feat: add whisper.cpp. use go:embed to archive .a and .h files

.github/workflows/ci.yaml

@@ -39,6 +39,8 @@ jobs:
           go-version: "1.24"
       - name: Build with CGO (llamacpp tag)
         run: CGO_ENABLED=1 go build -tags llamacpp ./ggml/llamacpp/...
+      - name: Build with CGO (whispercpp tag)
+        run: CGO_ENABLED=1 go build -tags whispercpp ./ggml/whispercpp/...
 
   build-libs:
     name: Build Libraries from Source (${{ matrix.os }})
@@ -68,9 +70,16 @@ jobs:
       - name: Build static libraries from source
         run: make build-libs
       - name: Verify CGO build with fresh libraries
-        run: CGO_ENABLED=1 go build -tags llamacpp ./ggml/llamacpp/...
-      - name: Upload prebuilt artifacts
+        run: |
+          CGO_ENABLED=1 go build -tags llamacpp ./ggml/llamacpp/...
+          CGO_ENABLED=1 go build -tags whispercpp ./ggml/whispercpp/...
+      - name: Upload llama.cpp prebuilt artifacts
+        uses: actions/upload-artifact@v4
+        with:
+          name: prebuilt-llamacpp-${{ matrix.platform }}
+          path: ggml/llamacpp/third_party/prebuilt/${{ matrix.platform }}/*.a
+      - name: Upload whisper.cpp prebuilt artifacts
         uses: actions/upload-artifact@v4
         with:
-          name: prebuilt-${{ matrix.platform }}
-          path: third_party/llama.cpp/prebuilt/${{ matrix.platform }}/*.a
+          name: prebuilt-whispercpp-${{ matrix.platform }}
+          path: ggml/whispercpp/third_party/prebuilt/${{ matrix.platform }}/*.a

.gitignore

@@ -1,7 +1,7 @@
 # Build artifacts (source downloads, cmake dirs)
-third_party/llama.cpp/src/
-third_party/whisper.cpp/src/
-third_party/**/build/
+ggml/llamacpp/third_party/src/
+ggml/whispercpp/third_party/src/
+**/build/
 out/
 
 # Prebuilt .a files and headers ARE committed — do not ignore them

Dockerfile.libs

@@ -1,4 +1,4 @@
-# Dockerfile.libs — build linux-amd64 static libraries for llama.cpp
+# Dockerfile.libs — build linux-amd64 static libraries for llama.cpp and whisper.cpp
 #
 # Usage:
 #   docker build -f Dockerfile.libs -o ./out .
@@ -18,27 +18,46 @@ COPY go.mod ./
 COPY version.go ./
 COPY cmd/versioncmd/ ./cmd/versioncmd/
 
-# Read versions and download sources
+# Download llama.cpp
 RUN LLAMA_VERSION=$(go run ./cmd/versioncmd llama.cpp) && \
     echo "Downloading llama.cpp ${LLAMA_VERSION}..." && \
     wget -qO llama.cpp.tar.gz "https://github.com/ggerganov/llama.cpp/archive/refs/tags/${LLAMA_VERSION}.tar.gz" && \
     mkdir -p llama-src && \
     tar xzf llama.cpp.tar.gz --strip-components=1 -C llama-src && \
     rm llama.cpp.tar.gz
 
+# Download whisper.cpp
+RUN WHISPER_VERSION=$(go run ./cmd/versioncmd whisper.cpp) && \
+    echo "Downloading whisper.cpp ${WHISPER_VERSION}..." && \
+    wget -qO whisper.cpp.tar.gz "https://github.com/ggerganov/whisper.cpp/archive/refs/tags/${WHISPER_VERSION}.tar.gz" && \
+    mkdir -p whisper-src && \
+    tar xzf whisper.cpp.tar.gz --strip-components=1 -C whisper-src && \
+    rm whisper.cpp.tar.gz
+
 # Build llama.cpp
 RUN cd llama-src && \
     cmake -B build -DBUILD_SHARED_LIBS=OFF && \
     cmake --build build --config Release -j$(nproc)
 
