Download validation.json from gatilin/hy4-flash-vision-encoder: direct link, hf CLI and curl.
- Browser
- Download file 2.97 kB
-
https://huggingface.co/gatilin/hy4-flash-vision-encoder/resolve/main/validation.json
- Command line
-
hf download hf://gatilin/hy4-flash-vision-encoder/validation.json
-
curl -L -o validation.json https://huggingface.co/gatilin/hy4-flash-vision-encoder/resolve/main/validation.json
2.97 kB
| { | |
| "passed": true, | |
| "source_revision": "d88c9dfd96ba4f2e6b92adba0b8e8f1c8696f959", | |
| "model_sha256": "5fdf462167ec22a48a231b79196bd196d96c6838a108e80716369690f95980bc", | |
| "num_tensors": 441, | |
| "strict_load": true, | |
| "native_model_layers": 27, | |
| "dtype_for_test": "float32 CPU; artifact retains source dtype", | |
| "runtime": "Pinned HYV4FlashVisionTransformer model/config/image processor files over installed Transformers 5.21.1 runtime; full pinned PR environment not independently installed in this validation", | |
| "weight_key_mapping": "matches vision renames in PR conversion_mapping.py; original file keys unchanged", | |
| "input_preprocessing": "upstream HYV4FlashImageProcessor from source processor_config.json", | |
| "images": [ | |
| { | |
| "image": "000000000009.png", | |
| "sha256": "ec671f039c1e2b16ebad3cef3412bc61fdf4286a9b8b98c7c59f8bceb6ce5fbb", | |
| "grid_thw": [ | |
| 1, | |
| 14, | |
| 14 | |
| ], | |
| "patch_features_shape": [ | |
| 1, | |
| 196, | |
| 1152 | |
| ], | |
| "projected_features_shape": [ | |
| 1, | |
| 56, | |
| 3072 | |
| ], | |
| "output_rms": 0.335646390914917, | |
| "repeat_max_abs_error": 0.0 | |
| }, | |
| { | |
| "image": "000000000030.png", | |
| "sha256": "1fbe2305d87d89ebd049c9d37e6795b76c10d7b0de5fab8c0cb58313cd1e4e11", | |
| "grid_thw": [ | |
| 1, | |
| 14, | |
| 14 | |
| ], | |
| "patch_features_shape": [ | |
| 1, | |
| 196, | |
| 1152 | |
| ], | |
| "projected_features_shape": [ | |
| 1, | |
| 56, | |
| 3072 | |
| ], | |
| "output_rms": 0.33717289566993713, | |
| "repeat_max_abs_error": 0.0 | |
| } | |
| ], | |
| "different_image_mean_abs_difference": 0.2504355013370514, | |
| "notes": "Two existing real COCO128 images, deterministic repeat and input sensitivity checks; not a downstream task accuracy or complete multimodal pipeline test.", | |
| "seconds": 7.7, | |
| "attention_check": { | |
| "image": "000000000009.png", | |
| "layers_zero_based": [ | |
| 0, | |
| 13, | |
| 26 | |
| ], | |
| "method": "full model native SDPA; explicitly reconstructed softmax(QK^T)\u00d7V compared at selected layers; one real image; CPU float32", | |
| "metrics": { | |
| "0": { | |
| "query_normalized_entropy": 0.8158265352249146, | |
| "explicit_vs_sdpa_max_abs_error": 1.6689300537109375e-06, | |
| "max_attention_row_sum_error": 7.152557373046875e-07, | |
| "heads": 16, | |
| "grid": [ | |
| 14, | |
| 14 | |
| ] | |
| }, | |
| "13": { | |
| "query_normalized_entropy": 0.7042220234870911, | |
| "explicit_vs_sdpa_max_abs_error": 2.0265579223632812e-06, | |
| "max_attention_row_sum_error": 4.76837158203125e-07, | |
| "heads": 16, | |
| "grid": [ | |
| 14, | |
| 14 | |
| ] | |
| }, | |
| "26": { | |
| "query_normalized_entropy": 0.6273640990257263, | |
| "explicit_vs_sdpa_max_abs_error": 4.76837158203125e-06, | |
| "max_attention_row_sum_error": 4.76837158203125e-07, | |
| "heads": 16, | |
| "grid": [ | |
| 14, | |
| 14 | |
| ] | |
| } | |
| } | |
| } | |
| } | |