{ "schema_version": "vision_question_clusters_resolved_v1", "dataset_family": "conditional_image_generation", "selection_metric": "test_energy_score", "profile_prefix": "conditional_image_generation_template_", "source_path": "profiles/image-generation-ranking-supplement-v1.yaml", "source_sha256": "b8eb9c87ec5dfa954a2c36762e28befd8cc02634df24c490ebd520e11af24000", "base_profile": "profiles/image-generation-ranking-base-v1.yaml", "default_num_choices": 3, "discovery_training_seeds": [ 40, 41, 42, 43 ], "confirmation_training_seeds": [ 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 254, 255 ], "selection_contract": { "discovery_split": "validation", "confirmation_split": "test", "minimum_mean_gap": 0.003, "require_nonoverlap": false, "multiplicity": "Independent supplement family: Holm across all six frozen adjacent hypotheses, including rejected instances.", "claim_scope": "Fixed synthetic image distribution and fixed independent evaluation sample streams; training-seed robustness only.", "difficulty_policy": "Evaluate every admitted supplemental triple and report all rejects separately from the original fixed batch; no answer-based selection or replacement." }, "statistics": { "discovery": { "n_seeds": 4, "split": "validation", "method": "direction_estimation", "ordering": "mean_validation_energy_score", "admission_gate": false, "purpose": "Freeze adjacent ranking directions before independent confirmation; no significance claim from four discovery seeds." }, "confirmation": { "n_seeds": 16, "split": "test", "test": "one_sided_paired_sign_test", "alpha": 0.05, "correction": "holm", "minimum_mean_gap": 0.003 }, "estimand": "Training-seed paired direction stability conditional on one fixed dataset and fixed evaluation-noise/reference stream, with observed mean energy-score gap; not a test of expected-score mean difference or a natural-image population claim." }, "protocol_config": { "schema_version": "vision_ranking_protocol_v1", "discovery": { "seeds": [ 40, 41, 42, 43 ], "metric": "final_validation_metric", "order": "ascending mean; ties use candidate_id", "purpose": "Freeze a direction for every template instance before independent confirmation. Weak discovery separation is retained and reported, not an automatic rejection." }, "confirmation": { "seeds": [ 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 254, 255 ], "metric": "final_test_energy_score", "higher_is_better": false, "test": "one-sided exact paired binomial sign test; ties count as nonpositive", "null": "P(oriented paired improvement > 0) <= 0.5 over training seed pairs, conditional on fixed materialized data and training protocol", "multiplicity": "Holm across every frozen adjacent comparison, including failed or missing comparisons with p=1", "alpha": 0.05, "minimum_adjacent_mean_gap": 0.003, "require_nonoverlapping_seed_ranges": false, "bootstrap": { "samples": 10000, "confidence": 0.95, "seed": 73013, "purpose": "Descriptive paired mean-gap interval only; no admission decision uses this interval" }, "failure_policy": "Any absent, duplicate, nonfinite or failed seed rejects that instance. No seed deletion, replacement, added seeds, reordered choices, or extra metric selection after confirmation." }, "admission": "Every adjacent comparison in the discovery-frozen order must pass Holm and the minimum mean-gap gate; both gates use oriented improvement in the frozen direction.", "independence": "Confirmation seeds must be disjoint from discovery seeds. Freeze full membership, candidate order, artifact hashes, metrics, gates, and hypothesis family before starting confirmation. Never select questions using tested models responses.", "comparison_contract": { "same_data": "Every choice shares one dataset instance and identical split file hashes.", "budgets": "Equal