Export trained reference models

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Hamza Cherkaoui
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,168 @@
+---
+library_name: gendynamics
+license: mit
+tags:
+- diffusion
+- unconditional-generation
+---
+
+# synthetic_alpha_stable_mixture
+
+Unconditional reference models trained with gendynamics. The weights correspond to the best validation epoch.
+
+## Dataset
+
+Synthetic unbalanced high-dimensional alpha-stable mixture. Generated by `gendynamics.datasets.fetch_synthetic_data('unbalanced_highdim_alpha_stable_mixture')`. Sample the configured mixture, then make a seeded random train/validation/test split. 
+The train/validation/test splits contain 10,000/1,000/10,000 samples
+of shape `[150]`. Normalization: None 
+The exact generator arguments or processed-file identity, split settings, and seed are in `dataset.json`.
+The source data themselves are not redistributed here; source-data terms remain with their provider.
+The preparation code is in [gendynamics](https://github.com/Diffusion-Research-Lab/gendynamics). API-backed sources can change; the processed-file checksum identifies this snapshot.
+
+## Architecture and training
+
+All independently trained models use this network architecture:
+
+```json
+{
+  "name": "mlp",
+  "params": {
+    "width": 240,
+    "depth": 4,
+    "time_dim": 128,
+    "dropout": 0.0,
+    "use_norm": true,
+    "dim": 150
+  }
+}
+```
+
+| Directory | Model | Prediction target | Trainable parameters | Sampling steps | Learning rate | Epochs run / best | Val loss | Test loss |
+|---|---|---|---:|---:|---:|---:|---:|---:|
+| ddpm-v | DDPM-V | velocity | 693,894 | 128 | 0.0002 | 250 / 150 | 6.379 | 5.197 |
+| dlpm-eps | DLPM-Eps | noise | 693,894 | 128 | 0.0005 | 640 / 558 | 0.8109 | 0.8493 |
+| tedm-origin | t-EDM | denoised data | 693,894 | 64 | 0.0005 | 179 / 79 | 1.069 | 2.891 |
+
+The complete **selected final** model and training settings are below and in each directory's `config.json`.
+The training settings include the maximum epoch budget; `epochs run / best` reports what actually happened.
+`provenance.json` records software versions and the source revision. Each model was reloaded and checked before upload;
+`export_check.json` records that check. Losses use each model's own objective, so they are not comparable sample-quality scores.
+The number of held-out rows used for each test loss is in `metrics.json`. Sample previews do not establish tail accuracy.
+
+### DDPM-V (`ddpm-v/`)
+
+```json
+{
+  "model_parameters": {
+    "n_steps": 128,
+    "sigma_max": 2.0,
+    "sampler": "ddpm"
+  },
+  "training": {
+    "device": "cuda",
+    "data_device": "cpu",
+    "batch_size": 1024,
+    "n_epochs": 640,
+    "early_stopping_patience": 100,
+    "lr": 0.0002,
+    "num_workers": 0,
+    "use_adamw": true,
+    "weight_decay": 1e-06,
+    "grad_clip_norm": 10.0,
+    "lr_schedule": "cosine",
+    "warmup_steps": 400,
+    "cosine_eta_min_ratio": 0.05,
+    "freq_logging": 50,
+    "stats_freq_epochs": 1,
+    "log_grad_norm": true,
+    "ckpt_freq_epochs": 64,
+    "ckpt_keep_last": 8
+  }
+}
+```
+### DLPM-Eps (`dlpm-eps/`)
+
+```json
+{
+  "model_parameters": {
+    "n_steps": 128,
+    "alpha": 1.9,
+    "sampler": "native"
+  },
+  "training": {
+    "device": "cuda",
+    "data_device": "cpu",
+    "batch_size": 1024,
+    "n_epochs": 640,
+    "early_stopping_patience": 100,
+    "lr": 0.0005,
+    "num_workers": 0,
+    "use_adamw": true,
+    "weight_decay": 1e-06,
+    "grad_clip_norm": 10.0,
+    "lr_schedule": "cosine",
+    "warmup_steps": 400,
+    "cosine_eta_min_ratio": 0.05,
+    "freq_logging": 50,
+    "stats_freq_epochs": 1,
+    "log_grad_norm": true,
+    "ckpt_freq_epochs": 64,
+    "ckpt_keep_last": 8
+  }
+}
+```
+### t-EDM (`tedm-origin/`)
+
+```json
+{
+  "model_parameters": {
+    "n_steps": 64,
+    "nu": 2.1,
+    "sigma_min": 0.005,
+    "sigma_max": 5.0,
+    "sigma_data": 1.0,
+    "solver": "edm_stochastic_heun"
+  },
+  "training": {
+    "device": "cuda",
+    "data_device": "cpu",
+    "batch_size": 1024,
+    "n_epochs": 640,
+    "early_stopping_patience": 100,
+    "lr": 0.0005,
+    "num_workers": 0,
+    "use_adamw": true,
+    "weight_decay": 1e-06,
+    "grad_clip_norm": 10.0,
+    "lr_schedule": "cosine",
+    "warmup_steps": 400,
+    "cosine_eta_min_ratio": 0.05,
+    "freq_logging": 50,
+    "stats_freq_epochs": 1,
+    "log_grad_norm": true,
+    "ckpt_freq_epochs": 64,
+    "ckpt_keep_last": 8
+  }
+}
+```
+
+DDIM-V (`ddim-v/`) shares the trained DDPM-V weights. Only the reverse sampler changes to deterministic DDIM (`eta=0`); it has no separate training run. Its exact sampling settings are in `ddim-v/config.json`.
+
+## Loading
+
+Install the training package and its dependencies:
+
+    python -m pip install git+https://github.com/Diffusion-Research-Lab/2_training_2026_tail_reference_models.git
+
+Then load one of the listed model directories by its underscored name:
+
+    from toolkit.huggingface import load_huggingface_model
+
+    model, normalization, data_scale = load_huggingface_model("Diffusion-Research-Lab/alpha-stable-mixture-1p7-150d-diffusion", "ddpm_v", device="cpu")
+    samples = model.sample(16).float().cpu()
+    if normalization is not None:
+        samples = samples * normalization["std"] + normalization["mean"]
+
+The model-specific files are `model.safetensors`, `config.json`, `dataset.json`, `metrics.json`,
+`provenance.json`, `export_check.json`, and, when applicable, `normalization.safetensors`.
+This repository is released under the MIT license. Cite the model repository revision when using these weights.

