: ./.github/actions/ccache-clear
@@ -521,8 +524,26 @@ jobs:
       - name: Pack artifacts
         id: pack_artifacts
         run: |
-          cp LICENSE ./build/ReleaseOV/bin/
-          tar -czvf llama-${{ steps.tag.outputs.name }}-bin-ubuntu-openvino-${{ env.OPENVINO_VERSION_MAJOR }}-x64.tar.gz --transform "s,^\.,llama-${{ steps.tag.outputs.name }}," -C ./build/ReleaseOV/bin .
+          dest=./build/ReleaseOV/bin/
+          OPENVINO_ROOT=./openvino_toolkit
+          ov_lib="$OPENVINO_ROOT/runtime/lib/intel64"
+
+          # Bundle OpenVINO runtime libs + TBB. Binaries built with RPATH=$ORIGIN
+          # load these siblings without setupvars.sh / LD_LIBRARY_PATH.
+          cp -P "$ov_lib"/libopenvino.so* \
+                "$ov_lib"/libopenvino_c.so* \
+                "$ov_lib"/libopenvino_*_plugin.so \
+                "$ov_lib"/libopenvino_intel_npu_compiler*.so \
+                "$OPENVINO_ROOT/runtime/3rdparty/tbb/lib"/*.so* \
+                "$dest"
+          cp -P /usr/lib/x86_64-linux-gnu/libOpenCL.so.1* "$dest" 2>/dev/null || true
+          cp "$ov_lib/cache.json" "$dest" 2>/dev/null || true
+
+          # OpenVINO licensing
+          cp -r "$OPENVINO_ROOT/docs/licensing" "$dest/openvino-licensing"
+
+          cp LICENSE "$dest"
+          tar -czvf llama-${{ steps.tag.outputs.name }}-bin-ubuntu-openvino-${{ env.OPENVINO_VERSION_MAJOR }}-x64.tar.gz --transform "s,^\.,llama-${{ steps.tag.outputs.name }}," -C "$dest" .
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v6
@@ -538,8 +559,8 @@ jobs:
 
     env:
       # Sync versions in build-openvino.yml, build-self-hosted.yml, release.yml, build-cache.yml, .devops/openvino.Dockerfile
-      OPENVINO_VERSION_MAJOR: "2026.2"
-      OPENVINO_VERSION_FULL: "2026.2.0.21903.52ddc073857"
+      OPENVINO_VERSION_MAJOR: "2026.2.1"
+      OPENVINO_VERSION_FULL: "2026.2.1.21919.ede283a88e3"
 
     steps:
       - name: Set OpenVINO version output
@@ -607,7 +628,9 @@ jobs:
             -A x64 ^
             -DCMAKE_BUILD_TYPE=Release ^
             -DGGML_OPENVINO=ON ^
-            -DCMAKE_TOOLCHAIN_FILE=C:\vcpkg\scripts\buildsystems\vcpkg.cmake
+            -DLLAMA_BUILD_BORINGSSL=ON ^
+            -DCMAKE_TOOLCHAIN_FILE=C:\vcpkg\scripts\buildsystems\vcpkg.cmake ^
+            ${{ env.CMAKE_ARGS }}
 
           cmake --build build\ReleaseOV --config Release -- /m
 
@@ -624,8 +647,29 @@ jobs:
         id: pack_artifacts
         shell: powershell
         run: |
-          Copy-Item LICENSE .\build\ReleaseOV\bin\
-          7z a -snl llama-${{ steps.tag.outputs.name }}-bin-win-openvino-${{ env.OPENVINO_VERSION_MAJOR }}-x64.zip .\build\ReleaseOV\bin\*
+          # Locate the extracted OpenVINO toolkit root (same pattern as the Build step).
+          $OPENVINO_ROOT = (Get-ChildItem -Directory openvino_toolkit | Select-Object -First 1).FullName
+          if (-not $OPENVINO_ROOT) {
+            Write-Error "OpenVINO toolkit folder not found under .\openvino_toolkit"
+            exit 1
+          }
+
+          $dest = ".\build\ReleaseOV\bin\Release"
+
+          $ovBin = Join-Path $OPENVINO_ROOT 'runtime\bin\intel64\Release'
+          Copy-Item -Path (Join-Path $ovBin '*.dll')       -Destination $dest -Force
+          Copy-Item -Path (Join-Path $ovBin 'cache.json')  -Destination $dest -Force
+
+          $tbbBin = Join-Path $OPENVINO_ROOT 'runtime\3rdparty\tbb\bin'
+          Copy-Item -Path (Join-Path $tbbBin 'tbb*.dll') -Destination $dest -Force
+
+          # OpenVINO licensing
+          $licensingDest = Join-Path $dest 'openvino-licensing'
+          New-Item -ItemType Directory -Force -Path $licensingDest | Out-Null
+          Copy-Item -Path (Join-Path $OPENVINO_ROOT 'docs\licensing\*') -Destination $licensingDest -Recurse -Force
+
+          Copy-Item LICENSE $dest
+          7z a -snl llama-${{ steps.tag.outputs.name }}-bin-win-openvino-${{ env.OPENVINO_VERSION_MAJOR }}-x64.zip $dest\*
 