-# Collect artifacts
+# Build whisper.cpp
+RUN cd whisper-src && \
+    cmake -B build -DBUILD_SHARED_LIBS=OFF && \
+    cmake --build build --config Release -j$(nproc)
+
+# Collect llama.cpp artifacts
 RUN mkdir -p /out/llama.cpp/linux-amd64 /out/llama.cpp/include /out/llama.cpp/ggml/include /out/llama.cpp/common && \
     find llama-src/build -name "*.a" -exec cp {} /out/llama.cpp/linux-amd64/ \; && \
     cp llama-src/include/*.h /out/llama.cpp/include/ && \
     cp llama-src/ggml/include/*.h /out/llama.cpp/ggml/include/ && \
     cp llama-src/common/common.h /out/llama.cpp/common/ && \
     cp llama-src/common/sampling.h /out/llama.cpp/common/
 
+# Collect whisper.cpp artifacts
+RUN mkdir -p /out/whisper.cpp/linux-amd64 /out/whisper.cpp/include /out/whisper.cpp/ggml/include && \
+    find whisper-src/build -name "*.a" -exec cp {} /out/whisper.cpp/linux-amd64/ \; && \
+    cp whisper-src/include/*.h /out/whisper.cpp/include/ && \
+    cp whisper-src/ggml/include/*.h /out/whisper.cpp/ggml/include/
+
 # Output stage — docker build -o extracts from here
 FROM scratch
 COPY --from=builder /out/ /

Makefile

@@ -2,11 +2,13 @@
 #
 # For consumers:
 #   go get github.com/footprintai/go-nativeml
-#   go build -tags llamacpp ./...    # just works — prebuilt .a files are in the module
+#   go build -tags llamacpp ./...     # just works — prebuilt .a files are in the module
+#   go build -tags whispercpp ./...   # whisper.cpp bindings
 #
 # For maintainers (rebuild .a files from source):
 #   make build-libs              # Build all libraries for current platform
 #   make build-libs-llama        # Build llama.cpp only
+#   make build-libs-whisper      # Build whisper.cpp only
 #   make build-libs-linux        # Build linux-amd64 .a files via Docker
 #   make build-libs-all          # Build native + linux-amd64
 #   make clean                   # Remove temp build dirs (keeps prebuilt .a + headers)
@@ -21,15 +23,18 @@ PLATFORM := $(shell go env GOOS)-$(shell go env GOARCH)
 # Parallel build cores
 NPROC := $(shell if which nproc > /dev/null 2>&1; then nproc; elif [ "$$(uname)" = "Darwin" ]; then sysctl -n hw.ncpu; else echo 4; fi)
 
-# Paths
-THIRD_PARTY := third_party
-LLAMA_DIR := $(THIRD_PARTY)/llama.cpp
-LLAMA_SRC := $(LLAMA_DIR)/src
-LLAMA_PREBUILT := $(LLAMA_DIR)/prebuilt/$(PLATFORM)
+# Paths — assets live inside the Go package directories
+LLAMA_THIRD_PARTY := ggml/llamacpp/third_party
+LLAMA_SRC := $(LLAMA_THIRD_PARTY)/src
+LLAMA_PREBUILT := $(LLAMA_THIRD_PARTY)/prebuilt/$(PLATFORM)
 
-.PHONY: build-libs build-libs-llama build-libs-linux build-libs-all clean verify
+WHISPER_THIRD_PARTY := ggml/whispercpp/third_party
+WHISPER_SRC := $(WHISPER_THIRD_PARTY)/src
+WHISPER_PREBUILT := $(WHISPER_THIRD_PARTY)/prebuilt/$(PLATFORM)
 