ddim-v/config.json

@@ -0,0 +1,165 @@
+{
+  "config": {
+    "run": {
+      "name": "synthetic_alpha_stable_mixture",
+      "seed": 0,
+      "dtype": "float64"
+    },
+    "dataset": {
+      "kind": "synthetic",
+      "name": "unbalanced_highdim_alpha_stable_mixture",
+      "params": {
+        "n_samples": 21000,
+        "alpha": 1.7,
+        "dim": 150,
+        "n_modes": 16,
+        "rank": 6,
+        "imbalance_tau": 1.2,
+        "mean_scale": 7.5,
+        "base_scale": 0.55,
+        "anisotropy": 1.0,
+        "structure_seed": 0
+      },
+      "split": {
+        "val_size": 0.047619047619047616,
+        "test_size": 0.47619047619047616,
+        "random_state": 0,
+        "standardize": false
+      }
+    },
+    "network": {
+      "name": "mlp",
+      "params": {
+        "width": 240,
+        "depth": 4,
+        "time_dim": 128,
+        "dropout": 0.0,
+        "use_norm": true
+      }
+    },
+    "models": {
+      "ddpm_v": {
+        "params": {
+          "n_steps": 128,
+          "sigma_max": 2.0,
+          "sampler": "ddpm"
+        }
+      },
+      "dlpm_eps": {
+        "params": {
+          "n_steps": 128,
+          "alpha": 1.7,
+          "sampler": "native"
+        }
+      },
+      "tedm_origin": {
+        "params": {
+          "n_steps": 64,
+          "nu": 2.1,
+          "sigma_min": 0.005,
+          "sigma_max": 5.0,
+          "sigma_data": 1.0,
+          "solver": "edm_stochastic_heun"
+        }
+      }
+    },
+    "pilot": {
+      "n_trials": 3,
+      "train": {
+        "n_epochs": 96,
+        "early_stopping_patience": 32,
+        "freq_logging": 16,
+        "stats_freq_epochs": 1
+      },
+      "sweep": {
+        "ddpm_v": {
+          "model": {
+            "sigma_max": [
+              2.0,
+              5.0,
+              8.0
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        },
+        "dlpm_eps": {
+          "model": {
+            "alpha": [
+              1.7,
+              1.9
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        },
+        "tedm_origin": {
+          "model": {
+            "nu": [
+              2.1,
+              3.0
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        }
+      }
+    },
+    "train": {
+      "device": "cuda",
+      "data_device": "cpu",
+      "batch_size": 1024,
+      "n_epochs": 640,
+      "early_stopping_patience": 100,
+      "lr": 0.0002,
+      "num_workers": 0,
+      "use_adamw": true,
+      "weight_decay": 1e-06,
+      "grad_clip_norm": 10.0,
+      "lr_schedule": "cosine",
+      "warmup_steps": 400,
+      "cosine_eta_min_ratio": 0.05,
+      "freq_logging": 50,
+      "stats_freq_epochs": 1,
+      "log_grad_norm": true,
+      "ckpt_freq_epochs": 64,
+      "ckpt_keep_last": 8
+    },
+    "test": {
+      "batch_size": 1024,
+      "loss_repeats": 4,
+      "n_samples": 32
+    },
+    "huggingface": {
+      "namespace": "Diffusion-Research-Lab",
+      "repository": "alpha-stable-mixture-1p7-150d-diffusion"
+    }
+  },
+  "model_name": "ddim_v",
+  "model_params": {
+    "n_steps": 128,
+    "sigma_max": 2.0,
+    "sampler": "ddim",
+    "eta": 0.0,
+    "sample_steps": null
+  },
+  "sample_shape": [
+    150
+  ],
+  "data_scale": 0.6880503018650794
+}

