responses: support namespace tool calls

.gitattributes

@@ -7,22 +7,9 @@ llama/**/*.cuh linguist-vendored
 llama/**/*.m linguist-vendored
 llama/**/*.metal linguist-vendored
 
-ml/backend/**/*.c linguist-vendored
-ml/backend/**/*.h linguist-vendored
-ml/backend/**/*.cpp linguist-vendored
-ml/backend/**/*.hpp linguist-vendored
-ml/backend/**/*.cu linguist-vendored
-ml/backend/**/*.cuh linguist-vendored
-ml/backend/**/*.m linguist-vendored
-ml/backend/**/*.metal linguist-vendored
-ml/backend/**/*.comp linguist-vendored
-ml/backend/**/*.glsl linguist-vendored
-ml/backend/**/CMakeLists.txt linguist-vendored
-
 app/webview linguist-vendored
 
 llama/build-info.cpp linguist-generated
-ml/backend/ggml/ggml/src/ggml-metal/ggml-metal-embed.s linguist-generated
 
 * text=auto
 *.go text eol=lf

.github/scripts/prepare_mlx_darwin.sh

@@ -9,7 +9,7 @@
 #
 # The payload also depends on Ollama's payload build rules (cmake glue and
 # carried mlx/compat patches) and the xgrammar native wrapper
-# (x/mlxrunner/xgrammar/native). Rule drift rebuilds the whole payload from
+# (mlxrunner/xgrammar/native). Rule drift rebuilds the whole payload from
 # source; wrapper-only drift rebuilds just libollama_xgrammar.dylib.
 #
 # If no release matches the MLX pins (e.g. right after a pin bump), the
@@ -52,17 +52,15 @@ read_pin() {
   tr -d '[:space:]' <"$1"
 }
 
-# Native wrapper sources compiled into libollama_xgrammar.dylib — keep in
-# sync with the ollama_xgrammar target in cmake/mlx/CMakeLists.txt.
-xgrammar_native_dir=x/mlxrunner/xgrammar/native
+xgrammar_native_dir=mlxrunner/xgrammar/native
 
 # Payload build rules beyond the MLX_VERSION/MLX_C_VERSION pins.
 payload_rule_files=(
   "cmake/local.cmake"
   "cmake/apply-git-patches.cmake"
   "cmake/mlx/CMakeLists.txt"
   "cmake/mlx/CMakePresets.json"
-  "x/mlxrunner/mlx/CMakeLists.txt"
+  "mlx/CMakeLists.txt"
 )
 
 # Build-rule inputs: the rule files plus carried MLX/MLX-C patch content.

.github/workflows/release.yaml

@@ -749,6 +749,10 @@ jobs:
     runs-on: ubuntu-latest
     environment: release
     needs: [darwin-build, windows-app, docker-build-push]
+    concurrency:
+      group: release-publish
+      queue: max
+      cancel-in-progress: false
     permissions:
       contents: write
     env:
@@ -784,28 +788,59 @@ jobs:
       - name: Generate checksum file
         run: find . -type f -not -name 'sha256sum.txt' | xargs sha256sum | tee sha256sum.txt
         working-directory: dist
-      - name: Create or update Release for tag
+      - name: Publish release artifacts
+        shell: bash
         run: |
-          RELEASE_VERSION="$(echo ${GITHUB_REF_NAME} | cut -f1 -d-)"
+          retry() {
+            local attempt
+            for attempt in 1 2 3; do
+              if "$@"; then
+                return 0
+              fi
+              if [ "$attempt" -lt 3 ]; then
+                echo "::warning::Attempt $attempt failed; retrying in $((attempt * 15))s: $*"
+                sleep "$((attempt * 15))"
+              fi
+            done
+            echo "::error::Failed after 3 attempts: $*"
+            return 1
+          }
+
+          delete_assets() {
+            local assets asset
+            # Re-list on each attempt: a failed DELETE may still have removed the asset.
+            assets=$(gh api --paginate "repos/${GITHUB_REPOSITORY}/releases/$1/assets" --jq '.[].url') || return 1
+            while IFS= read -r asset; do
+              [ -z "$asset" ] && continue
+              gh api --method DELETE "$asset" || return 1
+            done <<< "$assets"
+          }
+
+          RELEASE_VERSION="${GITHUB_REF_NAME%%-*}"
           echo "Looking for existing release for ${RELEASE_VERSION}"
-          OLD_TAG=$(gh release ls --json name,tagName | jq -r ".[] | select(.name == \"${RELEASE_VERSION}\") | .tagName")
+          OLD_TAG=$(gh release ls --json name,tagName | jq -r --arg version "$RELEASE_VERSION" '.[] | select(.name == $version) | .tagName')
           if [ -n "$OLD_TAG" ]; then
+            RELEASE_ID=$(gh release view "$OLD_TAG" --json databaseId --jq .databaseId)
+            echo "Deleting old assets from ${OLD_TAG}"
+            retry delete_assets "$RELEASE_ID"
             echo "Updating release ${RELEASE_VERSION} to point to new tag ${GITHUB_REF_NAME}"
-            gh release edit ${OLD_TAG} --tag ${GITHUB_REF_NAME}
+            gh release edit "$OLD_TAG" --tag "$GITHUB_REF_NAME"
           else
             echo "Creating new release ${RELEASE_VERSION} pointing to tag ${GITHUB_REF_NAME}"
-            gh release create ${GITHUB_REF_NAME} \
-              --title ${RELEASE_VERSION} \
+            gh release create "$GITHUB_REF_NAME" \
+              --title "$RELEASE_VERSION" \
               --draft \
               --generate-notes \
               --prerelease
           fi
-      - name: Upload release artifacts
-        run: |
+
+          shopt -s nullglob
           pids=()
           for payload in dist/*.txt dist/*.zip dist/*.tgz dist/*.tar.zst dist/*.exe dist/*.dmg dist/*.ps1 dist/*.sh ; do
+            # Publish the checksum manifest only after every payload has succeeded.
+            [ "$payload" = dist/sha256sum.txt ] && continue
             echo "Uploading $payload"
-            gh release upload ${GITHUB_REF_NAME} $payload --clobber &
+            retry gh release upload "$GITHUB_REF_NAME" "$payload" --clobber &
             pids+=($!)
             sleep 1
           done
@@ -821,4 +856,5 @@ jobs:
             echo "One or more uploads failed"
             exit 1
           fi
+          retry gh release upload "$GITHUB_REF_NAME" dist/sha256sum.txt --clobber
           echo "done"

