Merge branch 'main' into chao/fixqnngpu

Author: chinazhangchao

src/winml/modelkit/commands/build.py

@@ -436,22 +436,23 @@ def _validate_loader_tasks_for_model(
     help="Output directory for all build artifacts",
 )
 @click.option(
-    "--use-cache",
-    is_flag=True,
+    "--use-cache/--no-use-cache",
     default=False,
+    show_default=True,
     help="Use WinML CLI global cache (~/.cache/winml/). Mutually exclusive with -o.",
 )
 @click.option(
-    "--rebuild",
-    is_flag=True,
+    "--rebuild/--no-rebuild",
     default=False,
+    show_default=True,
     help="Overwrite existing artifacts and rebuild",
 )
 @click.option(
-    "--no-quant",
-    is_flag=True,
-    default=False,
-    help="Skip quantization (overrides config)",
+    "--quant/--no-quant",
+    "quant",
+    default=True,
+    show_default=True,
+    help="Enable quantization (use --no-quant to skip, overrides config)",
 )
 @click.option(
     "--no-compile/--compile",
@@ -472,16 +473,18 @@ def _validate_loader_tasks_for_model(
     optional_message="Default: auto-detect.",
 )
 @click.option(
-    "--no-analyze",
-    is_flag=True,
-    default=False,
-    help="Skip analyzer loop during build",
+    "--analyze/--no-analyze",
+    "analyze",
+    default=True,
+    show_default=True,
+    help="Run analyzer loop during build (use --no-analyze to skip)",
 )
 @click.option(
-    "--no-optimize",
-    is_flag=True,
-    default=False,
-    help="Skip optimization (for pre-quantized ONNX models)",
+    "--optimize/--no-optimize",
+    "optimize",
+    default=True,
+    show_default=True,
+    help="Run optimization (use --no-optimize to skip for pre-quantized ONNX models)",
 )
 @click.option(
     "--max-optim-iterations",
@@ -503,12 +506,12 @@ def build(
     output_dir: str | None,
     use_cache: bool,
     rebuild: bool,
-    no_quant: bool,
+    quant: bool,
     no_compile: bool | None,
-    no_optimize: bool,
+    optimize: bool,
     ep: EPNameOrAlias | None,
     device: str,
-    no_analyze: bool,
+    analyze: bool,
     max_optim_iterations: int | None,
     allow_unsupported_nodes: bool,
     trust_remote_code: bool,
@@ -579,7 +582,7 @@ def build(
         if config_file is not None:
             config_or_configs = _load_config(
                 config_file,
-                no_quant=no_quant,
+                no_quant=not quant,
                 no_compile=no_compile,
             )
         else:
@@ -592,7 +595,7 @@ def build(
                 trust_remote_code=trust_remote_code,
                 device=device,
             )
-            if no_quant:
+            if not quant:
                 config_or_configs.quant = None
             # Auto-generated configs: compile disabled by default unless
             # --compile was explicitly passed (no_compile=False).
@@ -610,7 +613,7 @@ def _patch_device(cfg: WinMLBuildConfig) -> None:
                 from ..config import resolve_quant_compile_config
 
                 resolved_quant, _ = resolve_quant_compile_config(device=device, ep=ep)
-                if no_quant or resolved_quant is None:
+                if not quant or resolved_quant is None:
                     cfg.quant = None
                 elif cfg.quant is None:
                     # Populate calibration identifiers from the loader/model
@@ -643,9 +646,7 @@ def _patch_device(cfg: WinMLBuildConfig) -> None:
         # scratch state when the user passes the wrong file or a
         # hand-edited config (#P1 UX).
         _configs_to_validate: list[WinMLBuildConfig] = (
-            config_or_configs
-            if isinstance(config_or_configs, list)
-            else [config_or_configs]
+            config_or_configs if isinstance(config_or_configs, list) else [config_or_configs]
         )
         try:
             for _cfg in _configs_to_validate:
@@ -661,9 +662,9 @@ def _patch_device(cfg: WinMLBuildConfig) -> None:
 