ddim-v/dataset.json

@@ -0,0 +1,39 @@
+{
+  "name": "unbalanced_highdim_alpha_stable_mixture",
+  "kind": "synthetic",
+  "params": {
+    "n_samples": 21000,
+    "alpha": 1.7,
+    "dim": 150,
+    "n_modes": 16,
+    "rank": 6,
+    "imbalance_tau": 1.2,
+    "mean_scale": 7.5,
+    "base_scale": 0.55,
+    "anisotropy": 1.0,
+    "structure_seed": 0
+  },
+  "split": {
+    "val_size": 0.047619047619047616,
+    "test_size": 0.47619047619047616,
+    "random_state": 0,
+    "standardize": false
+  },
+  "split_mode": "random",
+  "counts": {
+    "train": 10000,
+    "validation": 1000,
+    "test": 10000
+  },
+  "n_samples": 21000,
+  "sample_shape": [
+    150
+  ],
+  "seed": 0,
+  "normalization": "None",
+  "provenance": {
+    "origin": "Synthetic unbalanced high-dimensional alpha-stable mixture",
+    "generator": "gendynamics.datasets.fetch_synthetic_data('unbalanced_highdim_alpha_stable_mixture')",
+    "preparation": "Sample the configured mixture, then make a seeded random train/validation/test split."
+  }
+}

ddim-v/export_check.json

@@ -0,0 +1,21 @@
+{
+  "passed": true,
+  "shape": [
+    4,
+    150
+  ],
+  "finite": true,
+  "seed": 20000,
+  "batch_size": 1024,
+  "device": "cpu",
+  "model_dtype": "torch.float64",
+  "sample_dtype": "torch.float32",
+  "min": -4.537977695465088,
+  "max": 4.132815837860107,
+  "std": 1.3834874926946228,
+  "clamp_min": null,
+  "clipped_fraction": 0.0,
+  "scope": "Shape and numerical validity only; not distribution quality or tail accuracy.",
+  "reloaded_safetensors": true,
+  "trainable_parameters": 693894
+}

ddim-v/metrics.json

@@ -0,0 +1,6 @@
+{
+  "best_epoch": 150,
+  "validation_loss": 6.378554867091844,
+  "test_loss_samples": 10000,
+  "test_loss": 5.196955184231518
+}

ddim-v/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3dc36767f1245d4f37b896648a88e7a7995499206883b212c8df7445c7b650b1
+size 5554560