.github/workflows/test-xgrammar.yaml

@@ -0,0 +1,75 @@
+name: test-xgrammar
+
+on:
+  pull_request:
+    paths:
+      - 'cmake/mlx/CMakeLists.txt'
+      - 'mlxrunner/xgrammar/**'
+      - '.github/workflows/test-xgrammar.yaml'
+
+permissions:
+  contents: read
+
+jobs:
+  standalone:
+    runs-on: macos-latest
+    env:
+      CGO_ENABLED: '1'
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-go@v5
+        with:
+          go-version-file: go.mod
+      - name: Cache MLX Darwin release payload
+        uses: actions/cache@v4
+        with:
+          path: .cache/mlx-darwin-release
+          key: mlx-darwin-${{ hashFiles('MLX_VERSION', 'MLX_C_VERSION', 'cmake/local.cmake', 'cmake/apply-git-patches.cmake', 'cmake/mlx/CMakeLists.txt', 'cmake/mlx/CMakePresets.json', 'mlx/CMakeLists.txt', 'mlx/compat/**', 'mlxrunner/xgrammar/native/**') }}
+      - name: Prepare MLX Darwin release payload
+        run: .github/scripts/prepare_mlx_darwin.sh
+      - name: Install pinned xgrammar into a prefix
+        run: |
+          set -euo pipefail
+          version="$(sed -n 's/^set(XGRAMMAR_VERSION \(.*\))$/\1/p' cmake/mlx/CMakeLists.txt)"
+          [ -n "${version}" ] || { echo "XGRAMMAR_VERSION not found in cmake/mlx/CMakeLists.txt" >&2; exit 1; }
+          echo "XGRAMMAR_VERSION=${version}" >>"${GITHUB_ENV}"
+          src="${RUNNER_TEMP}/xgrammar-src"
+          build="${RUNNER_TEMP}/xgrammar-build"
+          git clone --depth 1 --branch "${version}" --recurse-submodules --shallow-submodules \
+            https://github.com/mlc-ai/xgrammar.git "${src}"
+          # xgrammar's cmake/config.cmake overrides -D options unless the build dir has its own.
+          mkdir -p "${build}"
+          echo 'set(XGRAMMAR_BUILD_PYTHON_BINDINGS OFF)' >"${build}/config.cmake"
+          cmake -S "${src}" -B "${build}" -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX="${RUNNER_TEMP}/xgrammar-prefix"
+          cmake --build "${build}" --parallel
+          cmake --install "${build}"
+      - name: Build xgrammar library against installed xgrammar
+        run: |
+          set -euo pipefail
+          cmake -S mlxrunner/xgrammar/native -B build/xgrammar-standalone \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DCMAKE_PREFIX_PATH="${RUNNER_TEMP}/xgrammar-prefix" \
+            -DXGRAMMAR_VERSION="${XGRAMMAR_VERSION}"
+          cmake --build build/xgrammar-standalone
+      - name: Verify only the ollama_xgrammar API is exported
+        run: |
+          set -euo pipefail
+          lib=build/xgrammar-standalone/libollama_xgrammar.dylib
+          if nm -gU "${lib}" | awk '{print $3}' | grep -v '^_ollama_xgrammar_'; then
+            echo "unexpected exported symbols in ${lib}" >&2
+            exit 1
+          fi
+      - name: Replace the payload's xgrammar library
+        run: |
+          set -euo pipefail
+          for variant in build/lib/ollama/mlx_metal_v*; do
+            cp -f build/xgrammar-standalone/libollama_xgrammar.dylib "${variant}/"
+          done
+      - name: go test
+        run: |
+          set -euo pipefail
+          # The tests skip when the library is missing; treat a skip as failure.
+          go test -count=1 -v ./mlxrunner/xgrammar/... | tee "${RUNNER_TEMP}/xgrammar-test.log"
+          if grep -- '--- SKIP' "${RUNNER_TEMP}/xgrammar-test.log"; then
+            exit 1
+          fi