         # Build extra kwargs for pipeline control
         extra_kwargs: dict[str, Any] = {}
-        if no_optimize:
+        if not optimize:
             extra_kwargs["skip_optimize"] = True
-        if no_analyze:
+        if not analyze:
             extra_kwargs["hack_max_optim_iterations"] = 0
         elif max_optim_iterations is not None:
             extra_kwargs["hack_max_optim_iterations"] = max_optim_iterations

src/winml/modelkit/commands/compile.py

@@ -85,9 +85,9 @@
     help="Path to QAIRT SDK root",
 )
 @click.option(
-    "--embed",
-    is_flag=True,
+    "--embed/--no-embed",
     default=False,
+    show_default=True,
     help="Embed EP context in ONNX file (default: external .bin file)",
 )
 @click.option(

src/winml/modelkit/commands/config.py

@@ -126,10 +126,11 @@ def _apply_stage_overrides(cfg: Any, *, no_quant: bool, no_compile: bool) -> Non
     help="Source library for TasksManager (default: transformers)",
 )
 @click.option(
-    "--no-quant",
-    is_flag=True,
-    default=False,
-    help="Exclude quantization from generated config (sets quant=None)",
+    "--quant/--no-quant",
+    "quant",
+    default=True,
+    show_default=True,
+    help="Include quantization in generated config (use --no-quant to exclude, sets quant=None)",
 )
 @click.option(
     "--no-compile/--compile",
@@ -156,7 +157,7 @@ def config(
     library_name: str,
     verbose: int,
     quiet: bool,
-    no_quant: bool,
+    quant: bool,
     no_compile: bool,
     trust_remote_code: bool,
 ) -> None:
@@ -289,7 +290,7 @@ def config(
             )
 
             # Apply --no-quant / --no-compile overrides
-            _apply_stage_overrides(config_obj, no_quant=no_quant, no_compile=no_compile)
+            _apply_stage_overrides(config_obj, no_quant=not quant, no_compile=no_compile)
 
             output_data = config_obj.to_dict()
             _is_onnx_mode = True
@@ -319,7 +320,7 @@ def config(
                     precision=precision,
                     trust_remote_code=trust_remote_code,
                     ep=ep,
-                    no_quant=no_quant,
+                    no_quant=not quant,
                     no_compile=no_compile,
                     output=output,
                     console=console,
@@ -347,15 +348,15 @@ def config(
             if module:
                 configs = generate_hf_build_config(module=module, **_shared_kwargs)
                 for cfg in configs:
-                    _apply_stage_overrides(cfg, no_quant=no_quant, no_compile=no_compile)
+                    _apply_stage_overrides(cfg, no_quant=not quant, no_compile=no_compile)
                 output_data = [cfg.to_dict() for cfg in configs]
                 _n_modules = len(configs)
                 # Use first config for display metadata
                 config_obj = configs[0] if configs else None
             else:
                 config_obj = generate_hf_build_config(**_shared_kwargs)
                 configs = []
-                _apply_stage_overrides(config_obj, no_quant=no_quant, no_compile=no_compile)
+                _apply_stage_overrides(config_obj, no_quant=not quant, no_compile=no_compile)
                 output_data = config_obj.to_dict()
                 _n_modules = 0
 

src/winml/modelkit/commands/eval.py

@@ -112,9 +112,9 @@
     help="Shuffle dataset before sampling.",
 )
 @click.option(
-    "--streaming",
-    is_flag=True,
+    "--streaming/--no-streaming",
     default=False,
+    show_default=True,
     help="Stream dataset instead of downloading fully.",
 )
 @click.option(

src/winml/modelkit/commands/export.py

@@ -71,24 +71,31 @@ def _delete_onnx_with_external_data(onnx_path: Path) -> None:
 )
 @cli_utils.output_option("Output ONNX file path (e.g., model.onnx)", required=True)
 @click.option(
-    "--with-report",
-    is_flag=True,
+    "--with-report/--no-with-report",
     default=False,
+    show_default=True,
     help="Generate full export reports (markdown, JSON, console tree)",
 )
+@click.option(
+    "--hierarchy/--no-hierarchy",
+    "hierarchy",
+    default=True,
+    show_default=True,
+    help="Embed hierarchy_tag metadata in ONNX output",
+)
 @click.option(
     "--clean-onnx",
-    "--no-hierarchy",
-    "no_hierarchy",
+    "clean_onnx",
     is_flag=True,
     default=False,
-    help="Skip embedding hierarchy_tag metadata in ONNX (clean ONNX output)",
+    hidden=True,
+    help="Deprecated alias for --no-hierarchy.",
 )
 @click.option(
-    "--dynamo",
+    "--dynamo/--no-dynamo",
     "dynamo",
-    is_flag=True,
     default=False,
+    show_default=True,
     help="Enable PyTorch 2.9+ dynamo export for rich node metadata",
 )
 @click.option(
@@ -132,7 +139,8 @@ def export(
     verbose: int,
     quiet: bool,
     with_report: bool,
-    no_hierarchy: bool,
+    hierarchy: bool,
+    clean_onnx: bool,
     dynamo: bool,
     torch_module: str | None,
     task: str | None,
@@ -195,15 +203,27 @@ def export(
         _build_export_dict = ec
         if not cli_utils.is_cli_provided(ctx, "task") and "task" in lc:
             task = lc["task"]
-        if not cli_utils.is_cli_provided(ctx, "no_hierarchy") and "enable_hierarchy_tags" in ec:
-            no_hierarchy = not ec["enable_hierarchy_tags"]
+        if (
+            not cli_utils.is_cli_provided(ctx, "hierarchy")
+            and not cli_utils.is_cli_provided(ctx, "clean_onnx")
+            and "enable_hierarchy_tags" in ec
+        ):
+            hierarchy = ec["enable_hierarchy_tags"]
         if not cli_utils.is_cli_provided(ctx, "dynamo") and "dynamo" in ec:
             dynamo = ec["dynamo"]
 
