Publish MIC usage, slim fixture and reviewed Transformers expectations

.mic-repo.json

@@ -0,0 +1,7 @@
+{
+  "format": "mic-umbrella-v1",
+  "contract_files": [
+    "integration.json",
+    "integrations/transformers/usage.md"
+  ]
+}

README.md

@@ -0,0 +1,37 @@
+---
+tags:
+  - mic
+  - test-artifact
+base_model: facebook/dinov2-small
+---
+
+# dinov2-small-mic
+
+Slim model weights, executable usage and framework expectations in one repository.
+
+- [Contract manifest](integration.json)
+- [Transformers usage](integrations/transformers/usage.md)
+- [Transformers expectations](integrations/transformers/expectations)
+- [Fixture provenance](build.json)
+
+## Contract
+
+### DINOv2 image features
+
+Runnable usage and the framework expectations directory are linked above.
+
+## Slim model
+
+### Slim fixture: dinov2-small-mic
+
+A 0.2M-parameter fixture for the [DINOv2 MIC](https://huggingface.co/huggingface-mic/dinov2-small-mic), covering [`facebook/dinov2-small`](https://huggingface.co/facebook/dinov2-small). It is a test artifact, not a usable model, tagged `test-artifact` to keep it out of candidate-model search.
+
+- **Fixture key `c18a1be82d3e`.** It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
+- **Same architecture, smaller config:** `Dinov2Model`, shrunk by the generic rule in [`sweep.py`](https://github.com/huggingface/transformers-mic/blob/main/experiments/sweep.py) (num_hidden_layers 2).
+- **Processor files copied byte for byte** from `facebook/dinov2-small`: `preprocessor_config.json`.
+- **Random weights (seed 17).** They passed the input-sensitivity gate: every perturbation of the contract's inputs moved the logits by at least 0.001 (weakest: `pixel_values:halve`, 0.331), so no training was needed.
+- **Same tensor names as the real checkpoint.** The real checkpoint has every parameter, and so does the fixture.
+- **Saved in float32**, the dtype `facebook/dinov2-small` loads in by default.
+- **Finite in float32, bfloat16, and float16** (forward, loss, gradients).
+
+Built by [`build_fixtures.py`](https://github.com/huggingface/transformers-mic/blob/main/example/build_fixtures.py); [`build.json`](build.json) records the pinned sources, gate results, and file hashes. This repository includes the slim weights, usage and framework expectations; rebuild the other files with `python example/build_fixtures.py --fixture dinov2-small-mic`.

