docs: bilingual section headers + intro (zh/en)

Author: KakaruHayate

BUILDING.md

@@ -1,12 +1,14 @@
-# Building from source
+# Building from source · 从源码构建
 
 This document covers how to build `game_ggml_cli` from source on Linux, macOS and
 Windows.  The project is self-contained — all dependencies are pulled via CMake
 FetchContent at configure time.
 
-## Prerequisites
+本文档覆盖在 Linux、macOS、Windows 上从源码构建 `game_ggml_cli`。项目自包含——所有依赖在 configure 时经 CMake FetchContent 拉取。
 
-### Linux
+## Prerequisites · 前置要求
+
+### Linux · Linux
 
 ```bash
 # Build tools
@@ -28,7 +30,7 @@ sudo apt install libvulkan-dev vulkan-tools
 sudo apt install ccache
 ```
 
-### macOS
+### macOS · macOS
 
 ```bash
 # Xcode Command Line Tools
@@ -42,7 +44,7 @@ brew install cmake ccache
 #   cmake -DCMAKE_OSX_ARCHITECTURES=x86_64 ...
 ```
 
-### Windows
+### Windows · Windows
 
 ```bash
 # Visual Studio 2022+ with "Desktop development with C++" workload
@@ -60,7 +62,7 @@ brew install cmake ccache
 # CUDA 12.6 supports Visual Studio 2022 / MSVC 193x.
 ```
 
-## Quick start
+## Quick start · 快速开始
 
 ```bash
 # Clone
@@ -79,7 +81,7 @@ cmake --build build -j
 build/bin/game_ggml_cli --version
 ```
 
-## Backend selection
+## Backend selection · 后端选择
 
 Pass one (or more) of these flags to the **Configure** step:
 
@@ -92,7 +94,7 @@ Pass one (or more) of these flags to the **Configure** step:
 If no GPU backend is enabled, the CPU backend is used as fallback.  The best
 available backend is selected automatically at runtime.
 
-### Examples
+### Examples · 示例
 
 ```bash
 # Linux + Vulkan
@@ -131,7 +133,7 @@ cmake -B build -DCMAKE_BUILD_TYPE=Release `
                -DCMAKE_CUDA_ARCHITECTURES="75"
 ```
 
-## Converting a PyTorch checkpoint to GGUF
+## Converting a PyTorch checkpoint to GGUF · 转换 PyTorch 权重为 GGUF
 
 The medium model checkpoint can be downloaded from
 [OpenVPI releases](https://github.com/openvpi/GAME/releases/download/v1.0.0/GAME-1.0-medium.zip).
@@ -159,7 +161,7 @@ Expected output for the medium model:
   tensors      : 671
 ```
 
-## Running inference
+## Running inference · 运行推理
 
 ```bash
 # Single file
@@ -209,15 +211,15 @@ recommended. The packages currently expect the CUDA 12 runtime and cuBLAS
 libraries to be installed on the target system; they do not bundle NVIDIA's
 runtime libraries or the driver (`nvcuda.dll` comes from the NVIDIA driver).
 
-## Troubleshooting
+## Troubleshooting · 故障排查
 
 ### CUDA Toolkit not found
 
 Ensure `nvcc --version` succeeds and that `CUDA_PATH` (or the platform-specific
 Toolkit environment) points to the intended installation. Delete `build/` before
 reconfiguring after changing CUDA Toolkit versions.
 
-### glslc not found (Linux/Windows Vulkan)
+### glslc not found (Linux/Windows Vulkan) · glslc 未找到（Linux/Windows Vulkan）
 
 The Vulkan SDK's `glslc` compiler is required by the ggml-vulkan backend.  If
 `FindVulkan` reports `glslc` as missing, ensure the SDK is installed and its
@@ -233,7 +235,7 @@ set VULKAN_SDK=C:\VulkanSDK\1.4.304.1
 set PATH=%VULKAN_SDK%\Bin;%PATH%
 ```
 
-### `unknown target CPU 'apple-m1'` (macOS x86_64 cross-compile)
+### `unknown target CPU 'apple-m1'` (macOS x86_64 cross-compile) · `unknown target CPU 'apple-m1'`（macOS x86_64 交叉编译）
 
 When cross-compiling for Intel Mac on an Apple Silicon runner, the CPU backend
 mistakenly detects the host as `apple-m1`.  Disable native CPU detection:
@@ -242,7 +244,7 @@ mistakenly detects the host as `apple-m1`.  Disable native CPU detection:
 cmake -B build -DGGML_NATIVE=OFF ...
 ```
 
-### `FetchContent` download failures
+### `FetchContent` download failures · `FetchContent` 下载失败
 
 Dependencies are fetched via Git at configure time.  If you are behind a proxy:
 

README.md

@@ -1,11 +1,13 @@
-# GAME ggml backend
+# GAME ggml backend · GAME ggml 后端
 
 Native C++ inference for the [GAME](https://github.com/openvpi/GAME) singing-voice-to-MIDI model, built on
 [ggml](https://github.com/ggerganov/ggml).  Runs on CPU, Metal (default on
 Apple Silicon), CUDA, or Vulkan.  Drop-in replacement for
 `python infer.py extract` with no Python dependency at runtime.
 