total_samples_seen unless the cluster explicitly intervenes on budget; always expose steps, batch size and total_samples_seen per choice. Equal samples with different batches does not imply equal optimizer updates.", "parameters": "Record measured parameter count. Parameter matching is a separate intervention, never inferred from width matching.", "streams": "Use paired training seeds and shared batch index stream when batch sizes agree; seed number alone does not establish matched initialization across architectures.", "interventions": "Record every changed spec field and final full-code hashes. Single-axis interpretation requires other fields fixed; multi-axis choices establish joint effects only.", "devices": [ "360-1", "ophis" ] }, "evaluation": { "models": [ "claude-opus-5-5", "gpt-6.1-sol", "claude-haiku-5-5", "gpt-6-luna" ], "reasoning_effort": "medium", "max_tokens": 8192, "tools": false, "calls_per_question": 1, "system": "Answer the supplied question. Tools are unavailable.", "response": "Public explanation followed by exact full ordering. Audit explanation independently of correctness.", "metrics": [ "exact full order", "pairwise concordance", "valid response rate", "answer-explanation consistency" ], "failure_policy": "Only transport failures without completed answer may retry. Preserve all attempts, identity, prompt hashes and public text. Never replace a completed wrong answer.", "baseline": "For k choices, uniform complete-ranking probability is 1/k!; pairwise chance is 1/2.", "prompt": "Use canonical renderer with complete shared base and independent diffs; byte-exact reconstruction checked. Fallback to full code when diff readability degrades." }, "difficulty": { "surface_heuristics": "Freeze interpretable surface-only rules before LLM responses; report abstentions and tie coverage. Do not use confirmation CE or model errors to tune rules.", "coverage": "Report every frozen template instance as accepted/rejected/failed and why. Do not only publish difficult successes.", "dependence": "Report results by template and teacher draw. Repeated teacher or dataset conditions are correlated; no independent-question binomial intervals or general model ranking claims.", "selection": "Do not use model mistakes to admit or drop items. Any later difficult subset is explicitly exploratory and kept separate from full predeclared batch." }, "causal_scope": "Confirmed performance differences identify the total effect of the exposed intervention at fixed data and training horizon. A mechanism is supported only to the extent that separately logged diagnostic/control evidence rules out alternatives. Image results concern the stated synthetic image distribution only.", "statistics": { "discovery": { "n_seeds": 4, "split": "validation", "method": "direction_estimation", "ordering": "mean_validation_energy_score", "admission_gate": false, "purpose": "Freeze adjacent ranking directions before independent confirmation; no significance claim from four discovery seeds." }, "confirmation": { "n_seeds": 16, "split": "test", "test": "one_sided_paired_sign_test", "alpha": 0.05, "correction": "holm", "minimum_mean_gap": 0.003 }, "estimand": "Training-seed paired direction stability conditional on one fixed dataset and fixed evaluation-noise/reference stream, with observed mean energy-score gap; not a test of expected-score mean difference or a natural-image population claim." }, "answer_format": "best_to_worst_v1" }, "surface_heuristics": [ { "id": "more_parameters", "field": "parameter_count", "higher_is_better": true }, { "id": "more_training_steps", "field": "training.training_steps", "higher_is_better": true }, { "id": "more_sampling_steps", "field": "training.sampling_steps", "higher_is_better": true } ], "prompt_contract": { "representation": "shared_complete_base_plus_independent_diff", "answer_format": "best_to_worst_v1", "analyst_only_fields": [ "hypothesis", "shortcut_checks", "explanation_target", "admission" ] }, "worker_sources": [ "tools/ntp_h200_worker.py", "tools/ntp_anchor_worker.py", "tools/vision_h200_worker.py" ], "counts": { "templates": 1, "instances": 3, "dataset_nodes": 3, "candidate_nodes": 12, "distinct_models": 4 }, "dataset_nodes": [ { "node_id": "dataset_f44ec2307822c0cf", "dataset_config": { "size": 16, "noise_steps": 32, "condition_dim": 6, "brightness_min": 0.65, "brightness_max": 1.0, "texture_strength": 0.12, "radius_min": 0.25, "radius_max": 0.42, "position_extent": 0.45, "reference_samples": 4, "train_size": 1024, "validation_size": 64, "test_size": 64, "ood_size": 32, "evaluation_shift": {}, "ood_shift": { "position_extent": 0.55 } }, "instance_seed": 852001, "context_ref": "textured" }, { "node_id": "dataset_fd33629bb65f3809", "dataset_config": { "size": 16, "noise_steps": 32, "condition_dim": 6, "brightness_min": 0.65, "brightness_max": 1.0, "texture_strength": 0.12, "radius_min": 0.25, "radius_max": 0.42, "position_extent": 0.45, "reference_samples": 4, "train_size": 1024, "validation_size": 64, "test_size": 64, "ood_size": 32, "evaluation_shift": {}, "ood_shift": { "position_extent": 0.55 } }, "instance_seed": 852002, "context_ref": "textured" }, { "node_id": "dataset_d8ad2ee2cf7b306a", "dataset_config": { "size": 16, "noise_steps": 32, "condition_dim": 6, "brightness_min": 0.65, "brightness_max": 1.0, "texture_strength": 0.12, "radius_min": 0.25, "radius_max": 0.42, "position_extent": 0.45, "reference_samples": 4, "train_size": 1024, "validation_size": 64, "test_size": 64, "ood_size": 32, "evaluation_shift": {}, "ood_shift": { "position_extent": 0.55 } }, "instance_seed": 852003, "context_ref": "textured" } ], "candidate_nodes": [ { "node_id": "candidate_ae7b6a4c1c102ac65308", "dataset_node_id": "dataset_f44ec2307822c0cf", "model": { "type": "image_generation_denoiser", "mode": "conv", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 19489, "model_code_sha256": "548e5eb2edfaab77381773b5c3a085cafe768785e53bd9377556e95e2cf4161d" }, { "node_id": "candidate_e9681c5d795db8946e1c", "dataset_node_id": "dataset_f44ec2307822c0cf", "model": { "type": "image_generation_denoiser", "mode": "unet", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "skip": true, "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 35089, "model_code_sha256": "5eb9b3c4a5893cf189f5b3571dd497ba383e394e0d9c190e56276b84a9d36229" }, { "node_id": "candidate_b83c1445855d4b8e3b87", "dataset_node_id": "dataset_f44ec2307822c0cf", "model": { "type": "image_generation_denoiser", "mode": "mlp", "width": 24, "depth": 3, "activation": "silu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 461344, "model_code_sha256": "31d958601950f09831a638b54b51b3db7d388a4599ccfe063782adc4637d4b2c" }, { "node_id": "candidate_c154ceb7fdd2523d4ae3", "dataset_node_id": "dataset_f44ec2307822c0cf", "model": { "type": "image_generation_denoiser", "mode": "patch_transformer", "width": 24, "depth": 3, "activation": "gelu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "patch_size": 2, "max_parameters": 2000000, "num_heads": 4, "ff_multiplier": 2, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 17908, "model_code_sha256": "6a1a3aae0326db52c89fa6c204e71cb771024431856ade894f8a7732e5886275" }, { "node_id": "candidate_8bceb0229ca0a46d125e", "dataset_node_id": "dataset_fd33629bb65f3809", "model": { "type": "image_generation_denoiser", "mode": "conv", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 19489, "model_code_sha256": "548e5eb2edfaab77381773b5c3a085cafe768785e53bd9377556e95e2cf4161d" }, { "node_id": "candidate_a2520835f9adbb7c0b84", "dataset_node_id": "dataset_fd33629bb65f3809", "model": { "type": "image_generation_denoiser", "mode": "unet", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "skip": true, "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 