ddim-v/provenance.json

@@ -0,0 +1,19 @@
+{
+  "recorded_at_utc": "2026-10-04T00:15:48.261096+00:00",
+  "python": "3.12.11",
+  "platform": "Linux-5.14.0-570.131.1.el9_6.x86_64-x86_64-with-glibc2.34",
+  "device": "cuda",
+  "gpu": "Tesla V100-SXM2-16GB",
+  "cuda_version": "12.8",
+  "source": {
+    "commit": null,
+    "dirty": null
+  },
+  "dependencies": {
+    "torch": "2.8.0",
+    "gendynamics": "0.1.0",
+    "benchtools": "0.1.0",
+    "safetensors": "0.6.2",
+    "huggingface-hub": "0.34.4"
+  }
+}

ddpm-v/config.json

@@ -0,0 +1,163 @@
+{
+  "config": {
+    "run": {
+      "name": "synthetic_alpha_stable_mixture",
+      "seed": 0,
+      "dtype": "float64"
+    },
+    "dataset": {
+      "kind": "synthetic",
+      "name": "unbalanced_highdim_alpha_stable_mixture",
+      "params": {
+        "n_samples": 21000,
+        "alpha": 1.7,
+        "dim": 150,
+        "n_modes": 16,
+        "rank": 6,
+        "imbalance_tau": 1.2,
+        "mean_scale": 7.5,
+        "base_scale": 0.55,
+        "anisotropy": 1.0,
+        "structure_seed": 0
+      },
+      "split": {
+        "val_size": 0.047619047619047616,
+        "test_size": 0.47619047619047616,
+        "random_state": 0,
+        "standardize": false
+      }
+    },
+    "network": {
+      "name": "mlp",
+      "params": {
+        "width": 240,
+        "depth": 4,
+        "time_dim": 128,
+        "dropout": 0.0,
+        "use_norm": true
+      }
+    },
+    "models": {
+      "ddpm_v": {
+        "params": {
+          "n_steps": 128,
+          "sigma_max": 2.0,
+          "sampler": "ddpm"
+        }
+      },
+      "dlpm_eps": {
+        "params": {
+          "n_steps": 128,
+          "alpha": 1.7,
+          "sampler": "native"
+        }
+      },
+      "tedm_origin": {
+        "params": {
+          "n_steps": 64,
+          "nu": 2.1,
+          "sigma_min": 0.005,
+          "sigma_max": 5.0,
+          "sigma_data": 1.0,
+          "solver": "edm_stochastic_heun"
+        }
+      }
+    },
+    "pilot": {
+      "n_trials": 3,
+      "train": {
+        "n_epochs": 96,
+        "early_stopping_patience": 32,
+        "freq_logging": 16,
+        "stats_freq_epochs": 1
+      },
+      "sweep": {
+        "ddpm_v": {
+          "model": {
+            "sigma_max": [
+              2.0,
+              5.0,
+              8.0
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        },
+        "dlpm_eps": {
+          "model": {
+            "alpha": [
+              1.7,
+              1.9
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        },
+        "tedm_origin": {
+          "model": {
+            "nu": [
+              2.1,
+              3.0
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        }
+      }
+    },
+    "train": {
+      "device": "cuda",
+      "data_device": "cpu",
+      "batch_size": 1024,
+      "n_epochs": 640,
+      "early_stopping_patience": 100,
+      "lr": 0.0002,
+      "num_workers": 0,
+      "use_adamw": true,
+      "weight_decay": 1e-06,
+      "grad_clip_norm": 10.0,
+      "lr_schedule": "cosine",
+      "warmup_steps": 400,
+      "cosine_eta_min_ratio": 0.05,
+      "freq_logging": 50,
+      "stats_freq_epochs": 1,
+      "log_grad_norm": true,
+      "ckpt_freq_epochs": 64,
+      "ckpt_keep_last": 8
+    },
+    "test": {
+      "batch_size": 1024,
+      "loss_repeats": 4,
+      "n_samples": 32
+    },
+    "huggingface": {
+      "namespace": "Diffusion-Research-Lab",
+      "repository": "alpha-stable-mixture-1p7-150d-diffusion"
+    }
+  },
+  "model_name": "ddpm_v",
+  "model_params": {
+    "n_steps": 128,
+    "sigma_max": 2.0,
+    "sampler": "ddpm"
+  },
+  "sample_shape": [
+    150
+  ],
+  "data_scale": 0.6880503018650794
+}

