| { | |
| "strict_load": true, | |
| "source_tensor_bytes_equal": true, | |
| "checks": [ | |
| { | |
| "name": "native_strict_load", | |
| "passed": true, | |
| "tensors": 504 | |
| }, | |
| { | |
| "name": "siglip_features", | |
| "passed": true, | |
| "shape": [ | |
| 1, | |
| 16, | |
| 1152 | |
| ] | |
| }, | |
| { | |
| "name": "compressed_64", | |
| "passed": true, | |
| "shape": [ | |
| 1, | |
| 16, | |
| 64 | |
| ] | |
| }, | |
| { | |
| "name": "decompressed_1152", | |
| "passed": true, | |
| "shape": [ | |
| 1, | |
| 16, | |
| 1152 | |
| ] | |
| }, | |
| { | |
| "name": "compression_shapes", | |
| "passed": true | |
| }, | |
| { | |
| "name": "latent_to_backbone", | |
| "passed": true, | |
| "shape": [ | |
| 1, | |
| 16, | |
| 3584 | |
| ] | |
| }, | |
| { | |
| "name": "backbone_to_latent", | |
| "passed": true, | |
| "shape": [ | |
| 1, | |
| 16, | |
| 64 | |
| ] | |
| }, | |
| { | |
| "name": "flow_shapes", | |
| "passed": true | |
| }, | |
| { | |
| "name": "explicit_1216_aligner", | |
| "passed": true, | |
| "shape": [ | |
| 1, | |
| 16, | |
| 3584 | |
| ] | |
| }, | |
| { | |
| "name": "upstream_1152_vs_1216_guard", | |
| "passed": true | |
| } | |
| ], | |
| "smoke": { | |
| "passed": true, | |
| "device": "cpu", | |
| "dtype": "float32" | |
| }, | |
| "elapsed_seconds": 8.918606996536255, | |
| "limitations": [ | |
| "Component tests only; full multimodal backbone and generation quality not tested.", | |
| "Published native direct SigLIP\u2192aligner helper has 1152 vs 1216 dimension mismatch; wrapper refuses implicit adaptation. Pixel reconstruction requires separate decoder and VAE." | |
| ] | |
| } | |