Upload checkpoints/normalized_registry.json

checkpoints/normalized_registry.json

@@ -0,0 +1,1871 @@
+{
+  "checkpoints": [
+    {
+      "auxiliary_files": [
+        "aux/MODEL_LICENSE.json",
+        "aux/preprocessing.json",
+        "aux/uspto_templates.csv.gz",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/ref_routes_reaction_hashes.txt",
+        "benchmarks/native/selected_reactions.csv"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "paroutes_reference_route_holdout",
+          "dataset_id": "aizynthfinder_uspto_onnx_test",
+          "dataset_ref": "paroutes2_reference_route_reactions",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tab_delimited_reactants_reagents_product_rows",
+          "native_source_paths": [
+            "benchmarks/native/selected_reactions.csv",
+            "benchmarks/native/ref_routes_reaction_hashes.txt"
+          ],
+          "notes": [
+            "Default bundled benchmark for the public AiZynthFinder USPTO ONNX policy.",
+            "The benchmark rows are reconstructed from PaRoutes selected_reactions.csv and ref_routes_reaction_hashes.txt so they match the held-out reference-route reactions used as the AiZynthTrain test set."
+          ],
+          "num_examples": 56657,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_model_onnx",
+      "input_schema": null,
+      "model_id": "aizynthfinder",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_model_onnx",
+      "output_schema": null,
+      "package_dir": "checkpoints/aizynthfinder/uspto_model_onnx",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_model_onnx",
+          "model_id": "aizynthfinder"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "aizynthtrain_uspto_selected_reactions_with_paroutes_holdout"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "aizynthfinder_template_policy"
+    },
+    {
+      "auxiliary_files": [
+        "aux/MAR-INF/MANIFEST.json",
+        "aux/mar_files/at_handler.py",
+        "aux/mar_files/utils.py",
+        "aux/mar_members.json",
+        "aux/opennmt_metadata.json",
+        "aux/opennmt_opt_summary.json",
+        "aux/safetensors_conversion.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "benchmark_pending",
+          "dataset_id": "askcos_retro_augmented_transformer_pistachio_23q3_benchmark_pending",
+          "dataset_ref": "askcos_pistachio_23q3_release",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "benchmark_not_available",
+          "native_source_paths": [],
+          "notes": [
+            "No redistribution-safe native benchmark split for this ASKCOS model archive is staged in this repository.",
+            "The bundle keeps benchmark metadata so downstream tooling can distinguish unavailable benchmarks from accidental omissions."
+          ],
+          "num_examples": 0,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "pistachio_23q3",
+      "input_schema": {
+        "product_field": "smiles",
+        "request_format": {
+          "smiles": [
+            "product_smiles"
+          ]
+        },
+        "target_field": "reactants"
+      },
+      "model_id": "askcos_retro_augmented_transformer",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "pistachio_23q3",
+      "output_schema": null,
+      "package_dir": "checkpoints/askcos_retro_augmented_transformer/pistachio_23q3",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "pistachio_23q3",
+          "model_id": "askcos_retro_augmented_transformer"
+        }
+      ],
+      "task_direction": "retrosynthesis",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "askcos_pistachio_23q3_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "askcos_augmented_transformer"
+    },
+    {
+      "auxiliary_files": [
+        "aux/MAR-INF/MANIFEST.json",
+        "aux/mar_files/at_handler.py",
+        "aux/mar_files/utils.py",
+        "aux/mar_members.json",
+        "aux/opennmt_metadata.json",
+        "aux/opennmt_opt_summary.json",
+        "aux/safetensors_conversion.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/src-test.txt",
+        "benchmarks/native/tgt-test.txt"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "askcos_uspto_full_r_smiles_compatibility_view",
+          "dataset_id": "askcos_retro_augmented_transformer_uspto_full_uspto_full_test",
+          "dataset_ref": "uspto_full_gln",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tokenized_parallel_txt",
+          "native_source_paths": [
+            "benchmarks/native/src-test.txt",
+            "benchmarks/native/tgt-test.txt"
+          ],
+          "notes": [
+            "ASKCOS does not redistribute a native USPTO-FULL benchmark CSV with this TorchServe archive in datasets/modelos_ASKCOS.",
+            "This bundle uses the public R-SMILES/GLN USPTO-FULL tokenized test split as a redistribution-safe compatibility view.",
+            "benchmarks/default_test.jsonl collapses the contiguous augmentation factor of 5 back to one benchmark case per underlying reaction."
