transcribe-cpp-native 0.1.0__tar.gz → 0.1.1__tar.gz

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/AGENTS.md

@@ -39,6 +39,25 @@ scripts/ci/clang-format.sh --check    # verify, no changes
   formatted. CI gates our C/C++ in
   `.github/workflows/clang-format.yml`.
 
+## C ABI Exception Discipline
+
+No C++ exception may escape a public entry point.
+
+- A new public entry point must either route through an `api_guard_*`
+  wrapper (`src/transcribe.cpp`) or be nothrow by construction. Device and
+  registry queries are not pure reads; guard them.
+- Entry points with ownership out-params enforce "non-OK => `*out == NULL`,
+  nothing leaked" in their forwarders on every error return.
+- Teardown never uses raw `ggml_backend_free` / `ggml_backend_buffer_free` /
+  `ggml_backend_sched_free` in library code: use `transcribe::safe_*` from
+  `src/transcribe-backend.h`. `tests/lint_teardown.cmake` fails CI on
+  violations.
+- Host log callbacks are contained at the emission site
+  (`transcribe_log_invoke`).
+- The `TRANSCRIBE_TEST_*` fault hooks (`_DEV_INIT_THROW`, `_TEARDOWN_THROW`)
+  intentionally ship in release artifacts for wheel clean-install CI.
+  Present-but-empty values are inert.
+
 ## Verification
 
 - End-to-end numerical checks:

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/CMakeLists.txt

@@ -387,6 +387,15 @@ if(TRANSCRIBE_GGML_BACKEND_DL)
     set(GGML_NATIVE OFF CACHE BOOL "" FORCE)
 endif()
 
+# MSVC: use /EHs so exceptions thrown by ggml's C-ABI frames unwind to
+# transcribe.cpp's public-entry guards. Apply before ggml and src are added.
+if(MSVC)
+    string(REPLACE "/EHsc" "/EHs" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
+    if(NOT CMAKE_CXX_FLAGS MATCHES "/EHs")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /EHs")
+    endif()
+endif()
+
 # ggml's CMake declares its own warning flags; let it.
 add_subdirectory(ggml)
 

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/Cargo.lock

@@ -315,7 +315,7 @@ dependencies = [
 
 [[package]]
 name = "transcribe-cpp"
-version = "0.1.0"
+version = "0.1.1"
 dependencies = [
  "hound",
  "log",
@@ -325,7 +325,7 @@ dependencies = [
 
 [[package]]
 name = "transcribe-cpp-sys"
-version = "0.1.0"
+version = "0.1.1"
 dependencies = [
  "cmake",
  "serde_json",

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/Cargo.toml

@@ -11,7 +11,7 @@
 
 [package]
 name = "transcribe-cpp-sys"
-version = "0.1.0"
+version = "0.1.1"
 edition = "2021"
 rust-version = "1.74"
 description = "Native FFI bindings for transcribe.cpp, a C/C++ speech-to-text library built on ggml"

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: transcribe-cpp-native
-Version: 0.1.0
+Version: 0.1.1
 Summary: Prebuilt native library for transcribe-cpp (speech-to-text, ggml)
 Keywords: transcription,speech-to-text,asr,ggml,native
 Author: The transcribe.cpp authors

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/README.md

@@ -1,6 +1,8 @@
 # transcribe.cpp
 
-C/C++ speech-to-text inference library. Runs diverse STT model families via [GGUF](https://github.com/ggerganov/gguf) models on the [ggml](https://github.com/ggml-org/ggml) runtime, with Metal, Vulkan, and CUDA backends for fast GPU inference.
+C/C++ speech-to-text inference library. Runs diverse STT model families via [GGUF](https://github.com/ggerganov/gguf) models on the [ggml](https://github.com/ggml-org/ggml) runtime, with Metal, Vulkan, and CUDA backends for fast GPU inference plus a tinyBLAS-accelerated CPU path.
+
+16 model families and 60+ variants, streaming and batch. Every model we publish under [`handy-computer`](https://huggingface.co/handy-computer) is numerically verified and WER-tested against its reference implementation
 
 **Supported models:**
 
@@ -53,6 +55,8 @@ cmake --build build
 
 `libopenblas-dev` is optional but recommended. It accelerates the host-side decoder ~10-15x. Without it the build falls back to a scalar path automatically.
 