.github/workflows/test.yaml

@@ -49,12 +49,11 @@ jobs:
             'LLAMA_CPP_VERSION' \
             'MLX_VERSION' \
             'MLX_C_VERSION' \
-            'llama/llama.cpp/**/*' \
-            'ml/backend/ggml/ggml/**/*' \
-            'x/imagegen/mlx/**' \
-            'x/imagegen/mlx/**/*' \
-            'x/mlxrunner/xgrammar/native/**' \
-            'x/mlxrunner/xgrammar/native/**/*' \
+            'mlx/CMakeLists.txt' \
+            'mlx/compat/**' \
+            'mlx/compat/**/*' \
+            'mlxrunner/xgrammar/native/**' \
+            'mlxrunner/xgrammar/native/**/*' \
             '.github/**/*') | tee -a $GITHUB_OUTPUT
           echo app_changed=$(changed 'app/**' 'app/**/*') | tee -a $GITHUB_OUTPUT
           echo go_mod_changed=$(changed 'go.mod') | tee -a $GITHUB_OUTPUT
@@ -419,7 +418,7 @@ jobs:
         with:
           path: .cache/mlx-darwin-release
           # Key on every payload input so a source-built payload survives pushes.
-          key: mlx-darwin-${{ hashFiles('MLX_VERSION', 'MLX_C_VERSION', 'cmake/local.cmake', 'cmake/apply-git-patches.cmake', 'cmake/mlx/CMakeLists.txt', 'cmake/mlx/CMakePresets.json', 'x/mlxrunner/mlx/CMakeLists.txt', 'mlx/compat/**', 'x/mlxrunner/xgrammar/native/**') }}
+          key: mlx-darwin-${{ hashFiles('MLX_VERSION', 'MLX_C_VERSION', 'cmake/local.cmake', 'cmake/apply-git-patches.cmake', 'cmake/mlx/CMakeLists.txt', 'cmake/mlx/CMakePresets.json', 'mlx/CMakeLists.txt', 'mlx/compat/**', 'mlxrunner/xgrammar/native/**') }}
       - name: Prepare MLX Darwin release payload
         if: ${{ startsWith(matrix.os, 'macos') }}
         run: .github/scripts/prepare_mlx_darwin.sh
@@ -439,11 +438,9 @@ jobs:
           cmake -S . -B build/mlx-generate -DOLLAMA_MLX_BACKENDS=cuda_v13
           cmake --build build/mlx-generate --target ollama-mlx-generate-wrappers
           git diff --exit-code -- \
-            x/imagegen/mlx/mlx.h \
-            x/imagegen/mlx/mlx.c \
-            x/mlxrunner/mlx/generated.h \
-            x/mlxrunner/mlx/generated.c \
-            x/mlxrunner/mlx/include/mlx/c
+            mlx/generated.h \
+            mlx/generated.c \
+            mlx/include/mlx/c
       - name: Run go generate
         run: go generate ./...
 
@@ -486,7 +483,7 @@ jobs:
         with:
           path: .cache/mlx-darwin-release
           # Key on every payload input so a source-built payload survives pushes.
-          key: mlx-darwin-${{ hashFiles('MLX_VERSION', 'MLX_C_VERSION', 'cmake/local.cmake', 'cmake/apply-git-patches.cmake', 'cmake/mlx/CMakeLists.txt', 'cmake/mlx/CMakePresets.json', 'x/mlxrunner/mlx/CMakeLists.txt', 'mlx/compat/**', 'x/mlxrunner/xgrammar/native/**') }}
+          key: mlx-darwin-${{ hashFiles('MLX_VERSION', 'MLX_C_VERSION', 'cmake/local.cmake', 'cmake/apply-git-patches.cmake', 'cmake/mlx/CMakeLists.txt', 'cmake/mlx/CMakePresets.json', 'mlx/CMakeLists.txt', 'mlx/compat/**', 'mlxrunner/xgrammar/native/**') }}
       - name: Prepare MLX Darwin release payload
         if: ${{ startsWith(matrix.os, 'macos') }}
         run: .github/scripts/prepare_mlx_darwin.sh