+          ],
+          "num_examples": 96023,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_full",
+      "input_schema": {
+        "product_field": "smiles",
+        "request_format": {
+          "smiles": [
+            "product_smiles"
+          ]
+        },
+        "target_field": "reactants"
+      },
+      "model_id": "askcos_retro_augmented_transformer",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_full",
+      "output_schema": null,
+      "package_dir": "checkpoints/askcos_retro_augmented_transformer/uspto_full",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_full",
+          "model_id": "askcos_retro_augmented_transformer"
+        }
+      ],
+      "task_direction": "retrosynthesis",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "askcos_uspto_full_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "askcos_augmented_transformer"
+    },
+    {
+      "auxiliary_files": [
+        "aux/MAR-INF/MANIFEST.json",
+        "aux/checkpoint_metadata.json",
+        "aux/graph2smiles_runtime/models/attention_xl.py",
+        "aux/graph2smiles_runtime/models/dgat.py",
+        "aux/graph2smiles_runtime/models/dgcn.py",
+        "aux/graph2smiles_runtime/models/graph2smiles.py",
+        "aux/graph2smiles_runtime/models/graphfeat.py",
+        "aux/graph2smiles_runtime/models/model_utils.py",
+        "aux/graph2smiles_runtime/models/transformer_decoder.py",
+        "aux/graph2smiles_runtime/predict.py",
+        "aux/graph2smiles_runtime/train.py",
+        "aux/graph2smiles_runtime/utils/c_calculate.so",
+        "aux/graph2smiles_runtime/utils/chem_utils.py",
+        "aux/graph2smiles_runtime/utils/ctypes_calculator.py",
+        "aux/graph2smiles_runtime/utils/data_utils.py",
+        "aux/graph2smiles_runtime/utils/parsing.py",
+        "aux/graph2smiles_runtime/utils/preprocess_utils.py",
+        "aux/graph2smiles_runtime/utils/rxn_graphs.py",
+        "aux/graph2smiles_runtime/utils/train_utils.py",
+        "aux/mar_files/handler.py",
+        "aux/mar_files/models.zip",
+        "aux/mar_files/predict.py",
+        "aux/mar_files/train.py",
+        "aux/mar_files/utils.zip",
+        "aux/mar_files/vocab.txt",
+        "aux/mar_members.json",
+        "aux/model_args.json",
+        "aux/safetensors_conversion.json",
+        "aux/vocab.txt",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "benchmark_pending",
+          "dataset_id": "askcos_retro_graph2smiles_pistachio_23q3_benchmark_pending",
+          "dataset_ref": "askcos_pistachio_23q3_release",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "benchmark_not_available",
+          "native_source_paths": [],
+          "notes": [
+            "No redistribution-safe native benchmark split for this ASKCOS model archive is staged in this repository.",
+            "The bundle keeps benchmark metadata so downstream tooling can distinguish unavailable benchmarks from accidental omissions."
+          ],
+          "num_examples": 0,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "pistachio_23q3",
+      "input_schema": {
+        "product_field": "smiles",
+        "request_format": {
+          "smiles": [
+            "product_smiles"
+          ]
+        },
+        "target_field": "reactants"
+      },
+      "model_id": "askcos_retro_graph2smiles",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "pistachio_23q3",
+      "output_schema": null,
+      "package_dir": "checkpoints/askcos_retro_graph2smiles/pistachio_23q3",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "pistachio_23q3",
+          "model_id": "askcos_retro_graph2smiles"
+        }
+      ],
+      "task_direction": "retrosynthesis",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "askcos_pistachio_23q3_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "askcos_graph2smiles"
+    },
+    {
+      "auxiliary_files": [
+        "aux/MAR-INF/MANIFEST.json",
+        "aux/checkpoint_metadata.json",
+        "aux/graph2smiles_runtime/models/attention_xl.py",
+        "aux/graph2smiles_runtime/models/dgat.py",
+        "aux/graph2smiles_runtime/models/dgcn.py",
+        "aux/graph2smiles_runtime/models/graph2smiles.py",
+        "aux/graph2smiles_runtime/models/graphfeat.py",
+        "aux/graph2smiles_runtime/models/model_utils.py",
+        "aux/graph2smiles_runtime/models/transformer_decoder.py",
+        "aux/graph2smiles_runtime/predict.py",
+        "aux/graph2smiles_runtime/train.py",
+        "aux/graph2smiles_runtime/utils/c_calculate.so",
+        "aux/graph2smiles_runtime/utils/chem_utils.py",
+        "aux/graph2smiles_runtime/utils/ctypes_calculator.py",
+        "aux/graph2smiles_runtime/utils/data_utils.py",
+        "aux/graph2smiles_runtime/utils/parsing.py",
+        "aux/graph2smiles_runtime/utils/preprocess_utils.py",
+        "aux/graph2smiles_runtime/utils/rxn_graphs.py",
+        "aux/graph2smiles_runtime/utils/train_utils.py",
+        "aux/mar_files/handler.py",
+        "aux/mar_files/models.zip",
+        "aux/mar_files/predict.py",
+        "aux/mar_files/train.py",
+        "aux/mar_files/utils.zip",
+        "aux/mar_files/vocab.txt",
+        "aux/mar_members.json",
+        "aux/model_args.json",
+        "aux/safetensors_conversion.json",
+        "aux/vocab.txt",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/src-test.txt",
+        "benchmarks/native/tgt-test.txt"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "askcos_uspto_full_r_smiles_compatibility_view",
+          "dataset_id": "askcos_retro_graph2smiles_uspto_full_uspto_full_test",
+          "dataset_ref": "uspto_full_gln",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tokenized_parallel_txt",
+          "native_source_paths": [
+            "benchmarks/native/src-test.txt",
+            "benchmarks/native/tgt-test.txt"
+          ],
+          "notes": [
+            "ASKCOS does not redistribute a native USPTO-FULL benchmark CSV with this TorchServe archive in datasets/modelos_ASKCOS.",
+            "This bundle uses the public R-SMILES/GLN USPTO-FULL tokenized test split as a redistribution-safe compatibility view.",
+            "benchmarks/default_test.jsonl collapses the contiguous augmentation factor of 5 back to one benchmark case per underlying reaction."