+tinyBLAS (Justine Tunney's `llamafile_sgemm` kernels) is on by default.
+
 To build the quantization tool:
 
 ```bash
@@ -108,6 +112,20 @@ Input must be 16 kHz mono WAV. Use `ffmpeg` or `sox` to convert other formats:
 ffmpeg -i input.mp3 -ar 16000 -ac 1 output.wav
 ```
 
+## Bindings
+
+Official bindings wrap the C API for other languages:
+
+| Language | Path |
+| --- | --- |
+| Python | [bindings/python](bindings/python) |
+| TypeScript / JavaScript | [bindings/typescript](bindings/typescript) |
+| Rust | [bindings/rust/transcribe-cpp](bindings/rust/transcribe-cpp) |
+| Swift / ObjC | [bindings/swift](bindings/swift) |
+
+See [`docs/bindings.md`](docs/bindings.md) for how the bindings are generated
+and kept in sync with the header.
+
 ## Tests
 
 ```bash
@@ -168,6 +186,7 @@ src/arch/parakeet/         Parakeet family implementation
 src/arch/cohere/           Cohere Transcribe family implementation
 examples/cli/              CLI binary source
 tools/transcribe-quantize/ Quantization tool source
+bindings/                  Python, TypeScript, Rust, and Swift bindings
 docs/                      Porting and validation guidance
 scripts/                   Python converter + test tooling
 ggml/                      Vendored ggml (see ggml/UPSTREAM for pinned SHA)

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/bindings/python/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "transcribe-cpp"
-version = "0.1.0"
+version = "0.1.1"
 description = "Python bindings for transcribe.cpp"
 readme = "README.md"
 # 3.8 is EOL (2024-10); 3.9 is the floor. The binding is ctypes-only, so there
@@ -34,13 +34,13 @@ classifiers = [
 # packaging fix still resolves); the import-time version/header-hash check in
 # _library.py is the runtime backstop. check_version_sync.py gates this pin
 # against include/transcribe.h.
-dependencies = ["transcribe-cpp-native==0.1.0.*"]
+dependencies = ["transcribe-cpp-native==0.1.1.*"]
 
 [project.optional-dependencies]
 # Opt-in accelerator providers — ADDITIVE: they install alongside the default
 # provider and the best one wins at runtime. Same base-version pin contract
 # as the hard dependency (gated by check_version_sync.py).
-cu12 = ["transcribe-cpp-native-cu12==0.1.0.*"]
+cu12 = ["transcribe-cpp-native-cu12==0.1.1.*"]
 # Test-only deps. Run with: uv run --extra test pytest (from bindings/python).
 # numpy is here so the numpy PCM-input tests run in every lane instead of
 # silently skipping wherever numpy happens to be absent.

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/bindings/python/src/transcribe_cpp/__init__.py

@@ -46,7 +46,7 @@ from .errors import (
     raise_for_status,
 )
 
-__version__ = "0.1.0"
+__version__ = "0.1.1"
 
 # String-enum types, exported so callers (and type checkers) can name them.
 Backend = Literal["auto", "cpu", "metal", "vulkan", "cpu_accel", "cuda"]

transcribe_cpp_native-0.1.1/bindings/python/tests/test_utf8_paths.py

@@ -0,0 +1,87 @@
+"""Non-ASCII (UTF-8) path handling through the C ABI.
+
+Regression coverage for Handy issue #1585: on Windows, model loading
+and backend artifact-dir validation failed for non-ASCII paths because
+the library's narrow ``::stat()`` / ``ifstream`` calls read the UTF-8
+path bytes in the process ANSI code page. src/transcribe-path.h now
+converts UTF-8 to wide paths on Windows.
+
+The C++ counterpart (tests/utf8_path_unit.cpp) pins the same contract,
+but ctest only runs on Linux/macOS in CI — this file is what executes