-build-libs: build-libs-llama
+.PHONY: build-libs build-libs-llama build-libs-whisper build-libs-linux build-libs-all clean verify
+
+build-libs: build-libs-llama build-libs-whisper
 
 # Build both native platform and linux-amd64 (via Docker)
 build-libs-all: build-libs build-libs-linux
@@ -46,13 +51,13 @@ $(LLAMA_PREBUILT): $(LLAMA_SRC)
 	@mkdir -p $(LLAMA_PREBUILT)
 	find $(LLAMA_SRC)/build -name "*.a" -exec cp {} $(LLAMA_PREBUILT)/ \;
 	@echo "==> Copying llama.cpp headers..."
-	@mkdir -p $(LLAMA_DIR)/include
-	cp $(LLAMA_SRC)/include/*.h $(LLAMA_DIR)/include/
-	@mkdir -p $(LLAMA_DIR)/ggml/include
-	cp $(LLAMA_SRC)/ggml/include/*.h $(LLAMA_DIR)/ggml/include/
-	@mkdir -p $(LLAMA_DIR)/common
-	cp $(LLAMA_SRC)/common/common.h $(LLAMA_DIR)/common/
-	cp $(LLAMA_SRC)/common/sampling.h $(LLAMA_DIR)/common/
+	@mkdir -p $(LLAMA_THIRD_PARTY)/include
+	cp $(LLAMA_SRC)/include/*.h $(LLAMA_THIRD_PARTY)/include/
+	@mkdir -p $(LLAMA_THIRD_PARTY)/ggml/include
+	cp $(LLAMA_SRC)/ggml/include/*.h $(LLAMA_THIRD_PARTY)/ggml/include/
+	@mkdir -p $(LLAMA_THIRD_PARTY)/common
+	cp $(LLAMA_SRC)/common/common.h $(LLAMA_THIRD_PARTY)/common/
+	cp $(LLAMA_SRC)/common/sampling.h $(LLAMA_THIRD_PARTY)/common/
 	@echo "==> llama.cpp $(LLAMA_VERSION) ready: $(LLAMA_PREBUILT)/"
 
 $(LLAMA_SRC):
@@ -62,37 +67,71 @@ $(LLAMA_SRC):
 	tar xzf llama.cpp.tar.gz --strip-components=1 -C $(LLAMA_SRC)
 	rm llama.cpp.tar.gz
 