ddpm-v/dataset.json

@@ -0,0 +1,39 @@
+{
+  "name": "unbalanced_highdim_alpha_stable_mixture",
+  "kind": "synthetic",
+  "params": {
+    "n_samples": 21000,
+    "alpha": 1.7,
+    "dim": 150,
+    "n_modes": 16,
+    "rank": 6,
+    "imbalance_tau": 1.2,
+    "mean_scale": 7.5,
+    "base_scale": 0.55,
+    "anisotropy": 1.0,
+    "structure_seed": 0
+  },
+  "split": {
+    "val_size": 0.047619047619047616,
+    "test_size": 0.47619047619047616,
+    "random_state": 0,
+    "standardize": false
+  },
+  "split_mode": "random",
+  "counts": {
+    "train": 10000,
+    "validation": 1000,
+    "test": 10000
+  },
+  "n_samples": 21000,
+  "sample_shape": [
+    150
+  ],
+  "seed": 0,
+  "normalization": "None",
+  "provenance": {
+    "origin": "Synthetic unbalanced high-dimensional alpha-stable mixture",
+    "generator": "gendynamics.datasets.fetch_synthetic_data('unbalanced_highdim_alpha_stable_mixture')",
+    "preparation": "Sample the configured mixture, then make a seeded random train/validation/test split."
+  }
+}

ddpm-v/export_check.json

@@ -0,0 +1,21 @@
+{
+  "passed": true,
+  "shape": [
+    4,
+    150
+  ],
+  "finite": true,
+  "seed": 20000,
+  "batch_size": 1024,
+  "device": "cpu",
+  "model_dtype": "torch.float64",
+  "sample_dtype": "torch.float32",
+  "min": -4.71538782119751,
+  "max": 3.6429829597473145,
+  "std": 1.3612132868099323,
+  "clamp_min": null,
+  "clipped_fraction": 0.0,
+  "scope": "Shape and numerical validity only; not distribution quality or tail accuracy.",
+  "reloaded_safetensors": true,
+  "trainable_parameters": 693894
+}

ddpm-v/metrics.json

@@ -0,0 +1,6 @@
+{
+  "best_epoch": 150,
+  "validation_loss": 6.378554867091844,
+  "test_loss_samples": 10000,
+  "test_loss": 5.196955184231518
+}

ddpm-v/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3dc36767f1245d4f37b896648a88e7a7995499206883b212c8df7445c7b650b1
+size 5554560

ddpm-v/provenance.json

@@ -0,0 +1,19 @@
+{
+  "recorded_at_utc": "2026-10-04T00:15:48.261096+00:00",
+  "python": "3.12.11",
+  "platform": "Linux-5.14.0-570.131.1.el9_6.x86_64-x86_64-with-glibc2.34",
+  "device": "cuda",
+  "gpu": "Tesla V100-SXM2-16GB",
+  "cuda_version": "12.8",
+  "source": {
+    "commit": null,
+    "dirty": null
+  },
+  "dependencies": {
+    "torch": "2.8.0",
+    "gendynamics": "0.1.0",
+    "benchtools": "0.1.0",
+    "safetensors": "0.6.2",
+    "huggingface-hub": "0.34.4"
+  }
+}