+on a real Windows runner (the wheel lanes run the pytest suite via
+scripts/ci/wheel_smoke.py).
+
+Model-free tests always run; the full-load test skips without a model
+(``TRANSCRIBE_SMOKE_MODEL``, set by the wheel lanes).
+"""
+
+from __future__ import annotations
+
+import os
+import shutil
+from pathlib import Path
+
+import pytest
+
+import transcribe_cpp as t
+from transcribe_cpp import errors
+
+# Same characters as the C++ counterpart's directory name.
+NON_ASCII_DIR = "transcribe-utf8-café-日本語"
+
+
+def _nonascii_dir(tmp_path: Path) -> Path:
+    d = tmp_path / NON_ASCII_DIR
+    try:
+        d.mkdir()
+    except OSError as e:  # filesystem that cannot represent the name
+        pytest.skip(f"cannot create non-ASCII directory: {e}")
+    return d
+
+
+def test_missing_file_in_nonascii_dir_raises_file_not_found(tmp_path):
+    # The not-found classification must survive the wide-path port: a
+    # genuinely absent file is FILE_NOT_FOUND, not some generic error.
+    d = _nonascii_dir(tmp_path)
+    with pytest.raises(t.ModelFileNotFound) as ei:
+        t.Model(d / "definitely-missing.gguf")
+    assert ei.value.status == errors.ERR_FILE_NOT_FOUND
+
+
+def test_junk_file_in_nonascii_dir_is_found_then_rejected(tmp_path):
+    # The #1585 failure shape: the file EXISTS at the non-ASCII path,
+    # but pre-fix Windows raised ModelFileNotFound (the ANSI-code-page
+    # stat could not see it). ModelLoadError is a sibling class, so
+    # this cannot pass via FILE_NOT_FOUND.
+    d = _nonascii_dir(tmp_path)
+    junk = d / "junk.gguf"
+    junk.write_bytes(b"this is not a gguf file" * 64)
+    with pytest.raises(t.ModelLoadError):
+        t.Model(junk)
+
+
+def test_smoke_model_loads_from_nonascii_dir(model_path, tmp_path):
+    # Full successful load through the non-ASCII path: existence
+    # pre-check, magic sniff, gguf parse, and the tensor-data streaming
+    # reopen all consume the path. Constructing the Model is the
+    # assertion.
+    d = _nonascii_dir(tmp_path)
+    local = d / model_path.name
+    shutil.copyfile(model_path, local)
+    with t.Model(local):
+        pass
+
+
+def test_init_backends_nonascii_dirs(tmp_path):
+    # The existing non-ASCII dir must not be misread as absent;
+    # FILE_NOT_FOUND specifically is the #1585 encoding bug. We do not
+    # assert OK: a module-less dir may legitimately report ERR_BACKEND
+    # depending on build posture.
+    lib = t._lib
+    d = _nonascii_dir(tmp_path)
+    st = lib.transcribe_init_backends(os.fspath(d).encode("utf-8"))
+    assert st != errors.ERR_FILE_NOT_FOUND
+
+    missing = d / "missing-subdir"
+    st = lib.transcribe_init_backends(os.fspath(missing).encode("utf-8"))
+    assert st == errors.ERR_FILE_NOT_FOUND

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/bindings/rust/transcribe-cpp/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "transcribe-cpp"
-version = "0.1.0"
+version = "0.1.1"
 edition = "2021"
 rust-version = "1.74"
 description = "Rust bindings for transcribe.cpp, a C/C++ speech-to-text library built on ggml"
@@ -40,7 +40,7 @@ dynamic-backends = ["shared", "transcribe-cpp-sys/dynamic-backends"]
 # `transcribe-cpp --no-default-features`, silently re-enabling Metal. The backend
 # features below forward explicitly, so the default (`metal`) still reaches sys
 # via `metal = ["transcribe-cpp-sys/metal"]`.
-transcribe-cpp-sys = { version = "0.1.0", path = "../../..", default-features = false }