build.json

@@ -0,0 +1,132 @@
+{
+ "fixture": "huggingface-mic/dinov2-small-mic",
+ "key": "c18a1be82d3e7a3a60b8a875a1608fe4f8e9a583b6f10c9620d046e7ee68ce64",
+ "key_fields": {
+  "architectures": [
+   "Dinov2Model"
+  ],
+  "output_hidden_states": false,
+  "return_dict": true,
+  "chunk_size_feed_forward": 0,
+  "is_encoder_decoder": false,
+  "id2label": {
+   "0": "LABEL_0",
+   "1": "LABEL_1"
+  },
+  "label2id": {
+   "LABEL_0": 0,
+   "LABEL_1": 1
+  },
+  "problem_type": null,
+  "hidden_size": 64,
+  "num_hidden_layers": 2,
+  "num_attention_heads": 4,
+  "mlp_ratio": 4,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.0,
+  "attention_probs_dropout_prob": 0.0,
+  "initializer_range": 0.02,
+  "layer_norm_eps": 1e-06,
+  "image_size": 518,
+  "patch_size": 14,
+  "num_channels": 3,
+  "qkv_bias": true,
+  "layerscale_value": 1.0,
+  "drop_path_rate": 0.0,
+  "use_swiglu_ffn": false,
+  "apply_layernorm": true,
+  "reshape_hidden_states": true,
+  "use_mask_token": true,
+  "stage_names": [
+   "stem",
+   "stage1",
+   "stage2",
+   "stage3",
+   "stage4",
+   "stage5",
+   "stage6",
+   "stage7",
+   "stage8",
+   "stage9",
+   "stage10",
+   "stage11",
+   "stage12"
+  ],
+  "model_type": "dinov2",
+  "output_attentions": false,
+  "out_features": [
+   "stage2"
+  ],
+  "out_indices": [
+   2
+  ]
+ },
+ "covers": {
+  "facebook/dinov2-small": "ed25f3a31f01632728cabb09d1542f84ab7b0056"
+ },
+ "built_from": {
+  "repo": "facebook/dinov2-small",
+  "revision": "ed25f3a31f01632728cabb09d1542f84ab7b0056"
+ },
+ "seed": 17,
+ "dtype": "float32",
+ "builder": "example/build_fixtures.py",
+ "model_class": "Dinov2Model",
+ "parameters": 225856,
+ "missing_parameters": [],
+ "checkpoint_inspection": "parameter headers/indexes; bounded binary fallback",
+ "numeric_health_scope": "CPU float32/bfloat16/float16; autocast for low-precision losses",
+ "captured": {
+  "call": "training step",
+  "inputs": [
+   "pixel_values"
+  ],
+  "class": "Dinov2Model"
+ },
+ "weights": "random",
+ "gate": {
+  "random": {
+   "pass": true,
+   "accepted_insensitive": {},
+   "threshold": 0.001,
+   "deltas": {
+    "pixel_values:roll_batch": 4.062774658203125,
+    "pixel_values:swap_channels": 4.1026763916015625,
+    "pixel_values:reverse": null,
+    "pixel_values:halve": 0.33074086904525757
+   },
+   "weakest": "pixel_values:halve",
+   "no_perturbation_defined": [],
+   "not_a_forward_parameter": []
+  }
+ },
+ "numeric_health": {
+  "float32": {
+   "outputs_finite": true,
+   "finite": true
+  },
+  "bfloat16": {
+   "outputs_finite": true,
+   "finite": true
+  },
+  "float16": {
+   "outputs_finite": true,
+   "finite": true
+  }
+ },
+ "copied_unchanged": [
+  "preprocessor_config.json"
+ ],
+ "environment": {
+  "python": "3.12.12",
+  "machine": "arm64",
+  "torch": "2.14.1",
+  "transformers": "5.22.0.dev0",
+  "tokenizers": "0.23.3"
+ },
+ "sha256": {
+  "config.json": "d4477978e933094096c9a6245903ab43754b1cb8ce7e6154557a62279b401d65",
+  "model.safetensors": "c3912762cb2c791b01d4b9b5b7cd51c5562cd6a1d3ad9448ff4c1a92efc753ed",
+  "preprocessor_config.json": "14e780d86fa1861f8751f868d7f45425b5feb55c38ca26f152ca5097ab30f828"
+ }
+}

config.json

@@ -0,0 +1,48 @@
+{
+  "apply_layernorm": true,
+  "architectures": [
+    "Dinov2Model"
+  ],
+  "attention_probs_dropout_prob": 0.0,
+  "drop_path_rate": 0.0,
+  "dtype": "float32",
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.0,
+  "hidden_size": 64,
+  "image_size": 518,
+  "initializer_range": 0.02,
+  "layer_norm_eps": 1e-06,
+  "layerscale_value": 1.0,
+  "mlp_ratio": 4,
+  "model_type": "dinov2",
+  "num_attention_heads": 4,
+  "num_channels": 3,
+  "num_hidden_layers": 2,
+  "out_features": [
+    "stage2"
+  ],
+  "out_indices": [
+    2
+  ],
+  "patch_size": 14,
+  "qkv_bias": true,
+  "reshape_hidden_states": true,
+  "stage_names": [
+    "stem",
+    "stage1",
+    "stage2",
+    "stage3",
+    "stage4",
+    "stage5",
+    "stage6",
+    "stage7",
+    "stage8",
+    "stage9",
+    "stage10",
+    "stage11",
+    "stage12"
+  ],
+  "transformers_version": "5.22.0.dev0",
+  "use_mask_token": true,
+  "use_swiglu_ffn": false
+}

integration.json

@@ -0,0 +1,83 @@
+{
+  "mic": "0.3",
+  "name": "DINOv2 image features",
+  "usage": "AutoImageProcessor + AutoModel, batched images, last_hidden_state probes, one SGD step on the CLS token",
+  "architecture": "Dinov2Model",
+  "checkpoints": {
+    "facebook/dinov2-small": {
+      "fixture": "huggingface-mic/dinov2-small-mic"
+    }
+  },
+  "fixtures": {
+    "huggingface-mic/dinov2-small-mic": {
+      "built_from": "facebook/dinov2-small",
+      "covers": [
+        "facebook/dinov2-small"
+      ],
+      "build": {
+        "seed": 17
+      }
+    }
+  },
+  "integrations": {
+    "transformers": {
+      "entry_point": "integrations/transformers/usage.md",
+      "format": "runnable-markdown",
+      "requires": [
+        "transformers>=5",
+        "torch",
+        "pillow"
+      ],
+      "targets": [
+        "fixtures",
+        "checkpoints"
+      ],
+      "variants": {
+        "default": {},
+        "eager": {
+          "attn_implementation": "eager"
+        },
+        "bfloat16": {
+          "dtype": "bfloat16"
+        }
+      },
+      "training": true,
+      "resources": {
+        "fixture": {
+          "memory_gb": 2,
+          "download_mb": 90
+        },
+        "facebook/dinov2-small": {
+          "memory_gb": 4,
+          "download_mb": 90
+        }
+      }
+    }
+  },
+  "outputs": {
+    "runtime": {
+      "compare": "exact"
+    },
+    "cls_features": {
+      "compare": "allclose",
+      "atol": 0.0001,
+      "rtol": 0.0001
+    },
+    "patch_features": {
+      "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"
+    }
+  }
+}

