Add --export-type (generic | optimized) selector to winml build

docs/commands/build.md

@@ -22,6 +22,7 @@ $ winml build [options]
 |---|---|---|---|---|
 | `--config` | `-c` | path | `None` | `WinMLBuildConfig` JSON file, generated by `winml config`. If omitted, config is auto-generated from `-m`. |
 | `--model` | `-m` | string | `None` | Hugging Face model ID or path to an existing `.onnx` file. |
+| `--export-type` | | choice | `generic` | Output selector: `generic` builds the stock single/composite ONNX model; `optimized` builds the family's registered runtime-optimized recipe (today the onnxruntime-genai NPU bundle), inferring its target. `optimized` fails fast if the family has no recipe or `--ep`/`--device` contradicts it. |
 | `--output-dir` | `-o` | path | `None` | Directory for all build artifacts. Mutually exclusive with `--use-cache`. |
 | `--use-cache/--no-use-cache` | | flag | `false` | Store artifacts in the winml-cli global cache (`~/.cache/winml/`). Mutually exclusive with `--output-dir`. |
 | `--rebuild/--no-rebuild` | | flag | `false` | Overwrite existing artifacts and re-run the full pipeline. |
@@ -58,11 +59,10 @@ single-pass build. Individual stages can be suppressed with `--no-quant`,
 
 ## Genai bundles for decoder LLMs (NPU + QNN)
 
-For a registered decoder-LLM family (currently **Qwen3**), targeting the NPU HTP
-with an explicit `--ep qnn` switches `winml build` to produce a complete
-[onnxruntime-genai](https://github.com/microsoft/onnxruntime-genai) **bundle**
-directory instead of the stock per-model ONNX output. Pair `--ep qnn` with
-`--device npu`, or with `--device auto` when auto-detection resolves to the NPU:
+For a registered decoder-LLM family (currently **Qwen3**), `--export-type
+optimized` switches `winml build` from the stock per-model ONNX output to a
+complete [onnxruntime-genai](https://github.com/microsoft/onnxruntime-genai)
+**bundle** directory:
 
 | File | Role | Precision |
 |---|---|---|
@@ -73,20 +73,36 @@ directory instead of the stock per-model ONNX output. Pair `--ep qnn` with
 | `genai_config.json` + tokenizer | onnxruntime-genai runtime metadata | — |
 
 ```bash
-# One command: HF decoder LLM → full onnxruntime-genai bundle on the NPU
-winml build -m Qwen/Qwen3-0.6B -o out/qwen3-bundle --device npu --ep qnn
-# or let device auto-detection pick the NPU:
-winml build -m Qwen/Qwen3-0.6B -o out/qwen3-bundle --device auto --ep qnn
+# One command: HF decoder LLM → full onnxruntime-genai bundle
+winml build -m Qwen/Qwen3-0.6B -o out/qwen3-bundle --export-type optimized
 ```
 
-The trigger is data-driven and opt-in: **`--ep qnn` must be passed explicitly**,
-alongside an NPU target (`--device npu`, or `--device auto` resolving to the NPU).
-Every other target — including Qwen3 on CPU/GPU, or an auto-detected NPU target
-without an explicit `--ep qnn` — keeps the existing behavior (the stock composite
-build). `--output-dir` is required (the bundle is a directory) and `--use-cache`
-is not supported. Use `--precision` to override the transformer precision; the CPU
-companions always use their bundle-standard precisions. See
-[Qwen3 — Genai Bundle](../samples/qwen3-genai-bundle.md) for the full walkthrough.
+`--export-type optimized` **infers** the recipe's target (Qwen3 builds for the
+NPU HTP via QNN), so `--ep`/`--device` are optional — and, unlike the shortcut
+below, the optimized path does not gate on detecting the target device locally.
+If you do pin `--ep`/`--device`, they must match the recipe: a contradicting
+value, an unregistered family, a pre-exported `.onnx` input, or module mode all
+fail fast rather than silently falling back to a generic build. `--output-dir` is
+required (the bundle is a directory) and `--use-cache` is not supported. Use
+`--precision` to override the transformer precision; the CPU companions always
+use their bundle-standard precisions.
+
+!!! note "Backward-compatible shortcut"
+    When `--export-type` is omitted, a registered family with an explicit
+    `--ep qnn` on an NPU target still routes to its optimized bundle. The NPU
+    target may be explicit (`--device npu`) or resolved from `auto` — whether
+    `--device auto` is typed or `--device` is left off entirely:
+
+    ```bash
+    winml build -m Qwen/Qwen3-0.6B -o out/qwen3-bundle --device npu --ep qnn
+    # or let device auto-detection pick the NPU (--device may be omitted):
+    winml build -m Qwen/Qwen3-0.6B -o out/qwen3-bundle --ep qnn
+    ```
+
+    `--export-type generic` always forces the stock composite build, even for a
+    registered family on the NPU.
+
+See [Qwen3 — Genai Bundle](../samples/qwen3-genai-bundle.md) for the full walkthrough.
 