+transcribe-cpp-sys = { version = "0.1.1", path = "../../..", default-features = false }
 thiserror = "2"
 log = "0.4"
 

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/bindings/swift/Sources/TranscribeCpp/TranscribeCpp.swift

@@ -17,7 +17,7 @@ public enum Transcribe {
     /// Version this binding was built against. Pinned here for the pre-1.0
     /// base-version load gate; the version-sync milestone will generate it from
     /// `include/transcribe.h` (the single source of truth) rather than hardcode.
-    public static let compiledVersion = "0.1.0"
+    public static let compiledVersion = "0.1.1"
 
     /// `MAJOR.MINOR.PATCH` of the linked native library.
     public static func version() -> String { String(cString: transcribe_version()) }

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/bindings/typescript/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "transcribe-cpp",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "transcribe-cpp",
-      "version": "0.1.0",
+      "version": "0.1.1",
       "license": "MIT",
       "dependencies": {
         "koffi": "^3.0.2"
@@ -19,11 +19,11 @@
         "node": ">=22"
       },
       "optionalDependencies": {
-        "@transcribe-cpp/darwin-arm64-metal": "0.1.0",
-        "@transcribe-cpp/darwin-x64-cpu": "0.1.0",
-        "@transcribe-cpp/linux-arm64-cpu-vulkan": "0.1.0",
-        "@transcribe-cpp/linux-x64-cpu-vulkan": "0.1.0",
-        "@transcribe-cpp/win32-x64-cpu-vulkan": "0.1.0"
+        "@transcribe-cpp/darwin-arm64-metal": "0.1.1",
+        "@transcribe-cpp/darwin-x64-cpu": "0.1.1",
+        "@transcribe-cpp/linux-arm64-cpu-vulkan": "0.1.1",
+        "@transcribe-cpp/linux-x64-cpu-vulkan": "0.1.1",
+        "@transcribe-cpp/win32-x64-cpu-vulkan": "0.1.1"
       }
     },
     "node_modules/@koromix/koffi-darwin-arm64": {

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/bindings/typescript/package.json

@@ -1,6 +1,6 @@
 {
   "name": "transcribe-cpp",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "description": "TypeScript/Node.js bindings for transcribe.cpp — a C/C++ speech-to-text library built on ggml",
   "type": "module",
   "exports": {
@@ -38,11 +38,11 @@
     "koffi": "^3.0.2"
   },
   "optionalDependencies": {
-    "@transcribe-cpp/darwin-arm64-metal": "0.1.0",
-    "@transcribe-cpp/darwin-x64-cpu": "0.1.0",
-    "@transcribe-cpp/linux-x64-cpu-vulkan": "0.1.0",
-    "@transcribe-cpp/linux-arm64-cpu-vulkan": "0.1.0",
-    "@transcribe-cpp/win32-x64-cpu-vulkan": "0.1.0"
+    "@transcribe-cpp/darwin-arm64-metal": "0.1.1",
+    "@transcribe-cpp/darwin-x64-cpu": "0.1.1",
+    "@transcribe-cpp/linux-x64-cpu-vulkan": "0.1.1",
+    "@transcribe-cpp/linux-arm64-cpu-vulkan": "0.1.1",
+    "@transcribe-cpp/win32-x64-cpu-vulkan": "0.1.1"
   },
   "devDependencies": {
     "@types/node": "^22.0.0",

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/docs/environment-variables.md

@@ -15,6 +15,13 @@ copies and writes raw activations to disk, so it carries I/O cost and a
 data-leak surface. Leave it unset in production unless you are deliberately
 collecting dumps.
 
+The two `TRANSCRIBE_TEST_*_THROW` fault-injection hooks also ship in release
+binaries on purpose: wheel clean-install CI proves exception containment
+against the exact artifact users install (see `AGENTS.md`, "C ABI Exception
+Discipline"). Unlike the boolean convention above, **any** non-empty value
