recipe(segformer): add nvidia b4 ADE recipe coverage

docs/reference/supported-models.md

@@ -191,6 +191,7 @@ testing. Use `winml catalog` to browse the full list interactively.
 | `mattmdjaga/segformer_b2_clothes` | Segformer |
 | `nvidia/segformer-b1-finetuned-ade-512-512` | Segformer |
 | `nvidia/segformer-b2-finetuned-ade-512-512` | Segformer |
+| `nvidia/segformer-b4-finetuned-ade-512-512` | Segformer |
 | `nvidia/segformer-b5-finetuned-ade-640-640` | Segformer |
 
 ### Image-to-Text

examples/recipes/README.md

@@ -14,7 +14,7 @@ Each *(model, task)* includes:
 
 ## Models
 
-Total: **75** (model, task) tuples that pass fp16 eval on all 10 (EP, device) buckets.
+Total: **77** (model, task) tuples that pass fp16 eval on all 10 (EP, device) buckets.
 
 | Model | Task |
 |---|---|
@@ -77,6 +77,7 @@ Total: **75** (model, task) tuples that pass fp16 eval on all 10 (EP, device) bu
 | monologg/koelectra-small-v2-distilled-korquad-384 | question-answering |
 | nvidia/segformer-b1-finetuned-ade-512-512 | image-segmentation |
 | nvidia/segformer-b2-finetuned-ade-512-512 | image-segmentation |
+| nvidia/segformer-b4-finetuned-ade-512-512 | image-segmentation |
 | nvidia/segformer-b5-finetuned-ade-640-640 | image-segmentation |
 | openai/clip-vit-base-patch16 | feature-extraction |
 | openai/clip-vit-base-patch16 | zero-shot-image-classification |

examples/recipes/nvidia_segformer-b4-finetuned-ade-512-512/image-segmentation_fp16_config.json

@@ -0,0 +1,54 @@
+{
+  "export": {
+    "opset_version": 17,
+    "batch_size": 1,
+    "export_params": true,
+    "do_constant_folding": true,
+    "verbose": false,
+    "dynamo": false,
+    "enable_hierarchy_tags": true,
+    "clean_onnx": false,
+    "hierarchy_tag_format": "full",
+    "input_tensors": [
+      {
+        "name": "pixel_values",
+        "dtype": "float32",
+        "shape": [
+          1,
+          3,
+          512,
+          512
+        ],
+        "value_range": [
+          0,
+          1
+        ]
+      }
+    ],
+    "output_tensors": [
+      {
+        "name": "logits"
+      }
+    ]
+  },
+  "optim": {},
+  "quant": null,
+  "compile": null,
+  "loader": {
+    "task": "image-segmentation",
+    "model_class": "AutoModelForSemanticSegmentation",
+    "model_type": "segformer"
+  },
+  "eval": {
+    "task": "image-segmentation",
+    "dataset": {
+      "path": "danjacobellis/scene_parse_150",
+      "split": "validation",
+      "samples": 100,
+      "label_mapping_file": "scripts/e2e_eval/datasets/ade20k_gt_to_model_label.json",
+      "columns_mapping": {
+        "annotation_column": "annotation"
+      }
+    }
+  }
+}

examples/recipes/nvidia_segformer-b4-finetuned-ade-512-512/image-segmentation_w8a16_config.json

@@ -0,0 +1,71 @@
+{
+  "export": {
+    "opset_version": 17,
+    "batch_size": 1,
+    "export_params": true,
+    "do_constant_folding": true,
+    "verbose": false,
+    "dynamo": false,
+    "enable_hierarchy_tags": true,
+    "clean_onnx": false,
+    "hierarchy_tag_format": "full",
+    "input_tensors": [
+      {
+        "name": "pixel_values",
+        "dtype": "float32",
+        "shape": [
+          1,
+          3,
+          512,
+          512
+        ],
+        "value_range": [
+          0,
+          1
+        ]
+      }
+    ],
+    "output_tensors": [
+      {
+        "name": "logits"
+      }
+    ]
+  },
+  "optim": {},
+  "quant": {
+    "mode": "qdq",
+    "samples": 10,
+    "calibration_method": "minmax",
+    "weight_type": "uint8",
+    "activation_type": "uint16",
+    "per_channel": false,
+    "symmetric": false,
+    "save_calibration": false,
+    "distribution": "uniform",
+    "seed": null,
+    "calibration_load_path": null,
+    "calibration_save_path": null,
+    "op_types_to_quantize": null,
+    "nodes_to_exclude": null,
+    "task": "image-segmentation",
+    "model_id": "nvidia/segformer-b4-finetuned-ade-512-512"
+  },
+  "compile": null,
+  "loader": {
+    "task": "image-segmentation",
+    "model_class": "AutoModelForSemanticSegmentation",
+    "model_type": "segformer"
+  },
+  "eval": {
+    "task": "image-segmentation",
+    "dataset": {
+      "path": "danjacobellis/scene_parse_150",
+      "split": "validation",
+      "samples": 1000,
+      "label_mapping_file": "scripts/e2e_eval/datasets/ade20k_gt_to_model_label.json",
+      "columns_mapping": {
+        "annotation_column": "annotation"
+      }
+    }
+  }
+}

scripts/e2e_eval/testsets/models_with_acc.json

@@ -516,6 +516,23 @@
       }
     }
   },
+  {
+    "hf_id": "nvidia/segformer-b4-finetuned-ade-512-512",
+    "task": "image-segmentation",
+    "model_type": "segformer",
+    "group": "Top200",
+    "priority": "P1",
+    "dataset_config": {
+      "path": "danjacobellis/scene_parse_150",
+      "split": "validation",
+      "metric": "mean_iou",
+      "winml_metric_key": "mean_iou",
+      "label_mapping_file": "scripts/e2e_eval/datasets/ade20k_gt_to_model_label.json",
+      "columns_mapping": {
+        "annotation_column": "annotation"
+      }
+    }
+  },
   {
     "hf_id": "nvidia/segformer-b5-finetuned-ade-640-640",
     "task": "image-segmentation",

src/winml/modelkit/data/hub_models.json

@@ -1569,6 +1569,38 @@
       },
       "size_mb": 27.5
     },
+    {
+      "model_id": "nvidia/segformer-b4-finetuned-ade-512-512",
+      "task": "image-segmentation",
+      "model_type": "segformer",
+      "supported_eps": {
+        "cpu": [
+          "CPU"
+        ],
+        "dml": [
+          "GPU"
+        ],
+        "qnn": [
+          "GPU",
+          "NPU"
+        ],
+        "openvino": [
+          "CPU",
+          "GPU",
+          "NPU"
+        ],
+        "vitisai": [
+          "NPU"
+        ],
+        "trtrtx": [
+          "GPU"
+        ],
+        "migraphx": [
+          "GPU"
+        ]
+      },
+      "size_mb": 64.1
+    },
     {
       "model_id": "nvidia/segformer-b5-finetuned-ade-640-640",
       "task": "image-segmentation",