{
  "mic": "0.3",
  "name": "PVT image classification",
  "usage": "PvtImageProcessor + PvtForImageClassification, batched synthetic images",
  "architecture": "PvtForImageClassification",
  "checkpoints": {
    "Zetatech/pvt-tiny-224": {
      "fixture": "huggingface-mic/pvt-tiny-224-mic"
    }
  },
  "fixtures": {
    "huggingface-mic/pvt-tiny-224-mic": {
      "built_from": "Zetatech/pvt-tiny-224",
      "covers": [
        "Zetatech/pvt-tiny-224"
      ],
      "build": {
        "seed": 17,
        "config_overrides": {
          "fields": {
            "num_attention_heads": [
              1,
              2,
              4,
              8
            ]
          },
          "reason": "The generic rule shrinks PVT's per-stage hidden_sizes [64,128,320,512] to [16,32,64,128] but leaves the per-stage num_attention_heads [1,2,5,8]; stage 3's 64 channels are not divisible by 5 heads, so the model cannot be constructed. [1,2,4,8] keeps the multi-head layout with valid head widths."
        }
      }
    }
  },
  "integrations": {
    "transformers": {
      "entry_point": "integrations/transformers/usage.md",
      "format": "runnable-markdown",
      "requires": [
        "transformers>=5",
        "torch"
      ],
      "targets": [
        "fixtures",
        "checkpoints"
      ],
      "variants": {
        "default": {},
        "eager": {
          "attn_implementation": "eager"
        },
        "bfloat16": {
          "dtype": "bfloat16"
        }
      },
      "training": true,
      "resources": {
        "fixture": {
          "memory_gb": 1,
          "download_mb": 20
        },
        "Zetatech/pvt-tiny-224": {
          "memory_gb": 1,
          "download_mb": 20
        }
      }
    }
  },
  "outputs": {
    "runtime": {
      "compare": "exact"
    },
    "predicted_classes": {
      "compare": "exact"
    },
    "logits_probe": {
      "compare": "allclose",
      "atol": 0.0001,
      "rtol": 0.0001
    },
    "training.loss": {
      "compare": "allclose",
      "atol": 0.0001,
      "rtol": 0.0001
    },
    "training.gradients": {
      "compare": "all_true"
    },
    "training.updated": {
      "compare": "true"
    }
  }
}