dlpm-eps/config.json

@@ -0,0 +1,163 @@
+{
+  "config": {
+    "run": {
+      "name": "synthetic_alpha_stable_mixture",
+      "seed": 0,
+      "dtype": "float64"
+    },
+    "dataset": {
+      "kind": "synthetic",
+      "name": "unbalanced_highdim_alpha_stable_mixture",
+      "params": {
+        "n_samples": 21000,
+        "alpha": 1.7,
+        "dim": 150,
+        "n_modes": 16,
+        "rank": 6,
+        "imbalance_tau": 1.2,
+        "mean_scale": 7.5,
+        "base_scale": 0.55,
+        "anisotropy": 1.0,
+        "structure_seed": 0
+      },
+      "split": {
+        "val_size": 0.047619047619047616,
+        "test_size": 0.47619047619047616,
+        "random_state": 0,
+        "standardize": false
+      }
+    },
+    "network": {
+      "name": "mlp",
+      "params": {
+        "width": 240,
+        "depth": 4,
+        "time_dim": 128,
+        "dropout": 0.0,
+        "use_norm": true
+      }
+    },
+    "models": {
+      "ddpm_v": {
+        "params": {
+          "n_steps": 128,
+          "sigma_max": 5.0,
+          "sampler": "ddpm"
+        }
+      },
+      "dlpm_eps": {
+        "params": {
+          "n_steps": 128,
+          "alpha": 1.9,
+          "sampler": "native"
+        }
+      },
+      "tedm_origin": {
+        "params": {
+          "n_steps": 64,
+          "nu": 2.1,
+          "sigma_min": 0.005,
+          "sigma_max": 5.0,
+          "sigma_data": 1.0,
+          "solver": "edm_stochastic_heun"
+        }
+      }
+    },
+    "pilot": {
+      "n_trials": 3,
+      "train": {
+        "n_epochs": 96,
+        "early_stopping_patience": 32,
+        "freq_logging": 16,
+        "stats_freq_epochs": 1
+      },
+      "sweep": {
+        "ddpm_v": {
+          "model": {
+            "sigma_max": [
+              2.0,
+              5.0,
+              8.0
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        },
+        "dlpm_eps": {
+          "model": {
+            "alpha": [
+              1.7,
+              1.9
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        },
+        "tedm_origin": {
+          "model": {
+            "nu": [
+              2.1,
+              3.0
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        }
+      }
+    },
+    "train": {
+      "device": "cuda",
+      "data_device": "cpu",
+      "batch_size": 1024,
+      "n_epochs": 640,
+      "early_stopping_patience": 100,
+      "lr": 0.0005,
+      "num_workers": 0,
+      "use_adamw": true,
+      "weight_decay": 1e-06,
+      "grad_clip_norm": 10.0,
+      "lr_schedule": "cosine",
+      "warmup_steps": 400,
+      "cosine_eta_min_ratio": 0.05,
+      "freq_logging": 50,
+      "stats_freq_epochs": 1,
+      "log_grad_norm": true,
+      "ckpt_freq_epochs": 64,
+      "ckpt_keep_last": 8
+    },
+    "test": {
+      "batch_size": 1024,
+      "loss_repeats": 4,
+      "n_samples": 32
+    },
+    "huggingface": {
+      "namespace": "Diffusion-Research-Lab",
+      "repository": "alpha-stable-mixture-1p7-150d-diffusion"
+    }
+  },
+  "model_name": "dlpm_eps",
+  "model_params": {
+    "n_steps": 128,
+    "alpha": 1.9,
+    "sampler": "native"
+  },
+  "sample_shape": [
+    150
+  ],
+  "data_scale": 0.6880503018650794
+}

dlpm-eps/dataset.json

@@ -0,0 +1,39 @@
+{
+  "name": "unbalanced_highdim_alpha_stable_mixture",
+  "kind": "synthetic",
+  "params": {
+    "n_samples": 21000,
+    "alpha": 1.7,
+    "dim": 150,
+    "n_modes": 16,
+    "rank": 6,
+    "imbalance_tau": 1.2,
+    "mean_scale": 7.5,
+    "base_scale": 0.55,
+    "anisotropy": 1.0,
+    "structure_seed": 0
+  },
+  "split": {
+    "val_size": 0.047619047619047616,
+    "test_size": 0.47619047619047616,
+    "random_state": 0,
+    "standardize": false
+  },
+  "split_mode": "random",
+  "counts": {
+    "train": 10000,
+    "validation": 1000,
+    "test": 10000
+  },
+  "n_samples": 21000,
+  "sample_shape": [
+    150
+  ],
+  "seed": 0,
+  "normalization": "None",
+  "provenance": {
+    "origin": "Synthetic unbalanced high-dimensional alpha-stable mixture",
+    "generator": "gendynamics.datasets.fetch_synthetic_data('unbalanced_highdim_alpha_stable_mixture')",
+    "preparation": "Sample the configured mixture, then make a seeded random train/validation/test split."
+  }
+}

dlpm-eps/export_check.json

@@ -0,0 +1,21 @@
+{
+  "passed": true,
+  "shape": [
+    4,
+    150
+  ],
+  "finite": true,
+  "seed": 20000,
+  "batch_size": 1024,
+  "device": "cpu",
+  "model_dtype": "torch.float64",
+  "sample_dtype": "torch.float32",
+  "min": -2.6529760360717773,
+  "max": 3.236975908279419,
+  "std": 0.9366270955322801,
+  "clamp_min": null,
+  "clipped_fraction": 0.0,
+  "scope": "Shape and numerical validity only; not distribution quality or tail accuracy.",
+  "reloaded_safetensors": true,
+  "trainable_parameters": 693894
+}