integrations/transformers/expectations/cpu-x86_64-bf16/dinov2-small-mic.bfloat16.json

@@ -0,0 +1,116 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "bfloat16",
+  "target": "dinov2-small-mic",
+  "kind": "fixture",
+  "revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.36,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64-bf16"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "bfloat16",
+   "attn_implementation": "sdpa"
+  },
+  "cls_features": [
+   [
+    -1.265625,
+    0.451171875,
+    -0.9765625,
+    -1.6953125,
+    -0.11083984375
+   ],
+   [
+    -0.158203125,
+    0.625,
+    -0.875,
+    -0.56640625,
+    -0.345703125
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     -0.458984375,
+     0.05810546875,
+     -0.51953125,
+     0.24609375,
+     -0.11474609375
+    ],
+    [
+     -0.419921875,
+     0.07763671875,
+     -0.49609375,
+     0.2890625,
+     -0.056396484375
+    ],
+    [
+     -0.3984375,
+     0.09619140625,
+     -0.51953125,
+     0.33984375,
+     -0.14453125
+    ]
+   ],
+   [
+    [
+     0.1396484375,
+     -0.54296875,
+     -0.248046875,
+     0.4140625,
+     -1.0078125
+    ],
+    [
+     0.1767578125,
+     -0.51953125,
+     -0.224609375,
+     0.466796875,
+     -0.9453125
+    ],
+    [
+     0.19921875,
+     -0.49609375,
+     -0.25,
+     0.515625,
+     -1.0390625
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.396484375,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cpu-x86_64-bf16/dinov2-small.bfloat16.json

@@ -0,0 +1,116 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "bfloat16",
+  "target": "dinov2-small",
+  "kind": "checkpoint",
+  "revision": "ed25f3a31f01632728cabb09d1542f84ab7b0056",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.59,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64-bf16"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "bfloat16",
+   "attn_implementation": "sdpa"
+  },
+  "cls_features": [
+   [
+    3.25,
+    1.390625,
+    3.53125,
+    0.6796875,
+    1.0703125
+   ],
+   [
+    2.71875,
+    1.4921875,
+    4.1875,
+    -1.125,
+    -0.75
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     2.671875,
+     -2.4375,
+     1.7109375,
+     -1.7421875,
+     0.039306640625
+    ],
+    [
+     2.65625,
+     -1.2734375,
+     3.8125,
+     -3.71875,
+     -0.4921875
+    ],
+    [
+     2.21875,
+     -6.4375,
+     1.015625,
+     -1.9296875,
+     -0.181640625
+    ]
+   ],
+   [
+    [
+     1.4921875,
+     -2.09375,
+     3.234375,
+     -2.859375,
+     -0.64453125
+    ],
+    [
+     0.50390625,
+     -1.6015625,
+     3.828125,
+     -3.8125,
+     -1.5625
+    ],
+    [
+     1.328125,
+     -5.9375,
+     2.828125,
+     -3.0,
+     -1.4296875
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.408203125,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cpu-x86_64/dinov2-small-mic.eager.json