+          ],
+          "num_examples": 96023,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_full",
+      "input_schema": {
+        "product_field": "smiles",
+        "request_format": {
+          "smiles": [
+            "product_smiles"
+          ]
+        },
+        "target_field": "reactants"
+      },
+      "model_id": "askcos_retro_graph2smiles",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_full",
+      "output_schema": null,
+      "package_dir": "checkpoints/askcos_retro_graph2smiles/uspto_full",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_full",
+          "model_id": "askcos_retro_graph2smiles"
+        }
+      ],
+      "task_direction": "retrosynthesis",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "askcos_uspto_full_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "askcos_graph2smiles"
+    },
+    {
+      "auxiliary_files": [
+        "aux/MAR-INF/MANIFEST.json",
+        "aux/mar_files/misc.py",
+        "aux/mar_files/models.py",
+        "aux/mar_files/templ_rel_handler.py",
+        "aux/mar_files/templ_rel_parser.py",
+        "aux/mar_files/utils.py",
+        "aux/mar_members.json",
+        "aux/model_args.json",
+        "aux/safetensors_conversion.json",
+        "aux/templates.jsonl",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/ref_routes_reaction_hashes.txt",
+        "benchmarks/native/selected_reactions.csv"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "askcos_paroutes_compatibility_view",
+          "dataset_id": "askcos_retro_template_relevance_bkms_metabolic_paroutes_reference_route_test",
+          "dataset_ref": "paroutes2_reference_route_reactions",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tab_delimited_reactants_reagents_product_rows",
+          "native_source_paths": [
+            "benchmarks/native/selected_reactions.csv",
+            "benchmarks/native/ref_routes_reaction_hashes.txt"
+          ],
+          "notes": [
+            "ASKCOS does not redistribute the native train/validation/test split for this TorchServe archive in datasets/modelos_ASKCOS.",
+            "This bundled benchmark is therefore a public PaRoutes one-step retrosynthesis compatibility view, not a claim about the model's original validation split."
+          ],
+          "num_examples": 56657,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "bkms_metabolic",
+      "input_schema": {
+        "product_field": "smiles",
+        "request_format": {
+          "smiles": [
+            "product_smiles"
+          ]
+        },
+        "target_field": "reactants"
+      },
+      "model_id": "askcos_retro_template_relevance",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "bkms_metabolic",
+      "output_schema": null,
+      "package_dir": "checkpoints/askcos_retro_template_relevance/bkms_metabolic",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "bkms_metabolic",
+          "model_id": "askcos_retro_template_relevance"
+        }
+      ],
+      "task_direction": "retrosynthesis",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "askcos_bkms_metabolic_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "askcos_template_relevance"
+    },
+    {
+      "auxiliary_files": [
+        "aux/MAR-INF/MANIFEST.json",
+        "aux/mar_files/misc.py",
+        "aux/mar_files/models.py",
+        "aux/mar_files/templ_rel_handler.py",
+        "aux/mar_files/templ_rel_parser.py",
+        "aux/mar_files/utils.py",
+        "aux/mar_members.json",
+        "aux/model_args.json",
+        "aux/safetensors_conversion.json",
+        "aux/templates.jsonl",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/ref_routes_reaction_hashes.txt",
+        "benchmarks/native/selected_reactions.csv"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "askcos_paroutes_compatibility_view",
+          "dataset_id": "askcos_retro_template_relevance_pistachio_paroutes_reference_route_test",
+          "dataset_ref": "paroutes2_reference_route_reactions",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tab_delimited_reactants_reagents_product_rows",
+          "native_source_paths": [
+            "benchmarks/native/selected_reactions.csv",
+            "benchmarks/native/ref_routes_reaction_hashes.txt"
+          ],
+          "notes": [
+            "ASKCOS does not redistribute the native train/validation/test split for this TorchServe archive in datasets/modelos_ASKCOS.",
+            "This bundled benchmark is therefore a public PaRoutes one-step retrosynthesis compatibility view, not a claim about the model's original validation split."
+          ],
+          "num_examples": 56657,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "pistachio",
+      "input_schema": {
+        "product_field": "smiles",
+        "request_format": {
+          "smiles": [
+            "product_smiles"
+          ]
+        },
+        "target_field": "reactants"
+      },
+      "model_id": "askcos_retro_template_relevance",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "pistachio",
+      "output_schema": null,
+      "package_dir": "checkpoints/askcos_retro_template_relevance/pistachio",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "pistachio",
+          "model_id": "askcos_retro_template_relevance"
+        }
+      ],
+      "task_direction": "retrosynthesis",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "askcos_pistachio_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "askcos_template_relevance"
+    },
+    {
+      "auxiliary_files": [
+        "aux/MAR-INF/MANIFEST.json",
+        "aux/mar_files/misc.py",
+        "aux/mar_files/models.py",
+        "aux/mar_files/templ_rel_handler.py",
+        "aux/mar_files/templ_rel_parser.py",
+        "aux/mar_files/utils.py",
+        "aux/mar_members.json",
+        "aux/model_args.json",
+        "aux/safetensors_conversion.json",
+        "aux/templates.jsonl",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/ref_routes_reaction_hashes.txt",
+        "benchmarks/native/selected_reactions.csv"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "askcos_paroutes_compatibility_view",
+          "dataset_id": "askcos_retro_template_relevance_pistachio_ringbreaker_paroutes_reference_route_test",
+          "dataset_ref": "paroutes2_reference_route_reactions",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tab_delimited_reactants_reagents_product_rows",
+          "native_source_paths": [
+            "benchmarks/native/selected_reactions.csv",
+            "benchmarks/native/ref_routes_reaction_hashes.txt"
+          ],
+          "notes": [
+            "ASKCOS does not redistribute the native train/validation/test split for this TorchServe archive in datasets/modelos_ASKCOS.",
+            "This bundled benchmark is therefore a public PaRoutes one-step retrosynthesis compatibility view, not a claim about the model's original validation split."
