Distribute and run LLMs with a single file.
C++ cross-platformggufllama-cpplocal-ailocal-inferencelocal-llmopen-source-aisingle-file-executablespeech-to-text
#-*-mode:makefile-gmake;indent-tabs-mode:t;tab-width:8;coding:utf-8-*-┐
#── vi: set noet ft=make ts=8 sw=8 fenc=utf-8 :vi ────────────────────┘
PKGS += TESTS
include tests/sgemm/BUILD.mk
include tests/strsm/BUILD.mk
# ==============================================================================
# Include paths (reuse llamafile includes)
# ==============================================================================
TESTS_CPPFLAGS := $(LLAMAFILE_INCLUDES)
# ==============================================================================
# Test: extract_data_uris_test
# ==============================================================================
# Dependencies for extract_data_uris test:
# - extract_data_uris.o: contains extract_data_uris function (isolated)
# - datauri.o: DataUri class for parsing data URIs
# - image.o: is_image function for validating images
# - string.o: lf::startscasewith helper
# - xterm.o: terminal utilities (required by image.o)
# - stb.a: stb_image for image validation
EXTRACT_DATA_URIS_TEST_DEPS := \
o/$(MODE)/llamafile/extract_data_uris.o \
o/$(MODE)/llamafile/datauri.o \
o/$(MODE)/llamafile/image.o \
o/$(MODE)/llamafile/string.o \
o/$(MODE)/llamafile/xterm.o \
o/$(MODE)/third_party/stb/stb.a \
o/$(MODE)/llama.cpp/common/build-info.cpp.o \
o/$(MODE)/llama.cpp/common/jinja/caps.cpp.o \
o/$(MODE)/llama.cpp/common/jinja/lexer.cpp.o \
o/$(MODE)/llama.cpp/common/jinja/parser.cpp.o \
o/$(MODE)/llama.cpp/common/jinja/runtime.cpp.o \
o/$(MODE)/llama.cpp/common/jinja/string.cpp.o \
o/$(MODE)/llama.cpp/common/jinja/value.cpp.o
o/$(MODE)/tests/extract_data_uris_test.o: tests/extract_data_uris_test.cpp
@mkdir -p $(@D)
$(CXX) $(CXXFLAGS) $(TESTS_CPPFLAGS) -c -o $@ $<
o/$(MODE)/tests/extract_data_uris_test: \
o/$(MODE)/tests/extract_data_uris_test.o \
$(EXTRACT_DATA_URIS_TEST_DEPS)
@mkdir -p $(@D)
$(CXX) $(LDFLAGS) -o $@ $^ $(LDLIBS)
# ==============================================================================
# Test: fa_helpers_test (issue #975 numerical equivalence)
# ==============================================================================
#
# Compares llamafile_fa_vec_dot_f16 / llamafile_fa_fp16_to_fp32_row
# against upstream ggml_vec_dot_f16 / ggml_fp16_to_fp32_row on the
# same random + edge-case inputs. Catches numerical regressions in
# the alternative AVX-512F implementations we ship via sgemm.cpp's
# dispatch. On CPUs without AVX-512F the helpers report unsupported
# and the corresponding assertions are skipped.
FA_HELPERS_TEST_DEPS := \
o/$(MODE)/llamafile/sgemm.o \
o/$(MODE)/llamafile/llamafile.o \
o/$(MODE)/llamafile/fa_helpers_amd_avx512f.o \
o/$(MODE)/llamafile/fa_helpers_unsupported.o \
o/$(MODE)/llamafile/fa_simd_gemm_amd_avx512f.o \
o/$(MODE)/llamafile/tinyblas_cpu_sgemm_amd_avx.o \
o/$(MODE)/llamafile/tinyblas_cpu_sgemm_amd_fma.o \
o/$(MODE)/llamafile/tinyblas_cpu_sgemm_amd_avx2.o \
o/$(MODE)/llamafile/tinyblas_cpu_sgemm_amd_avxvnni.o \
o/$(MODE)/llamafile/tinyblas_cpu_sgemm_amd_avx512f.o \
o/$(MODE)/llamafile/tinyblas_cpu_sgemm_amd_zen4.o \
o/$(MODE)/llamafile/tinyblas_cpu_sgemm_arm80.o \
o/$(MODE)/llamafile/tinyblas_cpu_sgemm_arm82.o \
o/$(MODE)/llamafile/tinyblas_cpu_unsupported.o \
o/$(MODE)/llamafile/tinyblas_cpu_mixmul_amd_avx.o \
o/$(MODE)/llamafile/tinyblas_cpu_mixmul_amd_fma.o \
o/$(MODE)/llamafile/tinyblas_cpu_mixmul_amd_avx2.o \
o/$(MODE)/llamafile/tinyblas_cpu_mixmul_amd_avxvnni.o \
o/$(MODE)/llamafile/tinyblas_cpu_mixmul_amd_avx512f.o \
o/$(MODE)/llamafile/tinyblas_cpu_mixmul_amd_zen4.o \
o/$(MODE)/llamafile/tinyblas_cpu_mixmul_arm80.o \
o/$(MODE)/llamafile/tinyblas_cpu_mixmul_arm82.o \
o/$(MODE)/llamafile/iqk_mul_mat_amd_avx2.o \
o/$(MODE)/llamafile/iqk_mul_mat_amd_zen4.o \
o/$(MODE)/llamafile/iqk_mul_mat_arm82.o \
o/$(MODE)/llama.cpp/llama.cpp.a
o/$(MODE)/tests/fa_helpers_test.o: tests/fa_helpers_test.cpp
@mkdir -p $(@D)
$(CXX) $(CXXFLAGS) $(TESTS_CPPFLAGS) -fopenmp -c -o $@ $<
o/$(MODE)/tests/fa_helpers_test: \
o/$(MODE)/tests/fa_helpers_test.o \
$(FA_HELPERS_TEST_DEPS)
@mkdir -p $(@D)
$(CXX) $(LDFLAGS) -fopenmp -o $@ $^ $(LDLIBS)