35089, "model_code_sha256": "5eb9b3c4a5893cf189f5b3571dd497ba383e394e0d9c190e56276b84a9d36229" }, { "node_id": "candidate_fa4a1e6560135ec374bf", "dataset_node_id": "dataset_fd33629bb65f3809", "model": { "type": "image_generation_denoiser", "mode": "mlp", "width": 24, "depth": 3, "activation": "silu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 461344, "model_code_sha256": "31d958601950f09831a638b54b51b3db7d388a4599ccfe063782adc4637d4b2c" }, { "node_id": "candidate_1f64af8282f367bccdb9", "dataset_node_id": "dataset_fd33629bb65f3809", "model": { "type": "image_generation_denoiser", "mode": "patch_transformer", "width": 24, "depth": 3, "activation": "gelu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "patch_size": 2, "max_parameters": 2000000, "num_heads": 4, "ff_multiplier": 2, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 17908, "model_code_sha256": "6a1a3aae0326db52c89fa6c204e71cb771024431856ade894f8a7732e5886275" }, { "node_id": "candidate_804cc8045512fe7cbec6", "dataset_node_id": "dataset_d8ad2ee2cf7b306a", "model": { "type": "image_generation_denoiser", "mode": "conv", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 19489, "model_code_sha256": "548e5eb2edfaab77381773b5c3a085cafe768785e53bd9377556e95e2cf4161d" }, { "node_id": "candidate_d4f3e193671168b46ced", "dataset_node_id": "dataset_d8ad2ee2cf7b306a", "model": { "type": "image_generation_denoiser", "mode": "unet", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "skip": true, "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 35089, "model_code_sha256": "5eb9b3c4a5893cf189f5b3571dd497ba383e394e0d9c190e56276b84a9d36229" }, { "node_id": "candidate_eaae951afaff90809747", "dataset_node_id": "dataset_d8ad2ee2cf7b306a", "model": { "type": "image_generation_denoiser", "mode": "mlp", "width": 24, "depth": 3, "activation": "silu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 461344, "model_code_sha256": "31d958601950f09831a638b54b51b3db7d388a4599ccfe063782adc4637d4b2c" }, { "node_id": "candidate_72f3500e4f5cd3e445c8", "dataset_node_id": "dataset_d8ad2ee2cf7b306a", "model": { "type": "image_generation_denoiser", "mode": "patch_transformer", "width": 24, "depth": 3, "activation": "gelu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "patch_size": 2, "max_parameters": 2000000, "num_heads": 4, "ff_multiplier": 2, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 17908, "model_code_sha256": "6a1a3aae0326db52c89fa6c204e71cb771024431856ade894f8a7732e5886275" } ], "model_validation_receipts": [ { "model_sha256": "dd4129b6d4781c142acdff5c7ff0a35b45aa0ca3d8581b9403809cf9f77f8ba3", "parameters": 19489, "code_sha256": "548e5eb2edfaab77381773b5c3a085cafe768785e53bd9377556e95e2cf4161d", "smoke": "standalone_render_forward_backward_and_canonical_parity", "output_shape": [ 2, 1, 16, 16 ], "canonical_max_abs_error": 0.0 }, { "model_sha256": "11bed1538d190d71ba4e61826c8b3adccb108da86c4c576eca4c5625546ca065", "parameters": 35089, "code_sha256": "5eb9b3c4a5893cf189f5b3571dd497ba383e394e0d9c190e56276b84a9d36229", "smoke": "standalone_render_forward_backward_and_canonical_parity", "output_shape": [ 2, 1, 16, 16 ], "canonical_max_abs_error": 0.0 }, { "model_sha256": "515429b2d9dde44af216dac7468298e06460cff8218f6cc1fad00160dd5c3d9b", "parameters": 461344, "code_sha256": "31d958601950f09831a638b54b51b3db7d388a4599ccfe063782adc4637d4b2c", "smoke": "standalone_render_forward_backward_and_canonical_parity", "output_shape": [ 2, 1, 16, 16 ], "canonical_max_abs_error": 0.0 }, { "model_sha256": "43fbaf15c628dfc6fd6110e83e18afe0ccb64077a14bd8feaf15649034279aa0", "parameters": 