-## Highlights
+基于 [ggml](https://github.com/ggerganov/ggml) 的 [GAME](https://github.com/openvpi/GAME) 歌声转 MIDI 模型原生 C++ 推理实现，支持 CPU、Metal（Apple Silicon 默认）、CUDA、Vulkan；是 `python infer.py extract` 的运行时零 Python 依赖替代品。
+
+## Highlights · 特性亮点
 
 - **End-to-end CLI** — WAV in, MIDI/TXT/CSV out (mirrors Python `extract`).
 - **Small footprint** — ~50 MB GGUF for the 1.0-medium checkpoint, ~50 M params.
@@ -16,7 +18,7 @@ Apple Silicon), CUDA, or Vulkan.  Drop-in replacement for
 - **Parity-tested** — full pipeline output matches the PyTorch reference bit-for-bit
   when the same RNG numbers are injected.
 
-## Architecture
+## Architecture · 架构
 
 ```
 waveform (44100 Hz mono)
@@ -40,7 +42,7 @@ waveform (44100 Hz mono)
    Gaussian-blurred pitch decode → notes (offset, duration, pitch, voiced)
 ```
 
-## Build
+## Build · 构建
 
 ```bash
 cmake -S ggml_backend -B ggml_backend/build \
@@ -59,7 +61,7 @@ Options:
 | `GAME_GGML_BUILD_CLI`   | `ON`              | Build `game_ggml_cli` |
 | `GAME_GGML_BUILD_TESTS` | `OFF`             | Build GoogleTest suite |
 
-## Convert a PyTorch checkpoint
+## Convert a PyTorch checkpoint · 转换 PyTorch 权重
 
 ```bash
 pip install -r ggml_backend/scripts/requirements.txt
@@ -78,7 +80,7 @@ Inspect the result:
 ./ggml_backend/build/bin/game_ggml_cli inspect ggml_backend/assets/game_small.gguf
 ```
 
-## Run inference
+## Run inference · 运行推理
 
 ```bash
 ./ggml_backend/build/bin/game_ggml_cli extract input.wav \
@@ -107,7 +109,7 @@ Inspect the result:
 | `--round-pitch`       | `--round-pitch` |
 | `--rng-replay <path>` | *(new)* — replay D3PM random numbers from a file for bit-exact parity with PyTorch |
 
-## Performance
+## Performance · 性能
 
 Measured on Apple M4 (macOS, 16-core Apple Silicon), 3 runs per side under
 `/usr/bin/time -l`, fresh subprocess each run (so peak RSS is clean).  Input:
@@ -161,7 +163,7 @@ estimator   ~0.06 s  (~ 6%)   x_est + regions → notes  (4× JEBF + joint attn)
 Segmenter dominates because D3PM loops it `--nsteps` times.  Default `1`
 keeps it cheap; bump to `4` or `8` for higher quality at linear cost.
 
-## Reproducing the benchmark
+## Reproducing the benchmark · 复现基准测试
 
 ```bash
 # 1. Resample to 44.1 kHz / mono (if not already)
@@ -186,7 +188,7 @@ python3 ggml_backend/scripts/benchmark_align.py /tmp/28_44100.wav \
     -l zh -o /tmp/bench_out --runs 3
 ```
 
-## Using as a third-party library
+## Using as a third-party library · 作为第三方库使用
 
 ```cmake
 add_subdirectory(path/to/GAME/ggml_backend)
@@ -223,7 +225,7 @@ include any ggml header.
 See [`examples/external_consumer/`](examples/external_consumer/) for a minimal
 standalone CMake project that builds against the library.
 
-## Tests
+## Tests · 测试
 
 ```bash
 ctest --test-dir ggml_backend/build --output-on-failure
@@ -247,7 +249,7 @@ The suite has 37 tests covering:
 Reference dumps are generated by `python scripts/dump_reference.py --category all`.
 Dumps are gitignored — regenerate them as part of CI.
 
-## Known limitations (v1)
+## Known limitations (v1) · 已知限制（v1）
 
 - **44100 Hz mono WAV only** — other sample rates raise `InvalidWav`.  Resampling
   is deliberately out-of-scope to keep the footprint small.
@@ -261,7 +263,7 @@ Dumps are gitignored — regenerate them as part of CI.
 - **Metal FP32 precision** — expected ~1e-3 per matmul; at boundary decoding
   this can flip one frame out of every few hundred vs the CPU reference.
 
-## Dependencies
+## Dependencies · 依赖
 
 Everything is fetched at configure time by CMake; nothing is vendored.  Source
 trees live under `build/_deps/<name>-src/` after the first configure.
@@ -277,7 +279,7 @@ Each upstream LICENSE file is preserved under `build/_deps/<name>-src/LICENSE*`
 after download.  To update a dependency, change its `GIT_TAG` in
 `cmake/Dependencies.cmake` and reconfigure.
 
-## License
+## License · 许可
 
 MIT — same as the parent [GAME project](https://github.com/openvpi/GAME).  Redistributions should
 also carry the upstream license notices listed in the table above.