@@ -0,0 +1,116 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "eager",
+  "target": "dinov2-small-mic",
+  "kind": "fixture",
+  "revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.33,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "float32",
+   "attn_implementation": "eager"
+  },
+  "cls_features": [
+   [
+    -1.2677825689315796,
+    0.44753769040107727,
+    -0.9715877175331116,
+    -1.6918056011199951,
+    -0.10733163356781006
+   ],
+   [
+    -0.16284990310668945,
+    0.6147457361221313,
+    -0.878085196018219,
+    -0.5765715837478638,
+    -0.34631210565567017
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     -0.4561893939971924,
+     0.06119873747229576,
+     -0.5198690891265869,
+     0.24606479704380035,
+     -0.11510999500751495
+    ],
+    [
+     -0.4219924509525299,
+     0.0810282751917839,
+     -0.49528130888938904,
+     0.2951686978340149,
+     -0.05753913149237633
+    ],
+    [
+     -0.4002307057380676,
+     0.1002686396241188,
+     -0.5202704071998596,
+     0.34013617038726807,
+     -0.14419952034950256
+    ]
+   ],
+   [
+    [
+     0.13712136447429657,
+     -0.5413686037063599,
+     -0.25230535864830017,
+     0.4157020151615143,
+     -1.0112773180007935
+    ],
+    [
+     0.17359992861747742,
+     -0.5173826813697815,
+     -0.22490383684635162,
+     0.46642279624938965,
+     -0.9491122364997864
+    ],
+    [
+     0.19613665342330933,
+     -0.4979003071784973,
+     -0.25296491384506226,
+     0.5143538117408752,
+     -1.0416719913482666
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.3960122764110565,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cpu-x86_64/dinov2-small-mic.json

@@ -0,0 +1,116 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "default",
+  "target": "dinov2-small-mic",
+  "kind": "fixture",
+  "revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.19,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "float32",
+   "attn_implementation": "sdpa"
+  },
+  "cls_features": [
+   [
+    -1.2677828073501587,
+    0.44753751158714294,
+    -0.9715878367424011,
+    -1.6918061971664429,
+    -0.10733203589916229
+   ],
+   [
+    -0.16284960508346558,
+    0.6147460341453552,
+    -0.8780848383903503,
+    -0.5765717625617981,
+    -0.34631192684173584
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     -0.45618942379951477,
+     0.06119873747229576,
+     -0.5198691487312317,
+     0.24606478214263916,
+     -0.11511000990867615
+    ],
+    [
+     -0.4219924509525299,
+     0.08102825284004211,
+     -0.49528130888938904,
+     0.2951686382293701,
+     -0.057539138942956924
+    ],
+    [
+     -0.40023073554039,
+     0.10026858747005463,
+     -0.5202704071998596,
+     0.3401360809803009,
+     -0.14419953525066376
+    ]
+   ],
+   [
+    [
+     0.13712137937545776,
+     -0.5413686037063599,
+     -0.2523052990436554,
+     0.4157020151615143,
+     -1.0112773180007935
+    ],
+    [
+     0.17359991371631622,
+     -0.5173826813697815,
+     -0.2249038815498352,
+     0.46642276644706726,
+     -0.9491119980812073
+    ],
+    [
+     0.19613665342330933,
+     -0.49790048599243164,
+     -0.25296491384506226,
+     0.5143538117408752,
+     -1.0416719913482666
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.39601224660873413,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cpu-x86_64/dinov2-small.eager.json

@@ -0,0 +1,116 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "eager",
+  "target": "dinov2-small",
+  "kind": "checkpoint",
+  "revision": "ed25f3a31f01632728cabb09d1542f84ab7b0056",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.78,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "float32",
+   "attn_implementation": "eager"
+  },
+  "cls_features": [
+   [
+    3.3632171154022217,
+    1.3631565570831299,
+    3.5031769275665283,
+    0.7399592399597168,
+    1.1561686992645264
+   ],
+   [
+    2.873883008956909,
+    1.558150291442871,
+    4.299252033233643,
+    -1.1491706371307373,
+    -0.7426658868789673
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     2.8144359588623047,
+     -2.566210985183716,
+     1.6931791305541992,
+     -1.627620816230774,
+     0.19263553619384766
+    ],
+    [
+     2.8032729625701904,
+     -1.2689285278320312,
+     3.8074841499328613,
+     -3.6497232913970947,
+     -0.40225517749786377
+    ],
+    [
+     2.3764963150024414,
+     -6.332672119140625,
+     0.9942713975906372,
+     -1.8182226419448853,
+     -0.050021421164274216
+    ]
+   ],
+   [
+    [
+     1.7134679555892944,
+     -2.1021156311035156,
+     3.2496793270111084,
+     -2.8441338539123535,
+     -0.6768681406974792
+    ],
+    [
+     0.6792188882827759,
+     -1.5355024337768555,
+     4.053866863250732,
+     -3.7946765422821045,
+     -1.4759902954101562
+    ],
+    [
+     1.5326447486877441,
+     -5.820225238800049,
+     2.909334659576416,
+     -2.9413650035858154,
+     -1.2621461153030396
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.40739622712135315,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cpu-x86_64/dinov2-small.json