+arms them (including `0`); unset or empty is inert. Leave them unset outside
+of tests.
+
 | Variable | Effect |
 | --- | --- |
 | `TRANSCRIBE_NO_FLASH` | Disable flash attention on encoder and decoder (forces the manual F32 path). |
@@ -24,6 +31,8 @@ collecting dumps.
 | `TRANSCRIBE_DUMP_DIR=<dir>` | Enable the per-stage tensor dumper; writes `<name>.f32` + `<name>.json` per dumped tensor into `<dir>`. The basis for the numerical-comparison harness (`scripts/compare_tensors.py`). |
 | `TRANSCRIBE_PERF_DEBUG` | Print a per-stage timing breakdown to stderr (DEBUG log) on the families that profile (`cohere`, `granite`, `canary`, `canary_qwen`, `moonshine`, `moonshine_streaming`, `qwen3_asr`, `whisper`). For whisper, a value containing `cpu` or `all` additionally prints the CPU sub-section breakdown. |
 | `TRANSCRIBE_VOXTRAL_REALTIME_STREAM_TIMING` | Print a per-component streaming wall-time breakdown at stream finalize (voxtral_realtime). |
+| `TRANSCRIBE_TEST_DEV_INIT_THROW=<match>` | Fault injection: backend device init (`ggml_backend_dev_init`) throws for devices whose name contains `<match>` (`*` matches every device). Exercises throw → skip → CPU-fallback in backend probing; an explicit backend request fails with `TRANSCRIBE_ERR_BACKEND`. Used by `backend_init_throw_unit` and `scripts/ci/vulkan_degradation_check.py`. |
+| `TRANSCRIBE_TEST_TEARDOWN_THROW` | Fault injection: any non-empty value injects a throw after each real free inside the `transcribe::safe_*` teardown wrappers, proving containment without leaking the handle. Used by `teardown_safety_unit`. |
 
 ## Tier 2 — validation hooks
 

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/include/transcribe.h

@@ -52,6 +52,12 @@
  *   releases. Consumers should rebuild against matching headers and
  *   should not assume any layout, enum value, or symbol set is frozen.
  *
+ * Exception safety (C ABI):
+ * - No C++ exception escapes a public entry point. Allocation failure maps
+ *   to TRANSCRIBE_ERR_OOM, backend/driver failure to TRANSCRIBE_ERR_BACKEND,
+ *   and free/teardown functions never fail. The log callback receives
+ *   exception detail when available.
+ *
  * Params (size-aware structs):
  * - Every caller-owned public struct crossing the ABI carries `struct_size`
  *   as field 0, both for inputs (params, family extensions) and outputs
@@ -147,7 +153,7 @@
  */
 #define TRANSCRIBE_VERSION_MAJOR 0
 #define TRANSCRIBE_VERSION_MINOR 1
-#define TRANSCRIBE_VERSION_PATCH 0
+#define TRANSCRIBE_VERSION_PATCH 1
 
 #define TRANSCRIBE_VERSION_STRINGIZE_(x) #x
 #define TRANSCRIBE_VERSION_STRINGIZE(x)  TRANSCRIBE_VERSION_STRINGIZE_(x)
@@ -407,6 +413,10 @@ typedef void (*transcribe_log_callback)(transcribe_log_level level, const char *
  *   cb == NULL            logging explicitly disabled: library and ggml
  *                         messages are dropped, nothing goes to stderr.
  *
+ * The callback must not throw. If it does anyway, the exception is contained
+ * at the emission site, the message is dropped, and a note is written to
+ * stderr.
+ *
  * Call once at process startup (see the threading contract above).
  */
 TRANSCRIBE_API void transcribe_log_set(transcribe_log_callback cb, void * userdata);
@@ -709,6 +719,9 @@ typedef enum {
  * is automatic under the supported usage (call once, before the first
  * model load).
  *
+ * artifact_dir is UTF-8, like every path crossing this API; on Windows
+ * it is converted to a wide path internally.
+ *
  * Returns:
  *   TRANSCRIBE_ERR_INVALID_ARG     artifact_dir is NULL or empty.
  *   TRANSCRIBE_ERR_FILE_NOT_FOUND  artifact_dir is not an existing directory.
@@ -1325,6 +1338,10 @@ TRANSCRIBE_API const char * transcribe_model_meta_val_str(const struct transcrib
 /*
  * Load a GGUF model from disk.
  *
+ * path is UTF-8, like every path crossing this API; on Windows it is
+ * converted to a wide path internally, so non-ASCII paths load
+ * correctly regardless of the process ANSI code page.
+ *
  * params may be NULL for all library defaults. To customize, initialize
  * a struct with transcribe_model_load_params_init() and set fields. A
  * struct with struct_size == 0 (including `{0}`) is rejected with
@@ -1433,6 +1450,10 @@ TRANSCRIBE_API const struct transcribe_model * transcribe_get_model(const struct
  * On success, results are populated on the session and may be read via
  * the accessors below. Calling transcribe_run() again replaces the
  * previous result on the same session.
+ *
+ * TRANSCRIBE_ERR_BACKEND from this function (and batch/streaming run calls)
+ * is recoverable in-process: free the session and model, reload with
+ * transcribe_model_load_params::backend = TRANSCRIBE_BACKEND_CPU, and retry.
  */
 TRANSCRIBE_API transcribe_status transcribe_run(struct transcribe_session *          session,
                                                 const float *                        pcm,

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/scripts/ci/vulkan_degradation_check.py

@@ -15,7 +15,10 @@ Tiers (one invocation per machine state):
 
   loader      A loader (+ driver, e.g. lavapipe or a real GPU) is present.
               The SAME artifacts must discover a Vulkan device. With
-              --model: a real transcription runs with backend="vulkan".
+              --model: a real transcription runs with backend="vulkan";
+              then, with TRANSCRIBE_TEST_DEV_INIT_THROW=* injecting a
+              device-init throw, explicit vulkan fails cleanly and default
+              backend falls back to CPU.
 
 Used by: provider-dl-vulkan (python-bindings.yml, hand-assembled provider
 dir), clean-install (python-wheels.yml, the shipped wheel — ships in the
@@ -29,6 +32,7 @@ from __future__ import annotations
 
 import argparse
 import array
+import os
 import sys
 import wave
 
@@ -97,6 +101,28 @@ def main() -> int:
             text = transcribe(t, args.model, args.audio, backend="vulkan")
             print("text (vulkan):", text.strip())
             assert "country" in text.lower(), text
+
+            # Same shipped artifact, injected device-init failure.
+            os.environ["TRANSCRIBE_TEST_DEV_INIT_THROW"] = "*"
+            try:
+                try:
+                    t.Model(args.model, backend="vulkan")
+                except t.TranscribeError as exc:
+                    print("throwing vulkan device correctly rejected:", exc)
+                else:
+                    raise AssertionError(
+                        "backend='vulkan' must fail when device init throws")
+                with t.Model(args.model) as m:
+                    # m.backend is the ggml backend name ("CPU", "Vulkan0"),
+                    # not a lowercase kind string.
+                    assert m.backend.lower() == "cpu", m.backend
+                    with m.session() as s:
+                        text = s.run(load_pcm(args.audio)).text
+                print("text (throwing devices, default backend):",