+          ],
+          "num_examples": 56657,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "pistachio_ringbreaker",
+      "input_schema": {
+        "product_field": "smiles",
+        "request_format": {
+          "smiles": [
+            "product_smiles"
+          ]
+        },
+        "target_field": "reactants"
+      },
+      "model_id": "askcos_retro_template_relevance",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "pistachio_ringbreaker",
+      "output_schema": null,
+      "package_dir": "checkpoints/askcos_retro_template_relevance/pistachio_ringbreaker",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "pistachio_ringbreaker",
+          "model_id": "askcos_retro_template_relevance"
+        }
+      ],
+      "task_direction": "retrosynthesis",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "askcos_pistachio_ringbreaker_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "askcos_template_relevance"
+    },
+    {
+      "auxiliary_files": [
+        "aux/MAR-INF/MANIFEST.json",
+        "aux/mar_files/misc.py",
+        "aux/mar_files/models.py",
+        "aux/mar_files/templ_rel_handler.py",
+        "aux/mar_files/templ_rel_parser.py",
+        "aux/mar_files/utils.py",
+        "aux/mar_members.json",
+        "aux/model_args.json",
+        "aux/safetensors_conversion.json",
+        "aux/templates.jsonl",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/ref_routes_reaction_hashes.txt",
+        "benchmarks/native/selected_reactions.csv"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "askcos_paroutes_compatibility_view",
+          "dataset_id": "askcos_retro_template_relevance_reaxys_paroutes_reference_route_test",
+          "dataset_ref": "paroutes2_reference_route_reactions",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tab_delimited_reactants_reagents_product_rows",
+          "native_source_paths": [
+            "benchmarks/native/selected_reactions.csv",
+            "benchmarks/native/ref_routes_reaction_hashes.txt"
+          ],
+          "notes": [
+            "ASKCOS does not redistribute the native train/validation/test split for this TorchServe archive in datasets/modelos_ASKCOS.",
+            "This bundled benchmark is therefore a public PaRoutes one-step retrosynthesis compatibility view, not a claim about the model's original validation split."
+          ],
+          "num_examples": 56657,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "reaxys",
+      "input_schema": {
+        "product_field": "smiles",
+        "request_format": {
+          "smiles": [
+            "product_smiles"
+          ]
+        },
+        "target_field": "reactants"
+      },
+      "model_id": "askcos_retro_template_relevance",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "reaxys",
+      "output_schema": null,
+      "package_dir": "checkpoints/askcos_retro_template_relevance/reaxys",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "reaxys",
+          "model_id": "askcos_retro_template_relevance"
+        }
+      ],
+      "task_direction": "retrosynthesis",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "askcos_reaxys_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "askcos_template_relevance"
+    },
+    {
+      "auxiliary_files": [
+        "aux/MAR-INF/MANIFEST.json",
+        "aux/mar_files/misc.py",
+        "aux/mar_files/models.py",
+        "aux/mar_files/templ_rel_handler.py",
+        "aux/mar_files/templ_rel_parser.py",
+        "aux/mar_files/utils.py",
+        "aux/mar_members.json",
+        "aux/model_args.json",
+        "aux/safetensors_conversion.json",
+        "aux/templates.jsonl",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/ref_routes_reaction_hashes.txt",
+        "benchmarks/native/selected_reactions.csv"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "askcos_paroutes_compatibility_view",
+          "dataset_id": "askcos_retro_template_relevance_reaxys_biocatalysis_paroutes_reference_route_test",
+          "dataset_ref": "paroutes2_reference_route_reactions",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tab_delimited_reactants_reagents_product_rows",
+          "native_source_paths": [
+            "benchmarks/native/selected_reactions.csv",
+            "benchmarks/native/ref_routes_reaction_hashes.txt"
+          ],
+          "notes": [
+            "ASKCOS does not redistribute the native train/validation/test split for this TorchServe archive in datasets/modelos_ASKCOS.",
+            "This bundled benchmark is therefore a public PaRoutes one-step retrosynthesis compatibility view, not a claim about the model's original validation split."
+          ],
+          "num_examples": 56657,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "reaxys_biocatalysis",
+      "input_schema": {
+        "product_field": "smiles",
+        "request_format": {
+          "smiles": [
+            "product_smiles"
+          ]
+        },
+        "target_field": "reactants"
+      },
+      "model_id": "askcos_retro_template_relevance",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "reaxys_biocatalysis",
+      "output_schema": null,
+      "package_dir": "checkpoints/askcos_retro_template_relevance/reaxys_biocatalysis",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "reaxys_biocatalysis",
+          "model_id": "askcos_retro_template_relevance"
+        }
+      ],
+      "task_direction": "retrosynthesis",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "askcos_reaxys_biocatalysis_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "askcos_template_relevance"
+    },
+    {
+      "auxiliary_files": [
+        "aux/MAR-INF/MANIFEST.json",
+        "aux/mar_files/misc.py",
+        "aux/mar_files/models.py",
+        "aux/mar_files/templ_rel_handler.py",
+        "aux/mar_files/templ_rel_parser.py",
+        "aux/mar_files/utils.py",
+        "aux/mar_members.json",
+        "aux/model_args.json",
+        "aux/safetensors_conversion.json",
+        "aux/templates.jsonl",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/ref_routes_reaction_hashes.txt",
+        "benchmarks/native/selected_reactions.csv"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "askcos_paroutes_compatibility_view",
+          "dataset_id": "askcos_retro_template_relevance_uspto_higher_level_paroutes_reference_route_test",
+          "dataset_ref": "paroutes2_reference_route_reactions",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tab_delimited_reactants_reagents_product_rows",
+          "native_source_paths": [
+            "benchmarks/native/selected_reactions.csv",
+            "benchmarks/native/ref_routes_reaction_hashes.txt"
+          ],
+          "notes": [
+            "ASKCOS does not redistribute the native train/validation/test split for this TorchServe archive in datasets/modelos_ASKCOS.",
+            "This bundled benchmark is therefore a public PaRoutes one-step retrosynthesis compatibility view, not a claim about the model's original validation split."