@@ -0,0 +1,116 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "default",
+  "target": "dinov2-small",
+  "kind": "checkpoint",
+  "revision": "ed25f3a31f01632728cabb09d1542f84ab7b0056",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cpu",
+   "torchvision": "0.29.2+cpu",
+   "torchaudio": "2.11.1+cpu",
+   "tokenizers": "0.23.3",
+   "numpy": "2.3.5",
+   "pillow": "13.0.0",
+   "accelerate": "1.15.0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.78,
+  "environment_differs": {
+   "torchvision": "0.29.2+cpu (pinned 0.29.1)",
+   "torchaudio": "2.11.1+cpu (pinned 2.11.0)",
+   "pillow": "13.0.0 (pinned 12.3.0)"
+  },
+  "platform": "cpu-x86_64"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cpu",
+   "dtype": "float32",
+   "attn_implementation": "sdpa"
+  },
+  "cls_features": [
+   [
+    3.363215684890747,
+    1.363155484199524,
+    3.503181219100952,
+    0.7399603128433228,
+    1.1561681032180786
+   ],
+   [
+    2.8738789558410645,
+    1.5581508874893188,
+    4.299252033233643,
+    -1.1491740942001343,
+    -0.7426685690879822
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     2.8144359588623047,
+     -2.566210985183716,
+     1.6931726932525635,
+     -1.6276195049285889,
+     0.1926337629556656
+    ],
+    [
+     2.8032710552215576,
+     -1.2689297199249268,
+     3.8074841499328613,
+     -3.6497254371643066,
+     -0.40224969387054443
+    ],
+    [
+     2.3764944076538086,
+     -6.33266544342041,
+     0.9942771792411804,
+     -1.8182300329208374,
+     -0.05001138150691986
+    ]
+   ],
+   [
+    [
+     1.7134637832641602,
+     -2.102106809616089,
+     3.249678373336792,
+     -2.844135284423828,
+     -0.6768750548362732
+    ],
+    [
+     0.6792117357254028,
+     -1.5354945659637451,
+     4.053877830505371,
+     -3.7946696281433105,
+     -1.475988745689392
+    ],
+    [
+     1.5326398611068726,
+     -5.820218086242676,
+     2.909342050552368,
+     -2.9413583278656006,
+     -1.2621601819992065
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.40739724040031433,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/dinov2-small-mic.bfloat16.json

@@ -0,0 +1,117 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "bfloat16",
+  "target": "dinov2-small-mic",
+  "kind": "fixture",
+  "revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.17,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "bfloat16",
+   "attn_implementation": "sdpa"
+  },
+  "cls_features": [
+   [
+    -1.265625,
+    0.451171875,
+    -0.9765625,
+    -1.6953125,
+    -0.11083984375
+   ],
+   [
+    -0.158203125,
+    0.625,
+    -0.875,
+    -0.56640625,
+    -0.345703125
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     -0.458984375,
+     0.05810546875,
+     -0.51953125,
+     0.24609375,
+     -0.11474609375
+    ],
+    [
+     -0.419921875,
+     0.0771484375,
+     -0.49609375,
+     0.2890625,
+     -0.056396484375
+    ],
+    [
+     -0.3984375,
+     0.09619140625,
+     -0.51953125,
+     0.33984375,
+     -0.14453125
+    ]
+   ],
+   [
+    [
+     0.1396484375,
+     -0.54296875,
+     -0.248046875,
+     0.4140625,
+     -1.0078125
+    ],
+    [
+     0.1767578125,
+     -0.51953125,
+     -0.224609375,
+     0.466796875,
+     -0.9453125
+    ],
+    [
+     0.19921875,
+     -0.49609375,
+     -0.25,
+     0.515625,
+     -1.0390625
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.396484375,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/dinov2-small-mic.eager.json

@@ -0,0 +1,117 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "eager",
+  "target": "dinov2-small-mic",
+  "kind": "fixture",
+  "revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.24,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "float32",
+   "attn_implementation": "eager"
+  },
+  "cls_features": [
+   [
+    -1.2679601907730103,
+    0.44750577211380005,
+    -0.971464216709137,
+    -1.6919364929199219,
+    -0.10752461850643158
+   ],
+   [
+    -0.16286683082580566,
+    0.6150127053260803,
+    -0.8777503371238708,
+    -0.5766773223876953,
+    -0.34621816873550415
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     -0.45605921745300293,
+     0.060752131044864655,
+     -0.5195971727371216,
+     0.246177077293396,
+     -0.11525414139032364
+    ],
+    [
+     -0.4218588173389435,
+     0.08058323711156845,
+     -0.49500611424446106,
+     0.2952865958213806,
+     -0.05767730250954628
+    ],
+    [
+     -0.40009447932243347,
+     0.09982605278491974,
+     -0.51999831199646,
+     0.3402586579322815,
+     -0.1443466991186142
+    ]
+   ],
+   [
+    [
+     0.13699482381343842,
+     -0.5414605736732483,
+     -0.25245267152786255,
+     0.4158841669559479,
+     -1.0113370418548584
+    ],
+    [
+     0.17346715927124023,
+     -0.5174773931503296,
+     -0.22505569458007812,
+     0.4665941894054413,
+     -0.949180006980896
+    ],
+    [
+     0.19600003957748413,
+     -0.4979996383190155,
+     -0.25311195850372314,
+     0.5145183801651001,
+     -1.0417256355285645
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.3959294557571411,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/dinov2-small-mic.json

