From 5d9cfc4ea03ccdd33cdeb9532da97a1ce7267f1e Mon Sep 17 00:00:00 2001 From: karl Date: Tue, 13 Oct 2026 08:13:33 +0200 Subject: [PATCH] Verify packaged workers from Unicode extraction directories --- docs/WINDOWS-UNICODE.md | 20 ++++++++++++++++++-- tools/embeddinggemma2/package_smoke.py | 3 ++- 2 files changed, 20 insertions(+), 3 deletions(-) diff --git a/docs/WINDOWS-UNICODE.md b/docs/WINDOWS-UNICODE.md index 256772c..ce0a0c5 100644 --- a/docs/WINDOWS-UNICODE.md +++ b/docs/WINDOWS-UNICODE.md @@ -57,6 +57,22 @@ $env:SEARCHI_GEMMA_WORKER = 'PATH/TO/CURRENT/searchi-gemma-worker.exe' cargo test -p searchi-semantic --test unicode_windows_cpu --locked -- --ignored --nocapture ``` -Native package restaging and installed Unicode-profile paths require their own -checks. This focused model-path regression does not certify every filesystem, +## Updated development archive + +The normal worker was restaged and the embedded desktop Release rebuilt in +2m 02s. Executable: 23,924,224 bytes, SHA-256 +`bce713e7d081095e650756d1b0ed497cb21e451ea2e47bead6676a22327729b9`. +The unsigned Windows archive is 145,409,579 bytes, SHA-256 +`512be4960da197a45710d49bdbc27b84c18e0270364e41bbcc546b8791826842`. + +Both actual package checks extracted under a Unicode directory and verified all +39 recorded files. Default `auto` mode used CPU with Vulkan registration disabled +only in the child: optimized indexing context 2048 and baseline-only query +context 512 both produced normalized 768-dimensional real Google vectors. +The packaged FFprobe started successfully in that directory. Local reports are +`.searchi/packages/smoke-unicode-{indexing,query-baseline}.json`. + +These are worker/dependency checks outside the build tree, not complete installed +desktop interaction, Unicode user-profile, clean-machine, signing or older +physical-CPU certification. This regression does not certify every filesystem, codec, launcher or installer path, or any GPU hardware. diff --git a/tools/embeddinggemma2/package_smoke.py b/tools/embeddinggemma2/package_smoke.py index f845c7c..e70553b 100644 --- a/tools/embeddinggemma2/package_smoke.py +++ b/tools/embeddinggemma2/package_smoke.py @@ -31,7 +31,7 @@ def main(): assert hashlib.file_digest(stream, "sha256").hexdigest() == manifest["files"][0]["sha256"] parent = ROOT / ".searchi/packages" parent.mkdir(parents=True, exist_ok=True) - with tempfile.TemporaryDirectory(prefix="smoke-", dir=parent) as folder: + with tempfile.TemporaryDirectory(prefix="smoke-Gemäß_日本_😀-", dir=parent) as folder: target = Path(folder).resolve() assert target.is_relative_to(parent.resolve()) with zipfile.ZipFile(args.archive) as archive: @@ -82,6 +82,7 @@ def main(): "vulkan_module_loaded": hello.get("vulkan_module_loaded"), "baseline_only": args.baseline, "worker_role": hello["role"], "context_tokens": hello["context_tokens"], + "unicode_extraction_directory": True, "archive_sha256": archive_checksum, "worker_requested_device": "auto", "worker_actual_device": hello["device"], "no_gpu_condition": "GGML_DISABLE_VULKAN=1 only in child; SEARCHI_FORCE_CPU absent",