+# ============================================================================
+# whisper.cpp
+# ============================================================================
+build-libs-whisper: $(WHISPER_PREBUILT)
+
+$(WHISPER_PREBUILT): $(WHISPER_SRC)
+	@echo "==> Building whisper.cpp $(WHISPER_VERSION) for $(PLATFORM)..."
+	cd $(WHISPER_SRC) && cmake -B build -DBUILD_SHARED_LIBS=OFF && \
+		cmake --build build --config Release -j$(NPROC)
+	@mkdir -p $(WHISPER_PREBUILT)
+	find $(WHISPER_SRC)/build -name "*.a" -exec cp {} $(WHISPER_PREBUILT)/ \;
+	@echo "==> Copying whisper.cpp headers..."
+	@mkdir -p $(WHISPER_THIRD_PARTY)/include
+	cp $(WHISPER_SRC)/include/*.h $(WHISPER_THIRD_PARTY)/include/
+	@mkdir -p $(WHISPER_THIRD_PARTY)/ggml/include
+	cp $(WHISPER_SRC)/ggml/include/*.h $(WHISPER_THIRD_PARTY)/ggml/include/
+	@echo "==> whisper.cpp $(WHISPER_VERSION) ready: $(WHISPER_PREBUILT)/"
+
+$(WHISPER_SRC):
+	@echo "==> Downloading whisper.cpp $(WHISPER_VERSION)..."
+	wget -qO whisper.cpp.tar.gz https://github.com/ggerganov/whisper.cpp/archive/refs/tags/$(WHISPER_VERSION).tar.gz
+	mkdir -p $(WHISPER_SRC)
+	tar xzf whisper.cpp.tar.gz --strip-components=1 -C $(WHISPER_SRC)
+	rm whisper.cpp.tar.gz
+
 # ============================================================================
 # Docker build for linux-amd64 (cross-compile from macOS)
 # ============================================================================
 build-libs-linux:
 	@echo "==> Building linux-amd64 static libraries via Docker..."
 	docker build -f Dockerfile.libs -o ./out .
-	@mkdir -p $(LLAMA_DIR)/prebuilt/linux-amd64
-	cp out/llama.cpp/linux-amd64/*.a $(LLAMA_DIR)/prebuilt/linux-amd64/
-	@# Copy headers if not already present
-	@mkdir -p $(LLAMA_DIR)/include $(LLAMA_DIR)/ggml/include $(LLAMA_DIR)/common
-	cp out/llama.cpp/include/*.h $(LLAMA_DIR)/include/
-	cp out/llama.cpp/ggml/include/*.h $(LLAMA_DIR)/ggml/include/
-	cp out/llama.cpp/common/common.h $(LLAMA_DIR)/common/
-	cp out/llama.cpp/common/sampling.h $(LLAMA_DIR)/common/
+	@# llama.cpp
+	@mkdir -p $(LLAMA_THIRD_PARTY)/prebuilt/linux-amd64
+	cp out/llama.cpp/linux-amd64/*.a $(LLAMA_THIRD_PARTY)/prebuilt/linux-amd64/
+	@mkdir -p $(LLAMA_THIRD_PARTY)/include $(LLAMA_THIRD_PARTY)/ggml/include $(LLAMA_THIRD_PARTY)/common
+	cp out/llama.cpp/include/*.h $(LLAMA_THIRD_PARTY)/include/
+	cp out/llama.cpp/ggml/include/*.h $(LLAMA_THIRD_PARTY)/ggml/include/
+	cp out/llama.cpp/common/common.h $(LLAMA_THIRD_PARTY)/common/
+	cp out/llama.cpp/common/sampling.h $(LLAMA_THIRD_PARTY)/common/
+	@# whisper.cpp
+	@mkdir -p $(WHISPER_THIRD_PARTY)/prebuilt/linux-amd64
+	cp out/whisper.cpp/linux-amd64/*.a $(WHISPER_THIRD_PARTY)/prebuilt/linux-amd64/
+	@mkdir -p $(WHISPER_THIRD_PARTY)/include $(WHISPER_THIRD_PARTY)/ggml/include
+	cp out/whisper.cpp/include/*.h $(WHISPER_THIRD_PARTY)/include/
+	cp out/whisper.cpp/ggml/include/*.h $(WHISPER_THIRD_PARTY)/ggml/include/
 	rm -rf out
 	@echo "==> linux-amd64 libraries ready"
 
 # ============================================================================
 # Verification
 # ============================================================================
 verify:
-	@echo "==> Verifying stub build (no tag)..."
+	@echo "==> Verifying stub builds (no tags)..."
 	go build ./ggml/llamacpp/...
-	@echo "==> Verifying CGO build (with tag)..."
+	go build ./ggml/whispercpp/...
+	@echo "==> Verifying CGO builds (with tags)..."
 	CGO_ENABLED=1 go build -tags llamacpp ./ggml/llamacpp/...
+	CGO_ENABLED=1 go build -tags whispercpp ./ggml/whispercpp/...
 	@echo "==> Running stub tests..."
 	go test ./ggml/llamacpp/...
+	go test ./ggml/whispercpp/...
 	@echo "==> All checks passed"
 
 # ============================================================================
 # Cleanup
 # ============================================================================
 clean:
-	rm -rf $(LLAMA_SRC) out
+	rm -rf $(LLAMA_SRC) $(WHISPER_SRC) out