| { |
| "strict_load": true, |
| "source_tensor_bytes_equal": true, |
| "checks": [ |
| { |
| "name": "1335_source_tensor_payloads", |
| "passed": true |
| }, |
| { |
| "name": "native_strict_load", |
| "passed": true |
| }, |
| { |
| "name": "eight_4096_by_8_codebooks", |
| "passed": true |
| }, |
| { |
| "name": "token_geometry", |
| "passed": true, |
| "shape": [ |
| 1, |
| 8, |
| 256 |
| ] |
| }, |
| { |
| "name": "per_codebook_ID_bounds", |
| "passed": true |
| }, |
| { |
| "name": "normalized_lookup_exact", |
| "passed": true |
| }, |
| { |
| "name": "decode_ID_lookup_exact", |
| "passed": true |
| }, |
| { |
| "name": "finite_RGB_reconstruction", |
| "passed": true, |
| "mse": 0.0007349226507358253 |
| }, |
| { |
| "name": "native_quantizer_reconstruction_bound", |
| "passed": true, |
| "max_abs": 1.0728836059570312e-06 |
| }, |
| { |
| "name": "semantic_image_features", |
| "passed": true |
| }, |
| { |
| "name": "semantic_text_features", |
| "passed": true |
| } |
| ], |
| "smoke": { |
| "passed": true, |
| "device": "cpu", |
| "dtype": "float32", |
| "resolution": [ |
| 256, |
| 256 |
| ] |
| }, |
| "environment": { |
| "torch": "2.14.1", |
| "torchvision": "0.29.1", |
| "timm": "1.0.8", |
| "safetensors": "0.8.0", |
| "typed-argument-parser": "1.12.0", |
| "pytz": "2026.5" |
| }, |
| "text_implementation": "Unchanged official project-vendored native.open_clip.transformer.TextTransformer, same configured default constructor branch. External OpenCLIP is not imported.", |
| "elapsed_seconds": 90.44274949999817, |
| "limitations": [ |
| "Synthetic native codec and semantic feature fixtures; no downstream image quality, retrieval, VQA or text-to-image benchmark.", |
| "This is the complete trained tokenizer mapping including its semantic text tower; no language generator is included.", |
| "The eight codebooks each have 4096 entries of width 8. The sum of entries is 32768; IDs are local to each codebook." |
| ] |
| } |
|
|
