Skip to content

Distribute and run LLMs with a single file.

fix(build): add missing link recipes for llama.cpp tools (#1034)

Dex Hunterc73be596a224
diff --git a/llama.cpp.patches/llamafile-files/BUILD.mk b/llama.cpp.patches/llamafile-files/BUILD.mk
index 3b3ccbdb..efcc7941 100644
--- a/llama.cpp.patches/llamafile-files/BUILD.mk
+++ b/llama.cpp.patches/llamafile-files/BUILD.mk
@@ -430,10 +430,20 @@ $(UI_CPP_GEN) $(UI_H_GEN) &: $(UI_EMBED_TOOL) $(UI_ASSETS_INDEX_HTML)
# ==============================================================================
# Tool source files
-TOOL_QUANTIZE_SRCS := llama.cpp/tools/quantize/quantize.cpp
+# quantize, perplexity and llama-bench keep main() in a separate main.cpp
+# (upstream builds the rest of each tool as a reusable *-impl library), so that
+# file has to be listed too or the tool won't link. imatrix defines main() in
+# imatrix.cpp.
+TOOL_QUANTIZE_SRCS := \
+ llama.cpp/tools/quantize/quantize.cpp \
+ llama.cpp/tools/quantize/main.cpp
TOOL_IMATRIX_SRCS := llama.cpp/tools/imatrix/imatrix.cpp
-TOOL_PERPLEXITY_SRCS := llama.cpp/tools/perplexity/perplexity.cpp
-TOOL_BENCH_SRCS := llama.cpp/tools/llama-bench/llama-bench.cpp
+TOOL_PERPLEXITY_SRCS := \
+ llama.cpp/tools/perplexity/perplexity.cpp \
+ llama.cpp/tools/perplexity/main.cpp
+TOOL_BENCH_SRCS := \
+ llama.cpp/tools/llama-bench/llama-bench.cpp \
+ llama.cpp/tools/llama-bench/main.cpp
TOOL_SERVER_SRCS := \
llama.cpp/tools/server/server.cpp \
@@ -598,25 +608,59 @@ o/$(MODE)/llama.cpp/llama-bench/llama-bench \
o/$(MODE)/llama.cpp/server/llama-server: \
private LDLIBS += -lpthread
+# Each tool needs an explicit link recipe. The generic `o/$(MODE)/%:
+# o/$(MODE)/%.o` rule in build/rules.mk cannot link them, because the object
+# lives at o/$(MODE)/llama.cpp/tools/<tool>/<tool>.cpp.o while the executable is
+# o/$(MODE)/llama.cpp/<tool>/<tool>, so the pattern never matches. Without a
+# recipe make treats the target as satisfied and reports "Nothing to be done",
+# which makes a missing binary look like a successful build.
+#
+# They link against the same support objects as llama-server, minus the
+# server-only ones, because llama.cpp.a itself pulls them in:
+# - TOOL_LLAMAFILE_OBJS: src/llama-mmap.cpp and ggml/src/gguf.cpp call
+# llamafile_open_gguf(), llamafile_read(), llamafile_ref(), ...
+# - HTTPLIB_OBJS + mbedtls.a: common/download.cpp, common/hf-cache.cpp and
+# common/license.cpp are built with -DLLAMA_USE_HTTPLIB.
+
o/$(MODE)/llama.cpp/quantize/quantize: \
$(TOOL_QUANTIZE_OBJS) \
+ $(HTTPLIB_OBJS) \
+ $(TOOL_LLAMAFILE_OBJS) \
$$(TINYBLAS_CPU_OBJS) \
- o/$(MODE)/llama.cpp/llama.cpp.a
+ o/$(MODE)/llama.cpp/llama.cpp.a \
+ o/$(MODE)/third_party/mbedtls/mbedtls.a
+ @mkdir -p $(dir $@)
+ $(LINK.o) $^ $(LOADLIBES) $(LDLIBS) -o $@
o/$(MODE)/llama.cpp/imatrix/imatrix: \
$(TOOL_IMATRIX_OBJS) \
+ $(HTTPLIB_OBJS) \
+ $(TOOL_LLAMAFILE_OBJS) \
$$(TINYBLAS_CPU_OBJS) \
- o/$(MODE)/llama.cpp/llama.cpp.a
+ o/$(MODE)/llama.cpp/llama.cpp.a \
+ o/$(MODE)/third_party/mbedtls/mbedtls.a
+ @mkdir -p $(dir $@)
+ $(LINK.o) $^ $(LOADLIBES) $(LDLIBS) -o $@
o/$(MODE)/llama.cpp/perplexity/perplexity: \
$(TOOL_PERPLEXITY_OBJS) \
+ $(HTTPLIB_OBJS) \
+ $(TOOL_LLAMAFILE_OBJS) \
$$(TINYBLAS_CPU_OBJS) \
- o/$(MODE)/llama.cpp/llama.cpp.a
+ o/$(MODE)/llama.cpp/llama.cpp.a \
+ o/$(MODE)/third_party/mbedtls/mbedtls.a
+ @mkdir -p $(dir $@)
+ $(LINK.o) $^ $(LOADLIBES) $(LDLIBS) -o $@
o/$(MODE)/llama.cpp/llama-bench/llama-bench: \
$(TOOL_BENCH_OBJS) \
+ $(HTTPLIB_OBJS) \
+ $(TOOL_LLAMAFILE_OBJS) \
$$(TINYBLAS_CPU_OBJS) \
- o/$(MODE)/llama.cpp/llama.cpp.a
+ o/$(MODE)/llama.cpp/llama.cpp.a \
+ o/$(MODE)/third_party/mbedtls/mbedtls.a
+ @mkdir -p $(dir $@)
+ $(LINK.o) $^ $(LOADLIBES) $(LDLIBS) -o $@
o/$(MODE)/llama.cpp/server/llama-server: \
$(TOOL_SERVER_OBJS) \