+          ],
+          "num_examples": 56657,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_higher_level",
+      "input_schema": {
+        "product_field": "smiles",
+        "request_format": {
+          "smiles": [
+            "product_smiles"
+          ]
+        },
+        "target_field": "reactants"
+      },
+      "model_id": "askcos_retro_template_relevance",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_higher_level",
+      "output_schema": null,
+      "package_dir": "checkpoints/askcos_retro_template_relevance/uspto_higher_level",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_higher_level",
+          "model_id": "askcos_retro_template_relevance"
+        }
+      ],
+      "task_direction": "retrosynthesis",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "askcos_uspto_higher_level_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "askcos_template_relevance"
+    },
+    {
+      "auxiliary_files": [
+        "aux/bart_vocab.json",
+        "aux/bart_vocab_downstream.json",
+        "aux/ds_config.json",
+        "aux/hparams.yaml",
+        "aux/hyper_parameters.json"
+      ],
+      "checkpoint_id": "combined_large_pre_trained_checkpoint",
+      "input_schema": null,
+      "model_id": "chemformer",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "combined_large_pre_trained_checkpoint",
+      "output_schema": null,
+      "package_dir": "checkpoints/chemformer/combined_large_pre_trained_checkpoint",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "combined_large_pre_trained_checkpoint",
+          "model_id": "chemformer"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "zinc15_100m_chemformer_pretraining"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "chemformer_bart"
+    },
+    {
+      "auxiliary_files": [
+        "aux/bart_vocab.json",
+        "aux/bart_vocab_downstream.json",
+        "aux/ds_config.json",
+        "aux/hparams.yaml",
+        "aux/hyper_parameters.json"
+      ],
+      "checkpoint_id": "combined_pre_trained_base_checkpoint",
+      "input_schema": null,
+      "model_id": "chemformer",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "combined_pre_trained_base_checkpoint",
+      "output_schema": null,
+      "package_dir": "checkpoints/chemformer/combined_pre_trained_base_checkpoint",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "combined_pre_trained_base_checkpoint",
+          "model_id": "chemformer"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "zinc15_100m_chemformer_pretraining"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "chemformer_bart"
+    },
+    {
+      "auxiliary_files": [
+        "aux/bart_vocab.json",
+        "aux/bart_vocab_downstream.json",
+        "aux/ds_config.json",
+        "aux/hparams.yaml",
+        "aux/hyper_parameters.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json"
+      ],
+      "benchmark_datasets": [
+        {
+          "dataset_id": "uspto_50_fine_tuned_retrosynthesis_checkpoint_test",
+          "dataset_ref": "uspto_50k_chemformer_project_release",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "pandas_pickled_dataframe",
+          "native_source_paths": [],
+          "notes": [
+            "Chemformer benchmark records are converted from the public Box release pickle into safe JSONL.",
+            "benchmarks/default_test.jsonl stores product/reactant SMILES converted from the pickled RDKit molecule objects in the test partition."
+          ],
+          "num_examples": 5004,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_50_fine_tuned_retrosynthesis_checkpoint",
+      "input_schema": null,
+      "model_id": "chemformer",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_50_fine_tuned_retrosynthesis_checkpoint",
+      "output_schema": null,
+      "package_dir": "checkpoints/chemformer/uspto_50_fine_tuned_retrosynthesis_checkpoint",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_50_fine_tuned_retrosynthesis_checkpoint",
+          "model_id": "chemformer"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_50k_chemformer_project_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "chemformer_bart"
+    },
+    {
+      "auxiliary_files": [
+        "aux/.gitattributes",
+        "aux/README.md",
+        "aux/UPSTREAM_URL.txt",
+        "aux/config.json",
+        "aux/generation_config.json",
+        "aux/hf_model_info.json",
+        "aux/provenance/README.txt",
+        "aux/provenance/ReactionT5v2_source_README.md",
+        "aux/provenance/datasets/ORD/all_ord_reaction_uniq_with_attr20240506_v3_test.csv",
+        "aux/provenance/datasets/ORD/all_ord_reaction_uniq_with_attr20240506_v3_train.csv",
+        "aux/provenance/datasets/ORD/all_ord_reaction_uniq_with_attr20240506_v3_valid.csv",
+        "aux/provenance/datasets/USPTO_50k/test.csv",
+        "aux/provenance/datasets/USPTO_50k/train.csv",
+        "aux/provenance/datasets/USPTO_50k/val.csv",
+        "aux/provenance/reactiont5v2_split_provenance.json",
+        "aux/special_tokens_map.json",
+        "aux/tokenizer.json",
+        "aux/tokenizer_config.json",
+        "aux/trainer_state.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json"
+      ],
+      "benchmark_datasets": [
+        {
+          "dataset_id": "sagawa_reactiont5v2_retrosynthesis_test",
+          "dataset_ref": "ord_retrosynthesis_pretraining",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "csv_reaction_rows",
+          "native_source_paths": [
+            "aux/provenance/datasets/ORD/all_ord_reaction_uniq_with_attr20240506_v3_test.csv"
+          ],
+          "notes": [
+            "Native benchmark split for the public ORD-trained retrosynthesis checkpoint.",
+            "USPTO_50k/test.csv was used upstream for duplicate-removal / leakage control, not as the native training benchmark for this checkpoint."
