| { |
| "strict_load": true, |
| "source_tensor_bytes_equal": true, |
| "checks": [ |
| { |
| "name": "strict_base_codec_load", |
| "passed": true, |
| "tensors": 1229 |
| }, |
| { |
| "name": "native_shared_codebook_object", |
| "passed": true |
| }, |
| { |
| "name": "shared_codebook_storage", |
| "passed": true |
| }, |
| { |
| "name": "patch_layout_inverse_112x112", |
| "passed": true, |
| "patches": [ |
| 64, |
| 1176 |
| ] |
| }, |
| { |
| "name": "rvq_ids_112x112", |
| "passed": true, |
| "shape": [ |
| 16, |
| 8 |
| ], |
| "minimum": 33, |
| "maximum": 16208 |
| }, |
| { |
| "name": "lookup_sum_112x112", |
| "passed": true |
| }, |
| { |
| "name": "base_reconstruction_112x112", |
| "passed": true, |
| "shape": [ |
| 3, |
| 112, |
| 112 |
| ], |
| "mse": 0.19658847153186798 |
| }, |
| { |
| "name": "patch_layout_inverse_56x84", |
| "passed": true, |
| "patches": [ |
| 24, |
| 1176 |
| ] |
| }, |
| { |
| "name": "rvq_ids_56x84", |
| "passed": true, |
| "shape": [ |
| 6, |
| 8 |
| ], |
| "minimum": 33, |
| "maximum": 16305 |
| }, |
| { |
| "name": "lookup_sum_56x84", |
| "passed": true |
| }, |
| { |
| "name": "base_reconstruction_56x84", |
| "passed": true, |
| "shape": [ |
| 3, |
| 56, |
| 84 |
| ], |
| "mse": 0.4664306044578552 |
| }, |
| { |
| "name": "language_visual_embedding_interface", |
| "passed": true |
| }, |
| { |
| "name": "refiner_vae_reconstruction", |
| "passed": true, |
| "latent_shape": [ |
| 1, |
| 16, |
| 8, |
| 8 |
| ] |
| }, |
| { |
| "name": "refiner_condition_projection", |
| "passed": true, |
| "shape": [ |
| 4, |
| 2048 |
| ] |
| }, |
| { |
| "name": "source_shared_aliases_exact", |
| "passed": true |
| }, |
| { |
| "name": "base_source_bytes_exact", |
| "passed": true |
| }, |
| { |
| "name": "refiner_strict_full_flow_refiner", |
| "passed": true, |
| "tensors": 828, |
| "transformer_tensors": 582 |
| }, |
| { |
| "name": "refiner_native_bfloat16_weights", |
| "passed": true |
| }, |
| { |
| "name": "refiner_native_refiner_conditioning", |
| "passed": true |
| }, |
| { |
| "name": "refiner_conditioned_flow_forward", |
| "passed": true, |
| "shape": [ |
| 1, |
| 16, |
| 4, |
| 4 |
| ], |
| "dtype": "torch.bfloat16" |
| }, |
| { |
| "name": "refiner_single_euler_update", |
| "passed": true, |
| "mean_abs_update": 0.07963985204696655 |
| }, |
| { |
| "name": "refiner_condition_affects_flow", |
| "passed": true, |
| "max_abs_difference": 0.99609375 |
| }, |
| { |
| "name": "refiner_refiner_source_tensor_bytes_exact", |
| "passed": true |
| } |
| ], |
| "smoke": { |
| "passed": true, |
| "device": "cpu", |
| "visual_dtype": "bfloat16", |
| "base_decoder_dtype": "float32", |
| "quantizer_dtype": "float32" |
| }, |
| "elapsed_seconds": 11.928704977035522, |
| "limitations": [ |
| "No full LongCat text/audio backbone or VQA/T2I benchmark.", |
| "CPU compatibility uses a vendored Qwen2.5-VL window-attention implementation and preserves native decoder behavior.", |
| "Full flow-refiner weights and tiny-latent conditioned forward are validated separately; no full-resolution quality benchmark.", |
| "Combined encoder+full-refiner CPU integration run was stopped due to host memory/swap pressure; no full-pipeline pass is claimed." |
| ], |
| "full_flow_refiner_validation": "refiner_validation.json", |
| "runtime_versions": { |
| "torch": "2.14.1", |
| "torchvision": "0.29.1", |
| "transformers": "5.20.1", |
| "diffusers": "0.41.0", |
| "safetensors": "0.8.0", |
| "accelerate": "1.15.0", |
| "numpy": "2.5.4", |
| "einops": "0.8.2" |
| } |
| } |
|
|