+                      text.strip())
+                assert "country" in text.lower(), text
+            finally:
+                del os.environ["TRANSCRIBE_TEST_DEV_INIT_THROW"]
         print("ok: loader present — the same artifacts discovered Vulkan")
     return 0
 

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/src/arch/canary/canary.h

@@ -60,7 +60,7 @@ struct CanaryKvCache {
 
     void free() {
         if (buffer != nullptr) {
-            ggml_backend_buffer_free(buffer);
+            safe_buffer_free(buffer);
             buffer = nullptr;
         }
         if (ctx != nullptr) {

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/src/arch/canary/model.cpp

@@ -46,7 +46,7 @@ static_assert(std::is_base_of_v<transcribe_session, CanarySession>);
 CanarySession::~CanarySession() {
     kv_cache.free();
     if (sched != nullptr) {
-        ggml_backend_sched_free(sched);
+        safe_sched_free(sched);
         sched = nullptr;
     }
     if (compute_ctx != nullptr) {
@@ -183,7 +183,7 @@ CanaryModel::~CanaryModel() {
         bn_fused_ctx = nullptr;
     }
     if (bn_fused_buffer != nullptr) {
-        ggml_backend_buffer_free(bn_fused_buffer);
+        safe_buffer_free(bn_fused_buffer);
         bn_fused_buffer = nullptr;
     }
     if (conv_pw_f32_ctx != nullptr) {
@@ -191,7 +191,7 @@ CanaryModel::~CanaryModel() {
         conv_pw_f32_ctx = nullptr;
     }
     if (conv_pw_f32_buffer != nullptr) {
-        ggml_backend_buffer_free(conv_pw_f32_buffer);
+        safe_buffer_free(conv_pw_f32_buffer);
         conv_pw_f32_buffer = nullptr;
     }
     if (ctx_meta != nullptr) {
@@ -199,11 +199,11 @@ CanaryModel::~CanaryModel() {
         ctx_meta = nullptr;
     }
     if (backend_buffer != nullptr) {
-        ggml_backend_buffer_free(backend_buffer);
+        safe_buffer_free(backend_buffer);
         backend_buffer = nullptr;
     }
     for (auto it = plan.scheduler_list.rbegin(); it != plan.scheduler_list.rend(); ++it) {
-        ggml_backend_free(*it);
+        safe_backend_free(*it);
     }
     plan.scheduler_list.clear();
     plan.primary      = nullptr;

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/src/arch/canary_qwen/model.cpp

@@ -61,7 +61,7 @@ CanaryQwenSession::~CanaryQwenSession() {
     kv_cache.free();
     kv_cache_batch.free();
     if (sched != nullptr) {
-        ggml_backend_sched_free(sched);
+        safe_sched_free(sched);
         sched = nullptr;
     }
     if (compute_ctx != nullptr) {
@@ -76,11 +76,11 @@ CanaryQwenModel::~CanaryQwenModel() {
         ctx_meta = nullptr;
     }
     if (backend_buffer != nullptr) {
-        ggml_backend_buffer_free(backend_buffer);
+        safe_buffer_free(backend_buffer);
         backend_buffer = nullptr;
     }
     if (bn_fused_buffer != nullptr) {
-        ggml_backend_buffer_free(bn_fused_buffer);
+        safe_buffer_free(bn_fused_buffer);
         bn_fused_buffer = nullptr;
     }
     if (bn_fused_ctx != nullptr) {
@@ -88,7 +88,7 @@ CanaryQwenModel::~CanaryQwenModel() {
         bn_fused_ctx = nullptr;
     }
     if (conv_pw_f32_buffer != nullptr) {
-        ggml_backend_buffer_free(conv_pw_f32_buffer);
+        safe_buffer_free(conv_pw_f32_buffer);
         conv_pw_f32_buffer = nullptr;
     }
     if (conv_pw_f32_ctx != nullptr) {
@@ -96,7 +96,7 @@ CanaryQwenModel::~CanaryQwenModel() {
         conv_pw_f32_ctx = nullptr;
     }
     if (linear_f32_buffer != nullptr) {
