fix(backends): enable sd-cpp CUDA support on Windows (#2601)

The sd-cpp descriptor restricted its cuda variant to Linux only, even
though the Windows CUDA download/runtime path in sdcpp_server.cpp was
already implemented and the upstream release still publishes a
windows-cuda asset (validated by validate_sdcpp.yml). Add "windows" back
to the cuda support row and add cuda to arg/bin variants so config.json
gets cuda_args/cuda_bin defaults, matching llama.cpp's pattern.

Regenerated defaults.json and docs via gen_backend_boilerplate.py.

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>

Author: kenvandine

README.md

@@ -256,7 +256,7 @@ Lemonade supports multiple inference engines for LLM, speech, TTS, and image gen
     <tr>
       <td><code>cuda</code></td>
       <td>NVIDIA GPUs (Turing or newer)**</td>
-      <td>Linux</td>
+      <td>Windows, Linux</td>
     </tr>
     <tr>
       <td><code>vulkan</code></td>

docs/dev/backends-reference.md

@@ -48,7 +48,7 @@ the generator instead. Prose outside the markers is preserved. -->
 | `openmoss` | vulkan | linux, windows | amd_gpu; cpu (x86_64); nvidia_gpu |
 | `ryzenai-llm` | npu | windows | amd_npu (XDNA2) |
 | `sd-cpp` | rocm | linux, windows | amd_gpu (gfx103X, gfx110X, gfx1150, gfx1151, gfx1152, gfx120X) |
-| `sd-cpp` | cuda | linux | nvidia_gpu (sm_100, sm_120, sm_121, sm_75, sm_80, sm_86, sm_89, sm_90) |
+| `sd-cpp` | cuda | linux, windows | nvidia_gpu (sm_100, sm_120, sm_121, sm_75, sm_80, sm_86, sm_89, sm_90) |
 | `sd-cpp` | vulkan | linux, windows | amd_gpu; cpu (x86_64); nvidia_gpu |
 | `sd-cpp` | cpu | linux, windows | cpu (x86_64) |
 | `sd-cpp` | metal | macos | metal |

docs/guide/configuration/README.md

@@ -97,6 +97,8 @@ Values set in the user's `config.json` always take precedence over these seeded
     "cfg_scale": 7.0,
     "cpu_args": "",
     "cpu_bin": "builtin",
+    "cuda_args": "",
+    "cuda_bin": "builtin",
     "height": 512,
     "rocm_args": "",
     "rocm_bin": "builtin",

src/cpp/include/lemon/backends/sdcpp/sdcpp.h

@@ -38,7 +38,7 @@ inline const BackendDescriptor descriptor = {
     /*support*/ {
         {"rocm", {"windows", "linux"},
          {{"amd_gpu", {"gfx1150", "gfx1151", "gfx1152", "gfx103X", "gfx110X", "gfx120X"}}}, "Supported AMD ROCm iGPU/dGPU families*"},
-        {"cuda", {"linux"},
+        {"cuda", {"windows", "linux"},
          {{"nvidia_gpu", {"sm_75", "sm_80", "sm_86", "sm_89", "sm_90", "sm_100", "sm_120", "sm_121"}}}, "NVIDIA GPUs (Turing or newer)**"},
         {"vulkan", {"windows", "linux"}, {{"cpu", {"x86_64"}}, {"amd_gpu", {}}, {"nvidia_gpu", {}}}, "Vulkan-capable GPUs"},
         {"cpu", {"windows", "linux"}, {{"cpu", {"x86_64"}}}, "x86_64 CPU"},
@@ -55,8 +55,8 @@ inline const BackendDescriptor descriptor = {
     /*version_policy*/  VersionPolicy::Exact,
     /*self_manages_downloads*/ false,
     /*takes_args*/      true,
-    /*arg_variants*/    {"cpu", "rocm", "vulkan"},
-    /*bin_variants*/    {"cpu", "rocm", "vulkan"},
+    /*arg_variants*/    {"cpu", "rocm", "vulkan", "cuda"},
+    /*bin_variants*/    {"cpu", "rocm", "vulkan", "cuda"},
     /*config_extra*/    {{"steps", 20}, {"cfg_scale", 7.0}, {"width", 512}, {"height", 512}},
 };
 

src/cpp/resources/defaults.json

@@ -62,6 +62,8 @@
     "cfg_scale": 7.0,
     "cpu_args": "",
     "cpu_bin": "builtin",
+    "cuda_args": "",
+    "cuda_bin": "builtin",
     "height": 512,
     "rocm_args": "",
     "rocm_bin": "builtin",