Links the local llama.cpp April build (libmtmd exports clip_* internals); generates deterministic PPM test images; dumps preprocessed f32 tensor and projected [n_tokens x 4096] embeddings. Reference stats recorded: radar 512x384 -> 192 tokens, cpu-vs-metal spread ~2e-4 abs on embd values. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
19 lines
728 B
Bash
Executable file
19 lines
728 B
Bash
Executable file
#!/usr/bin/env bash
|
|
# Build the clip.cpp oracle against the local llama.cpp checkout.
|
|
set -euo pipefail
|
|
LLAMA_CPP="${LLAMA_CPP:-/Users/admin/llama.cpp}"
|
|
# NOTE: build/ is stale (Aug 2025); build-arm64-apple-clang-release matches the
|
|
# April 2026 source and has qwen3vl mmproj support.
|
|
LLAMA_BIN="${LLAMA_BIN:-$LLAMA_CPP/build-arm64-apple-clang-release/bin}"
|
|
DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
|
|
|
clang++ -std=c++17 -O2 \
|
|
-I "$LLAMA_CPP/tools/mtmd" \
|
|
-I "$LLAMA_CPP/ggml/include" \
|
|
-I "$LLAMA_CPP/include" \
|
|
-I "$LLAMA_CPP/vendor" \
|
|
"$DIR/clip_dump.cpp" \
|
|
-L "$LLAMA_BIN" -lmtmd -lggml -lggml-base -lllama \
|
|
-Wl,-rpath,"$LLAMA_BIN" \
|
|
-o "$DIR/clip_dump"
|
|
echo "built $DIR/clip_dump"
|