 ## Examples
 
@@ -150,9 +166,9 @@ winml build -m microsoft/resnet-50 -o output/ \
   pre-exported ONNX inputs because the export step has already happened.
 - **Existing artifacts are reused by default.** Pass `--rebuild` to force a
   fresh run after changing the config.
-- **Genai bundles require `--output-dir`, not `--use-cache`.** A decoder-LLM
-  bundle (`--device npu --ep qnn`) writes a directory of components and rejects
-  `--use-cache`.
+- **Genai bundles require `--output-dir`, not `--use-cache`.** An optimized
+  decoder-LLM bundle (`--export-type optimized`, or the `--device npu --ep qnn`
+  shortcut) writes a directory of components and rejects `--use-cache`.
 
 ## See also
 

docs/samples/qwen3-genai-bundle.md

@@ -29,7 +29,7 @@ memory-bound companions stay on CPU.
 
 ```mermaid
 graph LR
-    A["winml build -m Qwen/Qwen3-0.6B --device npu --ep qnn"] --> B[Genai bundle recipe]
+    A["winml build -m Qwen/Qwen3-0.6B --export-type optimized"] --> B[Genai bundle recipe]
     B --> C[ctx.onnx / iter.onnx — NPU]
     B --> D[embeddings.onnx — CPU]
     B --> E[lm_head.onnx — CPU]
@@ -41,14 +41,12 @@ graph LR
 
 ## Step 1: Build the bundle (one command)
 
-Targeting the NPU HTP with an explicit `--ep qnn` switches `winml build` from the
-stock per-model ONNX output to the full genai bundle. Pair it with `--device npu`,
-or `--device auto` when auto-detection resolves to the NPU:
+`--export-type optimized` switches `winml build` from the stock per-model ONNX
+output to the full genai bundle. It **infers** the recipe's target (Qwen3 builds
+for the NPU HTP via QNN), so `--ep`/`--device` are optional:
 
 ```bash
-winml build -m Qwen/Qwen3-0.6B -o out/qwen3-bundle --device npu --ep qnn
-# or, letting device auto-detection pick the NPU:
-winml build -m Qwen/Qwen3-0.6B -o out/qwen3-bundle --device auto --ep qnn
+winml build -m Qwen/Qwen3-0.6B -o out/qwen3-bundle --export-type optimized
 ```
 
 This builds (or reuses from cache) all four components and assembles them, writing
@@ -63,11 +61,23 @@ The CPU companions likewise keep their bundle-standard precisions.
 
 Force a clean rebuild of every component with `--rebuild`.
 
-!!! note "Opt-in, non-destructive"
-    The bundle path only triggers on an **explicit `--ep qnn`** together with an
-    NPU target — either `--device npu` or a `--device auto` that resolves to the
-    NPU. Qwen3 on any other target — CPU, GPU, or an auto-detected NPU *without*
-    an explicit `--ep qnn` — still produces the stock composite build, unchanged.
+!!! note "Backward-compatible shortcut"
+    When `--export-type` is omitted, an **explicit `--ep qnn`** targeting the NPU
+    still routes a registered family to its bundle. The NPU target may be
+    explicit (`--device npu`) or resolved from `auto` — whether `--device auto`
+    is typed or `--device` is omitted:
+
+    ```bash
+    winml build -m Qwen/Qwen3-0.6B -o out/qwen3-bundle --device npu --ep qnn
+    # or, letting device auto-detection pick the NPU (--device may be omitted):
+    winml build -m Qwen/Qwen3-0.6B -o out/qwen3-bundle --ep qnn
+    ```
+
+    Qwen3 on any other target — CPU, GPU, or an auto-detected NPU *without* an
+    explicit `--ep qnn` — still produces the stock composite build. `--export-type
+    generic` always forces the stock build, even on the NPU. A pinned
+    `--ep`/`--device` that contradicts the recipe fails fast rather than silently
+    reverting.
 
 ## Step 2: Tune context and prefill lengths (optional)