17908, "code_sha256": "6a1a3aae0326db52c89fa6c204e71cb771024431856ade894f8a7732e5886275", "smoke": "standalone_render_forward_backward_and_canonical_parity", "output_shape": [ 2, 1, 16, 16 ], "canonical_max_abs_error": 0.0 } ], "templates": [ { "id": "operator_native_matched_s852001", "category": "operator", "context_ref": "textured", "budget_semantics": "matched_steps", "hypothesis": "Four native denoisers share absolute coordinate input, data, optimizer, target and update budget. Their operator, parameter count, normalization and connectivity differ as shown; finite-budget ordering requires actual sampled image quality.", "varying_axes": [ "model.mode" ], "controls": [ "All four available native operator cells are retained: convolution, one-level U-Net, flat MLP and patch Transformer.", "Every cell uses absolute coordinates, concat conditions, sinusoidal time, x0 target, 3000 updates, batch16, AdamW1e-3, the same materialized data and fixed evaluation noise/reference draws.", "This is a bundled native-recipe contrast: parameter count, activation, normalization and connectivity are explicitly shown; do not attribute the total difference exclusively to operator name.", "Primary choices were selected from original discovery validation only; the diagnostic Transformer is not promoted or substituted after new confirmation." ], "shortcut_checks": [ "always prefer U-Net for diffusion", "always prefer more parameters", "always prefer globally mixing operator" ], "explanation_target": "Explain both adjacent comparisons using exact native code and the fixed finite-budget task; use all four local cells and image/geometry/diversity diagnostics, while acknowledging unresolved parameter-count and optimization attribution.", "admission": "Freeze the new four-seed validation order, then independently test all six adjacent hypotheses from three new dataset instances using the unchanged16-seed sign-Holm rule and gap0.003. Keep rejected instances; no replacement.", "template_id": "operator_native_matched", "instance_seed": 852001, "dataset_node_id": "dataset_f44ec2307822c0cf", "dataset_config": { "size": 16, "noise_steps": 32, "condition_dim": 6, "brightness_min": 0.65, "brightness_max": 1.0, "texture_strength": 0.12, "radius_min": 0.25, "radius_max": 0.42, "position_extent": 0.45, "reference_samples": 4, "train_size": 1024, "validation_size": 64, "test_size": 64, "ood_size": 32, "evaluation_shift": {}, "ood_shift": { "position_extent": 0.55 } }, "comparison": { "budget_semantics": "matched_steps", "changed_fields": [ "model.block", "model.kernel", "model.mode", "model.norm", "model.residual", "model.skip" ], "undeclared_fields": [ "model.block", "model.kernel", "model.norm", "model.residual", "model.skip" ], "comparison_class": "bundled_recipe", "field_values": { "conv": { "model.block": "dense", "model.kernel": 3, "model.mode": "conv", "model.norm": "group", "model.residual": true, "model.skip": "" }, "unet": { "model.block": "dense", "model.kernel": 3, "model.mode": "unet", "model.norm": "group", "model.residual": true, "model.skip": true }, "mlp": { "model.block": "", "model.kernel": "", "model.mode": "mlp", "model.norm": "", "model.residual": "", "model.skip": "" } }, "actual_changed_fields": [ "model.block", "model.kernel", "model.mode", "model.norm", "model.residual", "model.skip" ] }, "choices": [ { "id": "conv", "role": "primary", "model": { "type": "image_generation_denoiser", "mode": "conv", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 19489, "model_code_sha256": "548e5eb2edfaab77381773b5c3a085cafe768785e53bd9377556e95e2cf4161d", "node_id": "candidate_ae7b6a4c1c102ac65308" }, { "id": "unet", "role": "primary", "model": { "type": "image_generation_denoiser", "mode": "unet", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "skip": true, "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 