+          ],
+          "num_examples": 225282,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "sagawa_reactiont5v2_retrosynthesis",
+      "input_schema": null,
+      "model_id": "reactiont5v2",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "sagawa_reactiont5v2_retrosynthesis",
+      "output_schema": null,
+      "package_dir": "checkpoints/reactiont5v2/sagawa_reactiont5v2_retrosynthesis",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "sagawa_reactiont5v2_retrosynthesis",
+          "model_id": "reactiont5v2"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "ord_retrosynthesis_pretraining"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "seq2seq_t5"
+    },
+    {
+      "auxiliary_files": [
+        "aux/.gitattributes",
+        "aux/README.md",
+        "aux/UPSTREAM_URL.txt",
+        "aux/config.json",
+        "aux/generation_config.json",
+        "aux/hf_model_info.json",
+        "aux/provenance/README.txt",
+        "aux/provenance/ReactionT5v2_source_README.md",
+        "aux/provenance/datasets/ORD/all_ord_reaction_uniq_with_attr20240506_v3_test.csv",
+        "aux/provenance/datasets/ORD/all_ord_reaction_uniq_with_attr20240506_v3_train.csv",
+        "aux/provenance/datasets/ORD/all_ord_reaction_uniq_with_attr20240506_v3_valid.csv",
+        "aux/provenance/datasets/USPTO_50k/test.csv",
+        "aux/provenance/datasets/USPTO_50k/train.csv",
+        "aux/provenance/datasets/USPTO_50k/val.csv",
+        "aux/provenance/reactiont5v2_split_provenance.json",
+        "aux/special_tokens_map.json",
+        "aux/tokenizer.json",
+        "aux/tokenizer_config.json",
+        "aux/trainer_state.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json"
+      ],
+      "benchmark_datasets": [
+        {
+          "dataset_id": "sagawa_reactiont5v2_retrosynthesis_uspto_50k_test",
+          "dataset_ref": "uspto_50k_project_release",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "csv_reaction_rows",
+          "native_source_paths": [
+            "aux/provenance/datasets/USPTO_50k/test.csv"
+          ],
+          "notes": [
+            "Native benchmark split for the public USPTO-50K fine-tuned retrosynthesis checkpoint."
+          ],
+          "num_examples": 5004,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "sagawa_reactiont5v2_retrosynthesis_uspto_50k",
+      "input_schema": null,
+      "model_id": "reactiont5v2",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "sagawa_reactiont5v2_retrosynthesis_uspto_50k",
+      "output_schema": null,
+      "package_dir": "checkpoints/reactiont5v2/sagawa_reactiont5v2_retrosynthesis_uspto_50k",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "sagawa_reactiont5v2_retrosynthesis_uspto_50k",
+          "model_id": "reactiont5v2"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_50k_project_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "seq2seq_t5"
+    },
+    {
+      "auxiliary_files": [
+        "aux/models.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "smiles_transformer/aux/checkpoint_metadata.json",
+        "smiles_transformer/aux/hyper_parameters.json",
+        "smiles_transformer/aux/vocab.txt",
+        "template_localization/aux/checkpoint_metadata.json",
+        "template_localization/aux/hyper_parameters.json",
+        "template_localization/aux/template_lib.json"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "benchmark_pending",
+          "dataset_id": "pistachio_test",
+          "dataset_ref": "pistachio_retrochimera_release",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "benchmark_not_available",
+          "native_source_paths": [],
+          "notes": [
+            "Benchmark companion for the end-to-end RetroChimera hybrid checkpoint on Pistachio.",
+            "These RetroChimera packages bundle both released ensemble components together and are intended for the upstream combined inference API.",
+            "No public benchmark split is bundled yet for this checkpoint; benchmarks/default_test.jsonl is intentionally empty for now.",
+            "The public RetroChimera checkpoint release documents Pistachio-trained checkpoints, but this repository does not yet bundle a public Pistachio evaluation split."
+          ],
+          "num_examples": 0,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "pistachio",
+      "input_schema": null,
+      "model_id": "retrochimera_hybrid",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "pistachio",
+      "output_schema": null,
+      "package_dir": "checkpoints/retrochimera_hybrid/pistachio",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "pistachio",
+          "model_id": "retrochimera_hybrid"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "pistachio_retrochimera_release"
+      ],
+      "weights_path": "template_localization/model.safetensors",
+      "wrapper_family": "retrochimera_hybrid"
+    },
+    {
+      "auxiliary_files": [
+        "aux/models.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/src-test.txt",
+        "benchmarks/native/tgt-test.txt",
+        "smiles_transformer/aux/checkpoint_metadata.json",
+        "smiles_transformer/aux/hyper_parameters.json",
+        "smiles_transformer/aux/vocab.txt",
+        "template_localization/aux/checkpoint_metadata.json",
+        "template_localization/aux/hyper_parameters.json",
+        "template_localization/aux/template_lib.json"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "retrochimera_hybrid_gln_compatibility_view",
+          "dataset_id": "uspto_50k_test",
+          "dataset_ref": "uspto_50k_gln",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tokenized_parallel_txt",
+          "native_source_paths": [
+            "benchmarks/native/src-test.txt",
+            "benchmarks/native/tgt-test.txt"
+          ],
+          "notes": [
+            "Benchmark companion for the end-to-end RetroChimera hybrid checkpoint on USPTO-50K via GLN/Coley split compatibility view.",
+            "These RetroChimera packages bundle both released ensemble components together and are intended for the upstream combined inference API.",
+            "The underlying test split is preserved from the local R-SMILES/GLN-compatible tokenized benchmark files already tracked by this repository.",
+            "The source files preserve rooted augmentation with factor 20; benchmarks/default_test.jsonl collapses each contiguous augmentation block back to one benchmark case per underlying reaction.",
+            "The native source files are tokenized parallel text; benchmarks/default_test.jsonl stores the corresponding de-tokenized product/reactant pairs.",
+            "RetroChimera cites Dai et al./GLN preprocessing for USPTO-50K; GLN in turn follows the Coley/RetroSim 80/10/10 split family."