       - name: Upload artifacts
         uses: actions/upload-artifact@v6

docs/backend/OPENVINO.md

@@ -237,8 +237,8 @@ chmod +x ubuntu-llamacpp-ov-install.sh
 # ============================================
 set -euo pipefail
 
-OPENVINO_VERSION_MAJOR="2026.2"
-OPENVINO_VERSION_FULL="2026.2.0.21903.52ddc073857"
+OPENVINO_VERSION_MAJOR="2026.2.1"
+OPENVINO_VERSION_FULL="2026.2.1.21919.ede283a88e3"
 
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 OPENVINO_INSTALL_DIR="/opt/intel/openvino_${OPENVINO_VERSION_MAJOR}"
@@ -334,7 +334,7 @@ echo "  ./build/ReleaseOV/bin/llama-cli -m model.gguf"
 ```
 
 > [!NOTE]
-> The script pins OpenVINO `2026.2` via the `OPENVINO_VERSION_MAJOR` / `OPENVINO_VERSION_FULL` variables at the top — edit them to track a different release.
+> The script pins OpenVINO `2026.2.1` via the `OPENVINO_VERSION_MAJOR` / `OPENVINO_VERSION_FULL` variables at the top — edit them to track a different release.
 
 </details>
 
@@ -364,8 +364,8 @@ REM ============================================
 REM llama.cpp OpenVINO Build Script (Ninja)
 REM ============================================
 
-set "OPENVINO_VERSION_MAJOR=2026.2"
-set "OPENVINO_VERSION_FULL=2026.2.0.21903.52ddc073857"
+set "OPENVINO_VERSION_MAJOR=2026.2.1"
+set "OPENVINO_VERSION_FULL=2026.2.1.21919.ede283a88e3"
 
 set "SCRIPT_DIR=%~dp0"
 set "VCPKG_DIR=C:\vcpkg"
@@ -547,7 +547,7 @@ endlocal
 ```
 
 > [!NOTE]
-> The script pins OpenVINO `2026.2` via the `OPENVINO_VERSION_MAJOR` / `OPENVINO_VERSION_FULL` variables at the top — edit them to track a different release. From any new shell, source the matching `setupvars` script via the junction — `call "C:\Intel\openvino\setupvars.bat"` from `cmd`, or `& "C:\Intel\openvino\setupvars.ps1"` from PowerShell. If `winget` cannot register Visual Studio Build Tools on first run, install them once manually and re-run the script from an elevated **Developer Command Prompt for VS 2022**.
+> The script pins OpenVINO `2026.2.1` via the `OPENVINO_VERSION_MAJOR` / `OPENVINO_VERSION_FULL` variables at the top — edit them to track a different release. From any new shell, source the matching `setupvars` script via the junction — `call "C:\Intel\openvino\setupvars.bat"` from `cmd`, or `& "C:\Intel\openvino\setupvars.ps1"` from PowerShell. If `winget` cannot register Visual Studio Build Tools on first run, install them once manually and re-run the script from an elevated **Developer Command Prompt for VS 2022**.
 