     from ..export import InputTensorSpec, OutputTensorSpec, WinMLExportConfig
     from ..export import export_pytorch as export_onnx
     from ..loader import load_hf_model
 
+    if clean_onnx:
+        click.echo(
+            "warning: --clean-onnx is deprecated; use --no-hierarchy instead.",
+            err=True,
+        )
+        if not cli_utils.is_cli_provided(ctx, "hierarchy"):
+            hierarchy = False
+
     # Configure logging — stderr only, shared format with the rest of the CLI.
     configure_logging(verbosity=verbose, quiet=quiet)
 
@@ -332,8 +352,8 @@ def export(
     # Layer 2: --export-config file overrides
     config_kwargs.update(export_config_dict)
     # Layer 3: explicit CLI options (highest precedence)
-    if cli_utils.is_cli_provided(ctx, "no_hierarchy"):
-        config_kwargs["enable_hierarchy_tags"] = not no_hierarchy
+    if cli_utils.is_cli_provided(ctx, "hierarchy") or cli_utils.is_cli_provided(ctx, "clean_onnx"):
+        config_kwargs["enable_hierarchy_tags"] = hierarchy
     if cli_utils.is_cli_provided(ctx, "verbose"):
         config_kwargs["verbose"] = bool(verbose)
     if cli_utils.is_cli_provided(ctx, "dynamo"):

src/winml/modelkit/commands/inspect.py

@@ -105,10 +105,10 @@ def _looks_like_local_path(model_id: str) -> bool:
     help="Override auto-detected task (e.g., image-classification, feature-extraction)",
 )
 @click.option(
-    "-H",
-    "--hierarchy",
-    is_flag=True,
+    "-H/-N",
+    "--hierarchy/--no-hierarchy",
     default=False,
+    show_default=True,
     help="Show HF module hierarchy (uses random weights, no weight download)",
 )
 @click.option(

src/winml/modelkit/commands/perf.py

@@ -1067,21 +1067,22 @@ def _run_simple_loop(
     help='JSON file with shape overrides (e.g., {"height": 480, "width": 480}).',
 )
 @click.option(
-    "--no-quantize",
-    is_flag=True,
-    default=False,
-    help="Skip quantization during model build",
+    "--quantize/--no-quantize",
+    "quantize",
+    default=True,
+    show_default=True,
+    help="Include quantization during model build (use --no-quantize to skip)",
 )
 @click.option(
-    "--rebuild",
-    is_flag=True,
+    "--rebuild/--no-rebuild",
     default=False,
+    show_default=True,
     help="Force rebuild even if cached artifacts exist",
 )
 @click.option(
-    "--ignore-cache",
-    is_flag=True,
+    "--ignore-cache/--no-ignore-cache",
     default=False,
+    show_default=True,
     help="Build from scratch in a temp folder (discard after benchmarking)",
 )
 @cli_utils.skip_build_option()
@@ -1097,9 +1098,9 @@ def _run_simple_loop(
     "not '--module encoder.layer.0.attention' (a path, will not match).",
 )
 @click.option(
-    "--monitor",
-    is_flag=True,
+    "--monitor/--no-monitor",
     default=False,
+    show_default=True,
     help="Show live hardware utilization chart for the benchmarked device (NPU, GPU, or CPU)",
 )
 @click.option(
@@ -1125,7 +1126,7 @@ def perf(
     output: Path | None,
     batch_size: int,
     shape_config_path: Path | None,
-    no_quantize: bool,
+    quantize: bool,
     rebuild: bool,
     ignore_cache: bool,
     skip_build: bool,
@@ -1216,7 +1217,7 @@ def perf(
             iterations=iterations,
             warmup=warmup,
             batch_size=batch_size,
-            no_quantize=no_quantize,
+            no_quantize=not quantize,
             output=output,
             verbose=bool(verbose),
             console=console,
@@ -1262,7 +1263,7 @@ def perf(
         warmup=warmup,
         batch_size=batch_size,
         output_path=output,
-        no_quantize=no_quantize,
+        no_quantize=not quantize,
         rebuild=rebuild,
         ignore_cache=ignore_cache,
         skip_build=skip_build,