.golangci.yaml

@@ -27,9 +27,14 @@ linters:
           files:
             - $all
             - "!$test"
+            - "!**/mlx/mlxtest/*.go"
           deny:
             - pkg: github.com/ollama/ollama/internal/testutil
               desc: test helpers may only be imported by test files
+            - pkg: github.com/ollama/ollama/mlx/mlxtest
+              desc: test helpers may only be imported by test files
+            - pkg: github.com/ollama/ollama/mlx/mlxthread/mlxthreadtest
+              desc: test helpers may only be imported by test files
     govet:
       disable:
         - unusedresult

Dockerfile

@@ -220,9 +220,8 @@ ENV CGO_LDFLAGS="-L/usr/local/cuda-13/lib64 -L/usr/local/cuda-13/targets/x86_64-
 WORKDIR /go/src/github.com/ollama/ollama
 COPY CMakeLists.txt CMakePresets.json .
 COPY cmake cmake
-COPY mlx/compat mlx/compat
-COPY x/mlxrunner/mlx x/mlxrunner/mlx
-COPY x/mlxrunner/xgrammar/native x/mlxrunner/xgrammar/native
+COPY mlx mlx
+COPY mlxrunner/xgrammar/native mlxrunner/xgrammar/native
 COPY go.mod go.sum .
 COPY MLX_VERSION MLX_C_VERSION .
 RUN curl -fsSL https://golang.org/dl/go$(awk '/^go/ { print $2 }' go.mod).linux-$(case $(uname -m) in x86_64) echo amd64 ;; aarch64) echo arm64 ;; esac).tar.gz | tar xz -C /usr/local

LLAMA_CPP_VERSION

@@ -1 +1 @@
-b10864
+b11081

MLX_VERSION

@@ -1 +1 @@
-d9add9d11f3154111a4c85f267ec2fd307ecd18e
+59d600b5e64c238427d0f8d897ab7c682ef4d3d2

anthropic/anthropic.go

@@ -20,6 +20,7 @@ import (
 	"github.com/ollama/ollama/auth"
 	internalcloud "github.com/ollama/ollama/internal/cloud"
 	"github.com/ollama/ollama/logutil"
+	"github.com/ollama/ollama/types/model"
 )
 
 // Error types matching Anthropic API
@@ -317,8 +318,9 @@ type StreamErrorEvent struct {
 	Error Error  `json:"error"`
 }
 
-// FromMessagesRequest converts an Anthropic MessagesRequest to an Ollama api.ChatRequest
-func FromMessagesRequest(r MessagesRequest) (*api.ChatRequest, error) {
+// FromMessagesRequest converts an Anthropic MessagesRequest to an Ollama api.ChatRequest.
+// An optional thinking descriptor preserves model-defined effort names for rendering.
+func FromMessagesRequest(r MessagesRequest, thinking ...*model.Thinking) (*api.ChatRequest, error) {
 	logutil.Trace("anthropic: converting request", "req", TraceMessagesRequest(r))
 
 	var messages []api.Message
@@ -402,24 +404,27 @@ func FromMessagesRequest(r MessagesRequest) (*api.ChatRequest, error) {
 	}
 
 	var think *api.ThinkValue
-	normalizedEffort := ""
-	if r.OutputConfig != nil {
-		normalizedEffort = strings.ToLower(strings.TrimSpace(r.OutputConfig.Effort))
-		if normalizedEffort == "xhigh" {
-			normalizedEffort = "high"
-		}
-	}
-
 	if r.Thinking != nil && r.Thinking.Type == "enabled" {
 		think = &api.ThinkValue{Value: true}
 	}
 	if r.Thinking != nil && r.Thinking.Type == "disabled" {
 		think = &api.ThinkValue{Value: false}
 	}
 	if think == nil && r.OutputConfig != nil {
-		switch normalizedEffort {
-		case "high", "medium", "low", "max":
-			think = &api.ThinkValue{Value: normalizedEffort}
+		effort := r.OutputConfig.Effort
+		if len(thinking) > 0 && thinking[0].Valid() {
+			if effort != "" {
+				think = &api.ThinkValue{Value: effort}
+			}
+		} else {
+			effort = strings.ToLower(strings.TrimSpace(effort))
+			if effort == "xhigh" {
+				effort = "high"
+			}
+			legacyThink := &api.ThinkValue{Value: effort}
+			if api.ValidateLegacyThinking(legacyThink) == nil {
+				think = legacyThink
+			}
 		}
 	}