35089, "model_code_sha256": "5eb9b3c4a5893cf189f5b3571dd497ba383e394e0d9c190e56276b84a9d36229", "node_id": "candidate_e9681c5d795db8946e1c" }, { "id": "mlp", "role": "primary", "model": { "type": "image_generation_denoiser", "mode": "mlp", "width": 24, "depth": 3, "activation": "silu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 461344, "model_code_sha256": "31d958601950f09831a638b54b51b3db7d388a4599ccfe063782adc4637d4b2c", "node_id": "candidate_b83c1445855d4b8e3b87" } ], "diagnostics": [ { "id": "transformer", "role": "diagnostic", "model": { "type": "image_generation_denoiser", "mode": "patch_transformer", "width": 24, "depth": 3, "activation": "gelu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "patch_size": 2, "max_parameters": 2000000, "num_heads": 4, "ff_multiplier": 2, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 17908, "model_code_sha256": "6a1a3aae0326db52c89fa6c204e71cb771024431856ade894f8a7732e5886275", "node_id": "candidate_c154ceb7fdd2523d4ae3" } ] }, { "id": "operator_native_matched_s852002", "category": "operator", "context_ref": "textured", "budget_semantics": "matched_steps", "hypothesis": "Four native denoisers share absolute coordinate input, data, optimizer, target and update budget. Their operator, parameter count, normalization and connectivity differ as shown; finite-budget ordering requires actual sampled image quality.", "varying_axes": [ "model.mode" ], "controls": [ "All four available native operator cells are retained: convolution, one-level U-Net, flat MLP and patch Transformer.", "Every cell uses absolute coordinates, concat conditions, sinusoidal time, x0 target, 3000 updates, batch16, AdamW1e-3, the same materialized data and fixed evaluation noise/reference draws.", "This is a bundled native-recipe contrast: parameter count, activation, normalization and connectivity are explicitly shown; do not attribute the total difference exclusively to operator name.", "Primary choices were selected from original discovery validation only; the diagnostic Transformer is not promoted or substituted after new confirmation." ], "shortcut_checks": [ "always prefer U-Net for diffusion", "always prefer more parameters", "always prefer globally mixing operator" ], "explanation_target": "Explain both adjacent comparisons using exact native code and the fixed finite-budget task; use all four local cells and image/geometry/diversity diagnostics, while acknowledging unresolved parameter-count and optimization attribution.", "admission": "Freeze the new four-seed validation order, then independently test all six adjacent hypotheses from three new dataset instances using the unchanged16-seed sign-Holm rule and gap0.003. Keep rejected instances; no replacement.", "template_id": "operator_native_matched", "instance_seed": 852002, "dataset_node_id": "dataset_fd33629bb65f3809", "dataset_config": { "size": 16, "noise_steps": 32, "condition_dim": 6, "brightness_min": 0.65, "brightness_max": 1.0, "texture_strength": 0.12, "radius_min": 0.25, "radius_max": 0.42, "position_extent": 0.45, "reference_samples": 4, "train_size": 1024, "validation_size": 64, "test_size": 64, "ood_size": 32, "evaluation_shift": {}, "ood_shift": { "position_extent": 0.55 } }, "comparison": { "budget_semantics": "matched_steps", "changed_fields": [ "model.block", "model.kernel", "model.mode", "model.norm", "model.residual", "model.skip" ], "undeclared_fields": [ "model.block", "model.kernel", "model.norm", "model.residual", "model.skip" ], "comparison_class": "bundled_recipe", "field_values": { "conv": { "model.block": "dense", "model.kernel": 3, "model.mode": "conv", "model.norm": "group", "model.residual": true, "model.skip": "" }, "unet": { "model.block": "dense", "model.kernel": 3, "model.mode": "unet", "model.norm": "group", "model.residual": true, "model.skip": true }, "mlp": { "model.block": "", "model.kernel": "", "model.mode": "mlp", "model.norm": "", "model.residual": "", "model.skip": "" } }, "actual_changed_fields": [ "model.block", "model.kernel", "model.mode", "model.norm", "model.residual", "model.skip" ] }, "choices": [ { "id": "conv", "role": "primary", "model": { "type": "image_generation_denoiser", "mode": "conv", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 19489, "model_code_sha256": "548e5eb2edfaab77381773b5c3a085cafe768785e53bd9377556e95e2cf4161d", "node_id": "candidate_8bceb0229ca0a46d125e" }, { "id": "unet", "role": "primary", "model": { "type": "image_generation_denoiser", "mode": "unet", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "skip": true, "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 35089, "model_code_sha256": "5eb9b3c4a5893cf189f5b3571dd497ba383e394e0d9c190e56276b84a9d36229", "node_id": "candidate_a2520835f9adbb7c0b84" }, { "id": "mlp", "role": "primary", "model": { "type": "image_generation_denoiser", "mode": "mlp", "width": 24, "depth": 3, "activation": "silu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 461344, "model_code_sha256": "31d958601950f09831a638b54b51b3db7d388a4599ccfe063782adc4637d4b2c", "node_id": "candidate_fa4a1e6560135ec374bf" } ], "diagnostics": [ { "id": "transformer", "role": "diagnostic", "model": { "type": "image_generation_denoiser", "mode": "patch_transformer", "width": 24, "depth": 3, "activation": "gelu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "patch_size": 2, "max_parameters": 2000000, "num_heads": 4, "ff_multiplier": 2, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 17908, "model_code_sha256": "6a1a3aae0326db52c89fa6c204e71cb771024431856ade894f8a7732e5886275", "node_id": "candidate_1f64af8282f367bccdb9" } ] }, { "id": "operator_native_matched_s852003", "category": "operator", "context_ref": "textured", "budget_semantics": "matched_steps", "hypothesis": "Four native denoisers share absolute coordinate input, data, optimizer, target and update budget. Their operator, parameter count, normalization and connectivity differ as shown; finite-budget ordering requires actual sampled image quality.", "varying_axes": [ "model.mode" ], "controls": [ "All four available native operator cells are retained: convolution, one-level U-Net, flat MLP and patch Transformer.", "Every cell uses absolute coordinates, concat conditions, sinusoidal time, x0 target, 3000 updates, batch16, AdamW1e-3, the same materialized data and fixed evaluation noise/reference draws.", "This is a bundled native-recipe contrast: parameter count, activation, normalization and connectivity are explicitly shown; do not attribute the total difference exclusively to operator name.", "Primary choices were selected from original discovery validation only; the diagnostic Transformer is not promoted or substituted after new confirmation." ], "shortcut_checks": [ "always prefer U-Net for diffusion", "always prefer more parameters", "always prefer globally mixing operator" ], "explanation_target": "Explain both adjacent comparisons using exact native code and the fixed finite-budget task; use all four local cells and image/geometry/diversity diagnostics, while acknowledging unresolved parameter-count and optimization attribution.", "admission": "Freeze the new four-seed validation order, then independently test all six adjacent hypotheses from three new dataset instances using the unchanged16-seed sign-Holm rule and gap0.003. Keep rejected instances; no replacement.", "template_id": "operator_native_matched", "instance_seed": 852003, "dataset_node_id": "dataset_d8ad2ee2cf7b306a", "dataset_config": { "size": 16, "noise_steps": 32, "condition_dim": 6, "brightness_min": 0.65, "brightness_max": 1.0, "texture_strength": 