dlpm-eps/metrics.json

@@ -0,0 +1,6 @@
+{
+  "best_epoch": 558,
+  "validation_loss": 0.810883340373693,
+  "test_loss_samples": 10000,
+  "test_loss": 0.8492910243112395
+}

dlpm-eps/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5064bc90ac0c04de7f65721adc5af6d914d9c7f11d99b7006b8630a110c9668c
+size 5554560

dlpm-eps/provenance.json

@@ -0,0 +1,19 @@
+{
+  "recorded_at_utc": "2026-10-04T00:15:40.661444+00:00",
+  "python": "3.12.11",
+  "platform": "Linux-5.14.0-570.131.1.el9_6.x86_64-x86_64-with-glibc2.34",
+  "device": "cuda",
+  "gpu": "Tesla V100-SXM2-16GB",
+  "cuda_version": "12.8",
+  "source": {
+    "commit": null,
+    "dirty": null
+  },
+  "dependencies": {
+    "torch": "2.8.0",
+    "gendynamics": "0.1.0",
+    "benchtools": "0.1.0",
+    "safetensors": "0.6.2",
+    "huggingface-hub": "0.34.4"
+  }
+}

tedm-origin/config.json

@@ -0,0 +1,166 @@
+{
+  "config": {
+    "run": {
+      "name": "synthetic_alpha_stable_mixture",
+      "seed": 0,
+      "dtype": "float64"
+    },
+    "dataset": {
+      "kind": "synthetic",
+      "name": "unbalanced_highdim_alpha_stable_mixture",
+      "params": {
+        "n_samples": 21000,
+        "alpha": 1.7,
+        "dim": 150,
+        "n_modes": 16,
+        "rank": 6,
+        "imbalance_tau": 1.2,
+        "mean_scale": 7.5,
+        "base_scale": 0.55,
+        "anisotropy": 1.0,
+        "structure_seed": 0
+      },
+      "split": {
+        "val_size": 0.047619047619047616,
+        "test_size": 0.47619047619047616,
+        "random_state": 0,
+        "standardize": false
+      }
+    },
+    "network": {
+      "name": "mlp",
+      "params": {
+        "width": 240,
+        "depth": 4,
+        "time_dim": 128,
+        "dropout": 0.0,
+        "use_norm": true
+      }
+    },
+    "models": {
+      "ddpm_v": {
+        "params": {
+          "n_steps": 128,
+          "sigma_max": 5.0,
+          "sampler": "ddpm"
+        }
+      },
+      "dlpm_eps": {
+        "params": {
+          "n_steps": 128,
+          "alpha": 1.7,
+          "sampler": "native"
+        }
+      },
+      "tedm_origin": {
+        "params": {
+          "n_steps": 64,
+          "nu": 2.1,
+          "sigma_min": 0.005,
+          "sigma_max": 5.0,
+          "sigma_data": 1.0,
+          "solver": "edm_stochastic_heun"
+        }
+      }
+    },
+    "pilot": {
+      "n_trials": 3,
+      "train": {
+        "n_epochs": 96,
+        "early_stopping_patience": 32,
+        "freq_logging": 16,
+        "stats_freq_epochs": 1
+      },
+      "sweep": {
+        "ddpm_v": {
+          "model": {
+            "sigma_max": [
+              2.0,
+              5.0,
+              8.0
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        },
+        "dlpm_eps": {
+          "model": {
+            "alpha": [
+              1.7,
+              1.9
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        },
+        "tedm_origin": {
+          "model": {
+            "nu": [
+              2.1,
+              3.0
+            ]
+          },
+          "train": {
+            "lr": [
+              0.0001,
+              0.0002,
+              0.0005
+            ]
+          }
+        }
+      }
+    },
+    "train": {
+      "device": "cuda",
+      "data_device": "cpu",
+      "batch_size": 1024,
+      "n_epochs": 640,
+      "early_stopping_patience": 100,
+      "lr": 0.0005,
+      "num_workers": 0,
+      "use_adamw": true,
+      "weight_decay": 1e-06,
+      "grad_clip_norm": 10.0,
+      "lr_schedule": "cosine",
+      "warmup_steps": 400,
+      "cosine_eta_min_ratio": 0.05,
+      "freq_logging": 50,
+      "stats_freq_epochs": 1,
+      "log_grad_norm": true,
+      "ckpt_freq_epochs": 64,
+      "ckpt_keep_last": 8
+    },
+    "test": {
+      "batch_size": 1024,
+      "loss_repeats": 4,
+      "n_samples": 32
+    },
+    "huggingface": {
+      "namespace": "Diffusion-Research-Lab",
+      "repository": "alpha-stable-mixture-1p7-150d-diffusion"
+    }
+  },
+  "model_name": "tedm_origin",
+  "model_params": {
+    "n_steps": 64,
+    "nu": 2.1,
+    "sigma_min": 0.005,
+    "sigma_max": 5.0,
+    "sigma_data": 1.0,
+    "solver": "edm_stochastic_heun"
+  },
+  "sample_shape": [
+    150
+  ],
+  "data_scale": 0.6880503018650794
+}