+          ],
+          "num_examples": 5007,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_50k",
+      "input_schema": null,
+      "model_id": "retrochimera_hybrid",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_50k",
+      "output_schema": null,
+      "package_dir": "checkpoints/retrochimera_hybrid/uspto_50k",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_50k",
+          "model_id": "retrochimera_hybrid"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_50k_gln"
+      ],
+      "weights_path": "template_localization/model.safetensors",
+      "wrapper_family": "retrochimera_hybrid"
+    },
+    {
+      "auxiliary_files": [
+        "aux/models.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/src-test.txt",
+        "benchmarks/native/tgt-test.txt",
+        "smiles_transformer/aux/checkpoint_metadata.json",
+        "smiles_transformer/aux/hyper_parameters.json",
+        "smiles_transformer/aux/vocab.txt",
+        "template_localization/aux/checkpoint_metadata.json",
+        "template_localization/aux/hyper_parameters.json",
+        "template_localization/aux/template_lib.json"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "retrochimera_hybrid_r_smiles_selected_full_view",
+          "dataset_id": "uspto_full_test",
+          "dataset_ref": "uspto_full_gln",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tokenized_parallel_txt",
+          "native_source_paths": [
+            "benchmarks/native/src-test.txt",
+            "benchmarks/native/tgt-test.txt"
+          ],
+          "notes": [
+            "Benchmark companion for the end-to-end RetroChimera hybrid checkpoint on USPTO-FULL via RetroChimera\u2192R-SMILES\u2192GLN compatibility view.",
+            "These RetroChimera packages bundle both released ensemble components together and are intended for the upstream combined inference API.",
+            "The underlying test split is preserved from the local R-SMILES/GLN-compatible tokenized benchmark files already tracked by this repository.",
+            "The source files preserve rooted augmentation with factor 5; benchmarks/default_test.jsonl collapses each contiguous augmentation block back to one benchmark case per underlying reaction.",
+            "The native source files are tokenized parallel text; benchmarks/default_test.jsonl stores the corresponding de-tokenized product/reactant pairs.",
+            "RetroChimera notes that multiple USPTO-FULL variants exist and selects the R-SMILES/Zhong et al. version for fair comparison; the R-SMILES release points that data lineage back to GLN."
+          ],
+          "num_examples": 96023,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_full",
+      "input_schema": null,
+      "model_id": "retrochimera_hybrid",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_full",
+      "output_schema": null,
+      "package_dir": "checkpoints/retrochimera_hybrid/uspto_full",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_full",
+          "model_id": "retrochimera_hybrid"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_full_gln"
+      ],
+      "weights_path": "template_localization/model.safetensors",
+      "wrapper_family": "retrochimera_hybrid"
+    },
+    {
+      "auxiliary_files": [
+        "aux/checkpoint_metadata.json",
+        "aux/hyper_parameters.json",
+        "aux/models.json",
+        "aux/template_lib.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "benchmark_pending",
+          "dataset_id": "pistachio_test",
+          "dataset_ref": "pistachio_retrochimera_release",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "benchmark_not_available",
+          "native_source_paths": [],
+          "notes": [
+            "Benchmark companion for RetroChimera component checkpoint on Pistachio.",
+            "These RetroChimera packages contain component checkpoints rather than the full end-to-end ensemble, so the bundled benchmark is best interpreted as a dataset compatibility view.",
+            "No public benchmark split is bundled yet for this checkpoint; benchmarks/default_test.jsonl is intentionally empty for now.",
+            "The public RetroChimera checkpoint release documents Pistachio-trained checkpoints, but this repository does not yet bundle a public Pistachio evaluation split."
+          ],
+          "num_examples": 0,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "pistachio",
+      "input_schema": null,
+      "model_id": "retrochimera_template",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "pistachio",
+      "output_schema": null,
+      "package_dir": "checkpoints/retrochimera_template/pistachio",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "pistachio",
+          "model_id": "retrochimera_template"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "pistachio_retrochimera_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "retrochimera_template_localization"
+    },
+    {
+      "auxiliary_files": [
+        "aux/checkpoint_metadata.json",
+        "aux/hyper_parameters.json",
+        "aux/models.json",
+        "aux/template_lib.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/src-test.txt",
+        "benchmarks/native/tgt-test.txt"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "retrochimera_component_gln_compatibility_view",
+          "dataset_id": "uspto_50k_test",
+          "dataset_ref": "uspto_50k_gln",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tokenized_parallel_txt",
+          "native_source_paths": [
+            "benchmarks/native/src-test.txt",
+            "benchmarks/native/tgt-test.txt"
+          ],
+          "notes": [
+            "Benchmark companion for RetroChimera component checkpoint on USPTO-50K via GLN/Coley split compatibility view.",
+            "These RetroChimera packages contain component checkpoints rather than the full end-to-end ensemble, so the bundled benchmark is best interpreted as a dataset compatibility view.",
+            "The underlying test split is preserved from the local R-SMILES/GLN-compatible tokenized benchmark files already tracked by this repository.",
+            "The source files preserve rooted augmentation with factor 20; benchmarks/default_test.jsonl collapses each contiguous augmentation block back to one benchmark case per underlying reaction.",
+            "The native source files are tokenized parallel text; benchmarks/default_test.jsonl stores the corresponding de-tokenized product/reactant pairs.",
+            "RetroChimera cites Dai et al./GLN preprocessing for USPTO-50K; GLN in turn follows the Coley/RetroSim 80/10/10 split family."