0.12, "radius_min": 0.25, "radius_max": 0.42, "position_extent": 0.45, "reference_samples": 4, "train_size": 1024, "validation_size": 64, "test_size": 64, "ood_size": 32, "evaluation_shift": {}, "ood_shift": { "position_extent": 0.55 } }, "comparison": { "budget_semantics": "matched_steps", "changed_fields": [ "model.block", "model.kernel", "model.mode", "model.norm", "model.residual", "model.skip" ], "undeclared_fields": [ "model.block", "model.kernel", "model.norm", "model.residual", "model.skip" ], "comparison_class": "bundled_recipe", "field_values": { "conv": { "model.block": "dense", "model.kernel": 3, "model.mode": "conv", "model.norm": "group", "model.residual": true, "model.skip": "" }, "unet": { "model.block": "dense", "model.kernel": 3, "model.mode": "unet", "model.norm": "group", "model.residual": true, "model.skip": true }, "mlp": { "model.block": "", "model.kernel": "", "model.mode": "mlp", "model.norm": "", "model.residual": "", "model.skip": "" } }, "actual_changed_fields": [ "model.block", "model.kernel", "model.mode", "model.norm", "model.residual", "model.skip" ] }, "choices": [ { "id": "conv", "role": "primary", "model": { "type": "image_generation_denoiser", "mode": "conv", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 19489, "model_code_sha256": "548e5eb2edfaab77381773b5c3a085cafe768785e53bd9377556e95e2cf4161d", "node_id": "candidate_804cc8045512fe7cbec6" }, { "id": "unet", "role": "primary", "model": { "type": "image_generation_denoiser", "mode": "unet", "width": 24, "depth": 3, "kernel": 3, "norm": "group", "activation": "silu", "residual": true, "block": "dense", "skip": true, "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 35089, "model_code_sha256": "5eb9b3c4a5893cf189f5b3571dd497ba383e394e0d9c190e56276b84a9d36229", "node_id": "candidate_d4f3e193671168b46ced" }, { "id": "mlp", "role": "primary", "model": { "type": "image_generation_denoiser", "mode": "mlp", "width": 24, "depth": 3, "activation": "silu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "max_parameters": 2000000, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 461344, "model_code_sha256": "31d958601950f09831a638b54b51b3db7d388a4599ccfe063782adc4637d4b2c", "node_id": "candidate_eaae951afaff90809747" } ], "diagnostics": [ { "id": "transformer", "role": "diagnostic", "model": { "type": "image_generation_denoiser", "mode": "patch_transformer", "width": 24, "depth": 3, "activation": "gelu", "conditioning": "concat", "coordinates": "absolute", "time_embedding": "sinusoidal", "patch_size": 2, "max_parameters": 2000000, "num_heads": 4, "ff_multiplier": 2, "input_shape": [ 8, 16, 16 ], "output_shape": [ 1, 16, 16 ], "output_kind": "diffusion", "condition_dim": 6, "noise_steps": 32 }, "optimizer": { "type": "AdamW", "lr": 0.001, "weight_decay": 0.01, "betas": [ 0.9, 0.999 ], "eps": 1e-08, "foreach": false }, "loss": { "loss_id": "diffusion_mse" }, "training": { "batch_size": 16, "training_steps": 3000, "total_samples_seen": 48000, "gradient_accumulation": 1, "clip_norm": 1.0, "schedule": "constant", "precision": "fp32", "tf32": false, "deterministic_algorithms": true, "warmup_updates": 0, "batch_rng_device": "cpu", "paired_cpu_batch_stream": true, "persist_initial_state": true, "batch_seed_offset": 500000, "noise_seed_offset": 600000, "evaluation_split": "validation", "eval_steps": [ 750, 1500, 3000 ], "prediction_type": "x0", "noise_steps": 32, "noise_schedule": "cosine", "loss_weighting": "uniform", "min_snr_gamma": 5.0, "sampling_steps": 16, "sampling_eta": 0.0, "evaluation_samples_per_condition": 4, "evaluation_seed": 314159, "evaluation_batch_conditions": 16 }, "parameter_count": 17908, "model_code_sha256": "6a1a3aae0326db52c89fa6c204e71cb771024431856ade894f8a7732e5886275", "node_id": "candidate_72f3500e4f5cd3e445c8" } ] } ] }