 </details>
 

ggml/src/ggml-openvino/ggml-decoder.cpp

@@ -1270,77 +1270,14 @@ void GgmlOvDecoder::visit_subgraph(std::function<void(std::shared_ptr<GgmlDecode
 }
 
 std::string GgmlOvDecoder::compute_op_type(const ggml_tensor * node) {
-    static const std::map<ggml_op, std::string> ops = {
-        {GGML_OP_NONE,            "GGML_OP_NONE"           },
-        {GGML_OP_ACC,             "GGML_OP_ACC"            },
-        {GGML_OP_ADD,             "GGML_OP_ADD"            },
-        {GGML_OP_ADD1,            "GGML_OP_ADD1"           },
-        {GGML_OP_ADD_ID,          "GGML_OP_ADD_ID"         },
-        {GGML_OP_CONCAT,          "GGML_OP_CONCAT"         },
-        {GGML_OP_CONT,            "GGML_OP_CONT"           },
-        {GGML_OP_DIV,             "GGML_OP_DIV"            },
-        {GGML_OP_DUP,             "GGML_OP_DUP"            },
-        {GGML_OP_GET_ROWS,        "GGML_OP_GET_ROWS"       },
-        {GGML_OP_MUL,             "GGML_OP_MUL"            },
-        {GGML_OP_MUL_MAT,         "GGML_OP_MUL_MAT"        },
-        {GGML_OP_MUL_MAT_ID,      "GGML_OP_MUL_MAT_ID"     },
-        {GGML_OP_PERMUTE,         "GGML_OP_PERMUTE"        },
-        {GGML_OP_RESHAPE,         "GGML_OP_RESHAPE"        },
-        {GGML_OP_RMS_NORM,        "GGML_OP_RMS_NORM"       },
-        {GGML_OP_NORM,            "GGML_OP_NORM"           },
-        {GGML_OP_ROPE,            "GGML_OP_ROPE"           },
-        {GGML_OP_SCALE,           "GGML_OP_SCALE"          },
-        {GGML_OP_SOFT_MAX,        "GGML_OP_SOFT_MAX"       },
-        {GGML_OP_SUM_ROWS,        "GGML_OP_SUM_ROWS"       },
-        {GGML_OP_SUB,             "GGML_OP_SUB"            },
-        {GGML_OP_TRANSPOSE,       "GGML_OP_TRANSPOSE"      },
-        {GGML_OP_VIEW,            "GGML_OP_VIEW"           },
-        {GGML_OP_SET_ROWS,        "GGML_OP_SET_ROWS"       },
-        {GGML_OP_CPY,             "GGML_OP_CPY"            },
-        {GGML_OP_FLASH_ATTN_EXT,  "GGML_OP_FLASH_ATTN_EXT" },
-        {GGML_OP_L2_NORM,         "GGML_OP_L2_NORM"        },
-        {GGML_OP_CLAMP,           "GGML_OP_CLAMP"          },
-        {GGML_OP_PAD,             "GGML_OP_PAD"            },
-        {GGML_OP_SSM_CONV,        "GGML_OP_SSM_CONV"       },
-        {GGML_OP_GATED_DELTA_NET, "GGML_OP_GATED_DELTA_NET"},
-        {GGML_OP_ARGSORT,         "GGML_OP_ARGSORT"        },
-        {GGML_OP_REPEAT,          "GGML_OP_REPEAT"         },
-        {GGML_OP_IM2COL,          "GGML_OP_IM2COL"         }
-    };
-    static const std::map<ggml_unary_op, std::string> unary_ops = {
-        {GGML_UNARY_OP_ABS,         "GGML_UNARY_OP_ABS"        },
-        {GGML_UNARY_OP_SGN,         "GGML_UNARY_OP_SGN"        },
-        {GGML_UNARY_OP_NEG,         "GGML_UNARY_OP_NEG"        },
-        {GGML_UNARY_OP_STEP,        "GGML_UNARY_OP_STEP"       },
-        {GGML_UNARY_OP_TANH,        "GGML_UNARY_OP_TANH"       },
-        {GGML_UNARY_OP_ELU,         "GGML_UNARY_OP_ELU"        },
-        {GGML_UNARY_OP_RELU,        "GGML_UNARY_OP_RELU"       },
-        {GGML_UNARY_OP_SIGMOID,     "GGML_UNARY_OP_SIGMOID"    },
-        {GGML_UNARY_OP_GELU,        "GGML_UNARY_OP_GELU"       },
-        {GGML_UNARY_OP_GELU_QUICK,  "GGML_UNARY_OP_GELU_QUICK" },
-        {GGML_UNARY_OP_SILU,        "GGML_UNARY_OP_SILU"       },
-        {GGML_UNARY_OP_SOFTPLUS,    "GGML_UNARY_OP_SOFTPLUS"   },
-        {GGML_UNARY_OP_HARDSWISH,   "GGML_UNARY_OP_HARDSWISH"  },
-        {GGML_UNARY_OP_HARDSIGMOID, "GGML_UNARY_OP_HARDSIGMOID"},
-        {GGML_UNARY_OP_EXP,         "GGML_UNARY_OP_EXP"        },
-        {GGML_UNARY_OP_COUNT,       "GGML_UNARY_OP_COUNT"      }
-    };
-    static const std::map<ggml_glu_op, std::string> glu_ops = {
-        {GGML_GLU_OP_SWIGLU, "GGML_GLU_OP_SWIGLU"},
-        {GGML_GLU_OP_GEGLU,  "GGML_GLU_OP_GEGLU" },
-        {GGML_GLU_OP_REGLU,  "GGML_GLU_OP_REGLU" }
-    };
-
     switch (node->op) {
     case GGML_OP_UNARY:
-        return unary_ops.at(ggml_get_unary_op(node));
+        return std::string("GGML_UNARY_OP_") + ggml_unary_op_name(ggml_get_unary_op(node));
     case GGML_OP_GLU:
-        return glu_ops.at(ggml_get_glu_op(node));
+        return std::string("GGML_GLU_OP_") + ggml_glu_op_name(ggml_get_glu_op(node));
     default:
-        return ops.at(node->op);
+        return std::string("GGML_OP_") + ggml_op_name(node->op);
     }
-    static const std::string unknown_op = "UNKNOWN_GGML_OP";
-    return unknown_op;
 }
 