tedm-origin/dataset.json

@@ -0,0 +1,39 @@
+{
+  "name": "unbalanced_highdim_alpha_stable_mixture",
+  "kind": "synthetic",
+  "params": {
+    "n_samples": 21000,
+    "alpha": 1.7,
+    "dim": 150,
+    "n_modes": 16,
+    "rank": 6,
+    "imbalance_tau": 1.2,
+    "mean_scale": 7.5,
+    "base_scale": 0.55,
+    "anisotropy": 1.0,
+    "structure_seed": 0
+  },
+  "split": {
+    "val_size": 0.047619047619047616,
+    "test_size": 0.47619047619047616,
+    "random_state": 0,
+    "standardize": false
+  },
+  "split_mode": "random",
+  "counts": {
+    "train": 10000,
+    "validation": 1000,
+    "test": 10000
+  },
+  "n_samples": 21000,
+  "sample_shape": [
+    150
+  ],
+  "seed": 0,
+  "normalization": "None",
+  "provenance": {
+    "origin": "Synthetic unbalanced high-dimensional alpha-stable mixture",
+    "generator": "gendynamics.datasets.fetch_synthetic_data('unbalanced_highdim_alpha_stable_mixture')",
+    "preparation": "Sample the configured mixture, then make a seeded random train/validation/test split."
+  }
+}

tedm-origin/export_check.json

@@ -0,0 +1,21 @@
+{
+  "passed": true,
+  "shape": [
+    4,
+    150
+  ],
+  "finite": true,
+  "seed": 20000,
+  "batch_size": 1024,
+  "device": "cpu",
+  "model_dtype": "torch.float64",
+  "sample_dtype": "torch.float32",
+  "min": -5.119678974151611,
+  "max": 5.398929119110107,
+  "std": 1.3876008204771824,
+  "clamp_min": null,
+  "clipped_fraction": 0.0,
+  "scope": "Shape and numerical validity only; not distribution quality or tail accuracy.",
+  "reloaded_safetensors": true,
+  "trainable_parameters": 693894
+}

tedm-origin/metrics.json

@@ -0,0 +1,6 @@
+{
+  "best_epoch": 79,
+  "validation_loss": 1.0691735461594984,
+  "test_loss_samples": 10000,
+  "test_loss": 2.891158615260325
+}

tedm-origin/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73cb3cc7c91bccee6cbdcc2f7257ffdf0131458ff454e4d2fef81cd522f29168
+size 5554560

tedm-origin/provenance.json

@@ -0,0 +1,19 @@
+{
+  "recorded_at_utc": "2026-10-04T00:15:18.189854+00:00",
+  "python": "3.12.11",
+  "platform": "Linux-5.14.0-570.131.1.el9_6.x86_64-x86_64-with-glibc2.34",
+  "device": "cuda",
+  "gpu": "Tesla V100-SXM2-16GB",
+  "cuda_version": "12.8",
+  "source": {
+    "commit": null,
+    "dirty": null
+  },
+  "dependencies": {
+    "torch": "2.8.0",
+    "gendynamics": "0.1.0",
+    "benchtools": "0.1.0",
+    "safetensors": "0.6.2",
+    "huggingface-hub": "0.34.4"
+  }
+}