# ==============================================================================
# Test: gpu_backend_test (issue #988 device-count gate / fallback)
# ==============================================================================
#
# Exercises the shared GPU backend probe core. The test injects stub entry
# points into a GpuBackend and provides its own doubles for gpu_backend.c's
# externs, so it links against gpu_backend.o alone — no DSO, no GPU, no
# llamafile.o/llama.cpp.a.
GPU_BACKEND_TEST_DEPS := \
o/$(MODE)/llamafile/gpu_backend.o
o/$(MODE)/tests/gpu_backend_test.o: tests/gpu_backend_test.cpp
@mkdir -p $(@D)
$(CXX) $(CXXFLAGS) $(TESTS_CPPFLAGS) -c -o $@ $<
o/$(MODE)/tests/gpu_backend_test: \
o/$(MODE)/tests/gpu_backend_test.o \
$(GPU_BACKEND_TEST_DEPS)
@mkdir -p $(@D)
$(CXX) $(LDFLAGS) -o $@ $^ $(LDLIBS)
# ==============================================================================
# Test: sandbox_test (issue #930 pledge/SECCOMP sandboxing)
# ==============================================================================
#
# Forks children that install the promise sets llamafile uses (server:
# stdio+anet, cli: stdio+rpath+tty) and asserts allowed syscalls still
# work while blocked ones fail with EPERM. Self-skips on OSes where
# pledge() can't be enforced (anything but Linux/OpenBSD). Links against
# sandbox.o alone: it stubs llamafile_has_gpu() to stay off the GPU stack.
SANDBOX_TEST_DEPS := \
o/$(MODE)/llamafile/sandbox.o
# sandbox_test.o is built by the generic %.o: %.c rule (-iquote. resolves
# the llamafile/llamafile.h include)
o/$(MODE)/tests/sandbox_test: \
o/$(MODE)/tests/sandbox_test.o \
$(SANDBOX_TEST_DEPS)
@mkdir -p $(@D)
$(CC) $(LDFLAGS) -o $@ $^ $(LDLIBS)
# ==============================================================================
# Test: backend_ops_test (upstream ggml test-backend-ops under llamafile)
# ==============================================================================
#
# Compares every ggml op on each runtime-loaded GPU backend against the CPU
# reference (NMSE thresholds). Used to verify numerical consistency of the
# dlopen'd backend DSOs (e.g. CPU vs Vulkan matmuls, issue #938 follow-up).
# Not part of the default check suite: it needs a GPU and the backend DSO at
# runtime. Build with: make o/$(MODE)/tests/backend_ops_test
# upstream test uses #include <ggml.h>, so the ggml include dirs must be on
# the system search path (-iquote from LLAMAFILE_INCLUDES is not enough)
BACKEND_OPS_INCLUDES := \
-isystem llama.cpp/ggml/include \
-isystem llama.cpp/include
o/$(MODE)/tests/test_backend_ops_impl.o: llama.cpp/tests/test-backend-ops.cpp
@mkdir -p $(@D)
$(CXX) $(CXXFLAGS) $(TESTS_CPPFLAGS) $(BACKEND_OPS_INCLUDES) -Dmain=backend_ops_main -c -o $@ $<
o/$(MODE)/tests/backend_ops_harness.o: tests/backend_ops_harness.cpp
@mkdir -p $(@D)
$(CXX) $(CXXFLAGS) $(TESTS_CPPFLAGS) -c -o $@ $<
o/$(MODE)/tests/backend_ops_test: \
o/$(MODE)/tests/backend_ops_harness.o \
o/$(MODE)/tests/test_backend_ops_impl.o \
o/$(MODE)/llamafile/llamafile.o \
o/$(MODE)/llamafile/gpu.a \
o/$(MODE)/llamafile/zip.o \
o/$(MODE)/llamafile/check_cpu.o \
$(GGML_OBJS) \
$(TINYBLAS_CPU_OBJS)
@mkdir -p $(@D)
$(CXX) $(LDFLAGS) -o $@ $^ $(LDLIBS)
# ==============================================================================
# Smoke test: transcribefile APE
# ==============================================================================
#
# Subprocess-style regression on the transcribefile binary itself.
# Layer 1: model-free probes (--help, WAV duration line) — always runs.
# Layer 2: parakeet end-to-end, gated on TRANSCRIBEFILE_PARAKEET_GGUF
# (skip-on-missing, never fails `make check` for absent model).
# The script is the test; we don't have an executable to plug %.runs into,
# so we define a one-off recipe that emits the stamp on success.
o/$(MODE)/tests/transcribefile_smoke.runs: \
o/$(MODE)/transcribefile/transcribefile \
tests/transcribefile_smoke.sh
@mkdir -p $(@D)
bash tests/transcribefile_smoke.sh o/$(MODE)/transcribefile/transcribefile
@touch $@
# ==============================================================================
# Phony targets
# ==============================================================================
.PHONY: o/$(MODE)/tests
o/$(MODE)/tests: \
o/$(MODE)/tests/extract_data_uris_test.runs \
o/$(MODE)/tests/fa_helpers_test.runs \
o/$(MODE)/tests/gpu_backend_test.runs \
o/$(MODE)/tests/sandbox_test.runs \
o/$(MODE)/tests/transcribefile_smoke.runs