 const std::string & GgmlOvDecoder::get_op_type(int node_idx) const {

ggml/src/ggml-openvino/openvino/op/add_id.cpp

@@ -17,6 +17,22 @@ namespace frontend {
 namespace ggml {
 namespace op {
 
+static ov::Output<ov::Node> reshape_add_id_input_to_2d(const ov::Output<ov::Node> & input,
+                                                       const ov::PartialShape & input_shape,
+                                                       const std::vector<int> & dims) {
+    const auto actual_shape = input.get_partial_shape();
+    if (actual_shape.rank().is_static() && actual_shape.rank().get_length() == 2) {
+        return input;
+    }
+
+    if (input_shape.rank().is_static() && input_shape.rank().get_length() == 2) {
+        return input;
+    }
+
+    auto shape = std::make_shared<ov::op::v3::ShapeOf>(input, ov::element::i64);
+    return std::make_shared<ov::op::v1::Reshape>(input, get_dimensions(shape, dims), false);
+}
+
 OutputVector translate_add_id(const NodeContext & context) {
     num_inputs_check(context, 3, 3);
 
@@ -28,11 +44,9 @@ OutputVector translate_add_id(const NodeContext & context) {
     //   input: [1, n_token, n_used, n_embd]
     //   bias:  [1, 1, n_expert, n_embd]
     //   ids:   [1, 1, n_token, n_used]
-    auto bias_shape_4d = std::make_shared<ov::op::v3::ShapeOf>(bias, ov::element::i64);
-    auto ids_shape_4d = std::make_shared<ov::op::v3::ShapeOf>(ids, ov::element::i64);
-
-    bias = std::make_shared<ov::op::v1::Reshape>(bias, get_dimensions(bias_shape_4d, {2, 3}), false);
-    ids = std::make_shared<ov::op::v1::Reshape>(ids, get_dimensions(ids_shape_4d, {2, 3}), false);
+    // Model bias constants may already be stored as [n_expert, n_embd].
+    bias = reshape_add_id_input_to_2d(bias, context.get_input_shape(1), {2, 3});
+    ids = reshape_add_id_input_to_2d(ids, context.get_input_shape(2), {2, 3});
 
     if (ids.get_element_type() != ov::element::i32 && ids.get_element_type() != ov::element::i64) {
         ids = std::make_shared<ov::op::v0::Convert>(ids, ov::element::i32);