-        ggml_backend_buffer_free(linear_f32_buffer);
+        safe_buffer_free(linear_f32_buffer);
         linear_f32_buffer = nullptr;
     }
     if (linear_f32_ctx != nullptr) {
@@ -105,7 +105,7 @@ CanaryQwenModel::~CanaryQwenModel() {
     }
     packed_gate_up.free();
     for (auto it = plan.scheduler_list.rbegin(); it != plan.scheduler_list.rend(); ++it) {
-        ggml_backend_free(*it);
+        safe_backend_free(*it);
     }
     plan.scheduler_list.clear();
     plan.primary      = nullptr;
@@ -357,7 +357,7 @@ transcribe_status promote_conv_pw_to_f32_on_cpu(CanaryQwenModel & m) {
     const auto * f16_traits = ggml_get_type_traits(GGML_TYPE_F16);
     if (f16_traits == nullptr || f16_traits->to_float == nullptr) {
         log_msg(TRANSCRIBE_LOG_LEVEL_WARN, "canary_qwen: no f16 to_float trait — skipping conv promotion");
-        ggml_backend_buffer_free(m.conv_pw_f32_buffer);
+        safe_buffer_free(m.conv_pw_f32_buffer);
         m.conv_pw_f32_buffer = nullptr;
         ggml_free(m.conv_pw_f32_ctx);
         m.conv_pw_f32_ctx = nullptr;
@@ -483,7 +483,7 @@ transcribe_status promote_linears_bf16_to_f32_on_cpu(CanaryQwenModel & m) {
     const auto * bf16_traits = ggml_get_type_traits(GGML_TYPE_BF16);
     if (bf16_traits == nullptr || bf16_traits->to_float == nullptr) {
         log_msg(TRANSCRIBE_LOG_LEVEL_WARN, "canary_qwen: no bf16 to_float trait — skipping linear promotion");
-        ggml_backend_buffer_free(m.linear_f32_buffer);
+        safe_buffer_free(m.linear_f32_buffer);
         m.linear_f32_buffer = nullptr;
         ggml_free(m.linear_f32_ctx);
         m.linear_f32_ctx = nullptr;

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/src/arch/cohere/cohere.h

@@ -78,7 +78,7 @@ struct CohereKvCache {
 
     void free() {
         if (buffer != nullptr) {
-            ggml_backend_buffer_free(buffer);
+            safe_buffer_free(buffer);
             buffer = nullptr;
         }
         if (ctx != nullptr) {

{transcribe_cpp_native-0.1.0 → transcribe_cpp_native-0.1.1}/src/arch/cohere/model.cpp

@@ -49,7 +49,7 @@ static_assert(std::is_base_of_v<transcribe_session, CohereSession>);
 CohereSession::~CohereSession() {
     kv_cache.free();
     if (sched != nullptr) {
-        ggml_backend_sched_free(sched);
+        safe_sched_free(sched);
         sched = nullptr;
     }
     if (compute_ctx != nullptr) {
@@ -193,7 +193,7 @@ CohereModel::~CohereModel() {
         bn_fused_ctx = nullptr;
     }
     if (bn_fused_buffer != nullptr) {
-        ggml_backend_buffer_free(bn_fused_buffer);
+        safe_buffer_free(bn_fused_buffer);
         bn_fused_buffer = nullptr;
     }
     if (conv_pw_f32_ctx != nullptr) {
@@ -201,7 +201,7 @@ CohereModel::~CohereModel() {
         conv_pw_f32_ctx = nullptr;
     }
     if (conv_pw_f32_buffer != nullptr) {
-        ggml_backend_buffer_free(conv_pw_f32_buffer);
+        safe_buffer_free(conv_pw_f32_buffer);
         conv_pw_f32_buffer = nullptr;
     }
     if (ctx_meta != nullptr) {
@@ -209,11 +209,11 @@ CohereModel::~CohereModel() {
         ctx_meta = nullptr;
     }
     if (backend_buffer != nullptr) {
-        ggml_backend_buffer_free(backend_buffer);
+        safe_buffer_free(backend_buffer);
         backend_buffer = nullptr;
     }
     for (auto it = plan.scheduler_list.rbegin(); it != plan.scheduler_list.rend(); ++it) {
-        ggml_backend_free(*it);
+        safe_backend_free(*it);
     }
     plan.scheduler_list.clear();
     plan.primary      = nullptr;