@@ -0,0 +1,117 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "default",
+  "target": "dinov2-small-mic",
+  "kind": "fixture",
+  "revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 7.32,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "float32",
+   "attn_implementation": "sdpa"
+  },
+  "cls_features": [
+   [
+    -1.2679606676101685,
+    0.4475054442882538,
+    -0.9714641571044922,
+    -1.6919360160827637,
+    -0.10752378404140472
+   ],
+   [
+    -0.16286709904670715,
+    0.615013062953949,
+    -0.8777508735656738,
+    -0.5766777396202087,
+    -0.34621813893318176
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     -0.45605921745300293,
+     0.060752153396606445,
+     -0.5195971131324768,
+     0.2461770921945572,
+     -0.11525411903858185
+    ],
+    [
+     -0.4218587279319763,
+     0.08058319985866547,
+     -0.4950060546398163,
+     0.2952865660190582,
+     -0.05767732858657837
+    ],
+    [
+     -0.4000944495201111,
+     0.09982606023550034,
+     -0.5199982523918152,
+     0.3402586579322815,
+     -0.14434663951396942
+    ]
+   ],
+   [
+    [
+     0.13699479401111603,
+     -0.5414605736732483,
+     -0.2524527311325073,
+     0.4158841669559479,
+     -1.0113370418548584
+    ],
+    [
+     0.17346712946891785,
+     -0.5174773931503296,
+     -0.2250557243824005,
+     0.46659424901008606,
+     -0.9491799473762512
+    ],
+    [
+     0.19600002467632294,
+     -0.49799954891204834,
+     -0.2531120479106903,
+     0.5145183801651001,
+     -1.041725754737854
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.395929217338562,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/dinov2-small.bfloat16.json

@@ -0,0 +1,117 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "bfloat16",
+  "target": "dinov2-small",
+  "kind": "checkpoint",
+  "revision": "ed25f3a31f01632728cabb09d1542f84ab7b0056",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 6.99,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "bfloat16",
+   "attn_implementation": "sdpa"
+  },
+  "cls_features": [
+   [
+    3.234375,
+    1.421875,
+    3.53125,
+    0.671875,
+    1.109375
+   ],
+   [
+    2.75,
+    1.4765625,
+    4.21875,
+    -1.1015625,
+    -0.7421875
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     2.640625,
+     -2.578125,
+     1.7109375,
+     -1.75,
+     0.1455078125
+    ],
+    [
+     2.640625,
+     -1.2734375,
+     3.765625,
+     -3.75,
+     -0.37890625
+    ],
+    [
+     2.1875,
+     -6.3125,
+     1.0859375,
+     -2.015625,
+     -0.09228515625
+    ]
+   ],
+   [
+    [
+     1.4921875,
+     -2.125,
+     3.1875,
+     -2.796875,
+     -0.59765625
+    ],
+    [
+     0.5234375,
+     -1.6875,
+     3.90625,
+     -3.75,
+     -1.53125
+    ],
+    [
+     1.375,
+     -5.9375,
+     2.71875,
+     -2.875,
+     -1.3125
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.41796875,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/dinov2-small.eager.json

