{ "strict_load": true, "source_tensor_bytes_equal": true, "checks": [ { "name": "vision_official_LFS_SHA256", "passed": true }, { "name": "projector_official_LFS_SHA256", "passed": true }, { "name": "native_vision_strict_load", "passed": true }, { "name": "native_projector_strict_load", "passed": true }, { "name": "all_1111_native_state_values", "passed": true }, { "name": "four_depths_shared_codebook", "passed": true }, { "name": "16384_codes_plus_padding_row", "passed": true }, { "name": "original_bfloat16_upsample_parity", "passed": true }, { "name": "native_encode_grid", "passed": true }, { "name": "code_ID_bounds_exclude_padding", "passed": true }, { "name": "straight_through_lookup_bound", "passed": true, "max_abs": 2.384185791015625e-07 }, { "name": "depth_lookup_sum_exact", "passed": true }, { "name": "native_decode_lookup_exact", "passed": true }, { "name": "finite_256px_RGB", "passed": true, "mse": 0.0035370811820030212 }, { "name": "language_projector_geometry", "passed": true }, { "name": "native_depth_teacher_forcing", "passed": true }, { "name": "native_four_depth_generation", "passed": true }, { "name": "generated_feature_lookup_exact", "passed": true }, { "name": "retained_trained_text_tower", "passed": true, "shape": [ 1, 1024 ] } ], "smoke": { "passed": true, "device": "cpu", "dtype": "float32" }, "elapsed_seconds": 175.93401333299698, "limitations": [ "Synthetic fixed 256px native image fixture and short depth-transformer sequences, not image-quality or VQA benchmarks.", "The separate Llama backbone and language tokenizer are excluded. Depth generation tests use visual-projector fixtures at that external language-state boundary.", "Four residual depths share one learned codebook; the 16385th stored row is padding, not a valid generated visual ID." ] }