@@ -0,0 +1,117 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "eager",
+  "target": "dinov2-small",
+  "kind": "checkpoint",
+  "revision": "ed25f3a31f01632728cabb09d1542f84ab7b0056",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 6.92,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "float32",
+   "attn_implementation": "eager"
+  },
+  "cls_features": [
+   [
+    3.3632185459136963,
+    1.3631588220596313,
+    3.503180980682373,
+    0.7399594187736511,
+    1.1561698913574219
+   ],
+   [
+    2.8738818168640137,
+    1.5581567287445068,
+    4.299259662628174,
+    -1.1491749286651611,
+    -0.742664635181427
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     2.8144304752349854,
+     -2.5662131309509277,
+     1.6931780576705933,
+     -1.62762451171875,
+     0.19264210760593414
+    ],
+    [
+     2.803272008895874,
+     -1.2689350843429565,
+     3.8074729442596436,
+     -3.649724245071411,
+     -0.4022521674633026
+    ],
+    [
+     2.3764851093292236,
+     -6.3326802253723145,
+     0.9942670464515686,
+     -1.8182239532470703,
+     -0.0500158928334713
+    ]
+   ],
+   [
+    [
+     1.7134647369384766,
+     -2.102114200592041,
+     3.249678373336792,
+     -2.8441355228424072,
+     -0.6768700480461121
+    ],
+    [
+     0.6792080998420715,
+     -1.5354911088943481,
+     4.0538835525512695,
+     -3.7946691513061523,
+     -1.4759985208511353
+    ],
+    [
+     1.5326409339904785,
+     -5.820221900939941,
+     2.909335136413574,
+     -2.9413630962371826,
+     -1.2621474266052246
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.40739744901657104,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/expectations/cuda/dinov2-small.json

@@ -0,0 +1,117 @@
+{
+ "recorded_with": {
+  "contract": "huggingface-mic/dinov2-small-mic",
+  "contract_name": "DINOv2",
+  "contract_revision": "2551b1df5f6f8020ee15b17aa9b7549e364fbfdb",
+  "framework": "transformers",
+  "variant": "default",
+  "target": "dinov2-small",
+  "kind": "checkpoint",
+  "revision": "ed25f3a31f01632728cabb09d1542f84ab7b0056",
+  "environment": {
+   "python": "3.11.18",
+   "transformers": "5.22.0.dev0",
+   "torch": "2.14.0+cu130",
+   "torchvision": "0.29.2",
+   "torchaudio": "2.11.0+cu130",
+   "tokenizers": "0.23.3",
+   "numpy": "2.4.6",
+   "pillow": "13.0.0",
+   "accelerate": "1.16.0.dev0",
+   "safetensors": "0.8.0",
+   "huggingface-hub": "2.3.0",
+   "sentencepiece": "0.2.2",
+   "protobuf": "7.36.2",
+   "librosa": "0.11.0",
+   "soundfile": "0.14.0",
+   "scipy": "1.17.1"
+  },
+  "seconds": 6.67,
+  "environment_differs": {
+   "torchvision": "0.29.2 (pinned 0.29.1)",
+   "numpy": "2.4.6 (pinned 2.3.5)",
+   "pillow": "13.0.0 (pinned 12.3.0)",
+   "accelerate": "1.16.0.dev0 (pinned 1.15.0)"
+  },
+  "platform": "cuda"
+ },
+ "outputs": {
+  "runtime": {
+   "device": "cuda",
+   "dtype": "float32",
+   "attn_implementation": "sdpa"
+  },
+  "cls_features": [
+   [
+    3.3632168769836426,
+    1.3631563186645508,
+    3.503182888031006,
+    0.7399498820304871,
+    1.1561646461486816
+   ],
+   [
+    2.8738813400268555,
+    1.5581508874893188,
+    4.299262523651123,
+    -1.1491715908050537,
+    -0.7426652908325195
+   ]
+  ],
+  "patch_features": [
+   [
+    [
+     2.81443452835083,
+     -2.5662147998809814,
+     1.6931688785552979,
+     -1.627625823020935,
+     0.19263756275177002
+    ],
+    [
+     2.803272247314453,
+     -1.2689265012741089,
+     3.8074750900268555,
+     -3.6497342586517334,
+     -0.40226009488105774
+    ],
+    [
+     2.376483201980591,
+     -6.33268404006958,
+     0.9942582845687866,
+     -1.8182166814804077,
+     -0.05002152547240257
+    ]
+   ],
+   [
+    [
+     1.7134617567062378,
+     -2.102113723754883,
+     3.2496745586395264,
+     -2.8441359996795654,
+     -0.6768668293952942
+    ],
+    [
+     0.6792129874229431,
+     -1.5354958772659302,
+     4.0538763999938965,
+     -3.794677734375,
+     -1.4759998321533203
+    ],
+    [
+     1.5326385498046875,
+     -5.820228099822998,
+     2.9093282222747803,
+     -2.941356897354126,
+     -1.2621431350708008
+    ]
+   ]
+  ],
+  "training": {
+   "loss": 0.40739744901657104,
+   "gradients": {
+    "embeddings.patch_embeddings.projection.weight": true,
+    "encoder.layer.0.attention.q_proj.weight": true
+   },
+   "updated": true
+  }
+ }
+}

integrations/transformers/usage.md

@@ -0,0 +1,76 @@
+# DINOv2: image features with Transformers
+
+This page is the Transformers contract of the DINOv2 MIC and also a usage example. [doc-builder](https://github.com/huggingface/doc-builder/blob/main/docs/runnable-code-blocks.md) renders the `runnable` blocks below as documentation, and its pytest plugin runs them as written against `facebook/dinov2-small`. The MIC runner runs the same blocks against the slim fixture and the checkpoint in [`integration.json`](../../integration.json). Lines marked `# doc-builder: hide` only matter there: they swap in the pinned checkpoint or variant and print the result record.
+
+## Extract image features
+
+```py runnable:dinov2
+import torch
+from transformers import AutoImageProcessor, AutoModel, set_seed
+from PIL import Image
+import json  # doc-builder: hide
+import os  # doc-builder: hide
+
+model_id = "facebook/dinov2-small"
+model_id = os.environ.get("MIC_MODEL", model_id)  # doc-builder: hide
+variant = json.loads(os.environ.get("MIC_MODEL_KWARGS", "{}"))  # doc-builder: hide
+device = "cuda" if torch.cuda.is_available() else "cpu"
+set_seed(0)  # seeds torch, NumPy, and Python: loading and any dropout draw random numbers
+
+processor = AutoImageProcessor.from_pretrained(model_id)
+model = AutoModel.from_pretrained(model_id).to(device)
+if variant:  # doc-builder: hide
+    del model  # doc-builder: hide
+    model = AutoModel.from_pretrained(model_id, **variant).to(device)  # doc-builder: hide
+model.eval()
+
+# Two synthetic images: a red square and a blue square, distinct enough that a batch probe is meaningful.
+images = [Image.new("RGB", (224, 224), (200, 30, 30)), Image.new("RGB", (224, 224), (30, 30, 200))]
+inputs = processor(images=images, return_tensors="pt")
+inputs = inputs.to(device, dtype=model.dtype)  # pixel_values are float: cast to the model dtype
+
+with torch.no_grad():
+    outputs = model(**inputs)
+
+cls_features = outputs.last_hidden_state[:, 0, :5].float()
+patch_features = outputs.last_hidden_state[:, 1:4, :5].float()
+print("cls:", cls_features.tolist())
+print("patches:", patch_features.tolist())
+```
+
+## One training step
+
+```py runnable:dinov2:2
+# A small seeded SGD step on the CLS token: the target is the mean of the batch's own CLS features,
+# so the loss is well defined without external labels.
+model.train()
+optimizer = torch.optim.SGD([p for p in model.parameters() if p.requires_grad], lr=0.01)
+
+with torch.no_grad():
+    target = outputs.last_hidden_state[:, 0, :].mean(dim=0, keepdim=True).detach()
+
+loss = torch.nn.functional.mse_loss(model(**inputs).last_hidden_state[:, 0, :], target.expand_as(model(**inputs).last_hidden_state[:, 0, :]))
+loss.backward()
+before = {n: p.detach().cpu().clone() for n, p in model.named_parameters() if p.requires_grad}  # doc-builder: hide
+optimizer.step()
+print(f"loss: {loss.item():.4f}")
+
+if os.environ.get("MIC_RUNNER"):  # doc-builder: hide
+    named = dict(model.named_parameters())
+    checked = ("embeddings.patch_embeddings.projection.weight", "encoder.layer.0.attention.q_proj.weight")
+    result = {
+        "runtime": {
+            "device": model.device.type,
+            "dtype": str(model.dtype).removeprefix("torch."),
+            "attn_implementation": model.config._attn_implementation,
+        },
+        "cls_features": cls_features.tolist(),
+        "patch_features": patch_features.tolist(),
+        "training": {
+            "loss": loss.item(),
+            "gradients": {n: bool(named[n].grad is not None and torch.count_nonzero(named[n].grad) > 0 and torch.isfinite(named[n].grad).all()) for n in checked},
+            "updated": any(not torch.equal(p.detach().cpu(), before[n]) for n, p in model.named_parameters() if n in before),
+        },
+    }
+    print(json.dumps(result))
+```

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3912762cb2c791b01d4b9b5b7cd51c5562cd6a1d3ad9448ff4c1a92efc753ed
+size 907728

preprocessor_config.json

@@ -0,0 +1,27 @@
+{
+  "crop_size": {
+    "height": 224,
+    "width": 224
+  },
+  "do_center_crop": true,
+  "do_convert_rgb": true,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "image_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "image_processor_type": "BitImageProcessor",
+  "image_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "resample": 3,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "shortest_edge": 256
+  }
+}