+          ],
+          "num_examples": 5007,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_50k",
+      "input_schema": null,
+      "model_id": "retrochimera_template",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_50k",
+      "output_schema": null,
+      "package_dir": "checkpoints/retrochimera_template/uspto_50k",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_50k",
+          "model_id": "retrochimera_template"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_50k_gln"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "retrochimera_template_localization"
+    },
+    {
+      "auxiliary_files": [
+        "aux/checkpoint_metadata.json",
+        "aux/hyper_parameters.json",
+        "aux/models.json",
+        "aux/template_lib.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/src-test.txt",
+        "benchmarks/native/tgt-test.txt"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "retrochimera_component_r_smiles_selected_full_view",
+          "dataset_id": "uspto_full_test",
+          "dataset_ref": "uspto_full_gln",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tokenized_parallel_txt",
+          "native_source_paths": [
+            "benchmarks/native/src-test.txt",
+            "benchmarks/native/tgt-test.txt"
+          ],
+          "notes": [
+            "Benchmark companion for RetroChimera component checkpoint on USPTO-FULL via RetroChimera\u2192R-SMILES\u2192GLN compatibility view.",
+            "These RetroChimera packages contain component checkpoints rather than the full end-to-end ensemble, so the bundled benchmark is best interpreted as a dataset compatibility view.",
+            "The underlying test split is preserved from the local R-SMILES/GLN-compatible tokenized benchmark files already tracked by this repository.",
+            "The source files preserve rooted augmentation with factor 5; benchmarks/default_test.jsonl collapses each contiguous augmentation block back to one benchmark case per underlying reaction.",
+            "The native source files are tokenized parallel text; benchmarks/default_test.jsonl stores the corresponding de-tokenized product/reactant pairs.",
+            "RetroChimera notes that multiple USPTO-FULL variants exist and selects the R-SMILES/Zhong et al. version for fair comparison; the R-SMILES release points that data lineage back to GLN."
+          ],
+          "num_examples": 96023,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_full",
+      "input_schema": null,
+      "model_id": "retrochimera_template",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_full",
+      "output_schema": null,
+      "package_dir": "checkpoints/retrochimera_template/uspto_full",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_full",
+          "model_id": "retrochimera_template"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_full_gln"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "retrochimera_template_localization"
+    },
+    {
+      "auxiliary_files": [
+        "aux/checkpoint_metadata.json",
+        "aux/hyper_parameters.json",
+        "aux/models.json",
+        "aux/vocab.txt",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "benchmark_pending",
+          "dataset_id": "pistachio_test",
+          "dataset_ref": "pistachio_retrochimera_release",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "benchmark_not_available",
+          "native_source_paths": [],
+          "notes": [
+            "Benchmark companion for RetroChimera component checkpoint on Pistachio.",
+            "These RetroChimera packages contain component checkpoints rather than the full end-to-end ensemble, so the bundled benchmark is best interpreted as a dataset compatibility view.",
+            "No public benchmark split is bundled yet for this checkpoint; benchmarks/default_test.jsonl is intentionally empty for now.",
+            "The public RetroChimera checkpoint release documents Pistachio-trained checkpoints, but this repository does not yet bundle a public Pistachio evaluation split."
+          ],
+          "num_examples": 0,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "pistachio",
+      "input_schema": null,
+      "model_id": "retrochimera_transformer",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "pistachio",
+      "output_schema": null,
+      "package_dir": "checkpoints/retrochimera_transformer/pistachio",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "pistachio",
+          "model_id": "retrochimera_transformer"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "pistachio_retrochimera_release"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "retrochimera_smiles_transformer"
+    },
+    {
+      "auxiliary_files": [
+        "aux/checkpoint_metadata.json",
+        "aux/hyper_parameters.json",
+        "aux/models.json",
+        "aux/vocab.txt",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/src-test.txt",
+        "benchmarks/native/tgt-test.txt"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "retrochimera_component_gln_compatibility_view",
+          "dataset_id": "uspto_50k_test",
+          "dataset_ref": "uspto_50k_gln",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tokenized_parallel_txt",
+          "native_source_paths": [
+            "benchmarks/native/src-test.txt",
+            "benchmarks/native/tgt-test.txt"
+          ],
+          "notes": [
+            "Benchmark companion for RetroChimera component checkpoint on USPTO-50K via GLN/Coley split compatibility view.",
+            "These RetroChimera packages contain component checkpoints rather than the full end-to-end ensemble, so the bundled benchmark is best interpreted as a dataset compatibility view.",
+            "The underlying test split is preserved from the local R-SMILES/GLN-compatible tokenized benchmark files already tracked by this repository.",
+            "The source files preserve rooted augmentation with factor 20; benchmarks/default_test.jsonl collapses each contiguous augmentation block back to one benchmark case per underlying reaction.",
+            "The native source files are tokenized parallel text; benchmarks/default_test.jsonl stores the corresponding de-tokenized product/reactant pairs.",
+            "RetroChimera cites Dai et al./GLN preprocessing for USPTO-50K; GLN in turn follows the Coley/RetroSim 80/10/10 split family."
+          ],
+          "num_examples": 5007,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_50k",
+      "input_schema": null,
+      "model_id": "retrochimera_transformer",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_50k",
+      "output_schema": null,
+      "package_dir": "checkpoints/retrochimera_transformer/uspto_50k",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_50k",
+          "model_id": "retrochimera_transformer"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_50k_gln"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "retrochimera_smiles_transformer"
+    },
+    {
+      "auxiliary_files": [
+        "aux/checkpoint_metadata.json",
+        "aux/hyper_parameters.json",
+        "aux/models.json",
+        "aux/vocab.txt",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/src-test.txt",
+        "benchmarks/native/tgt-test.txt"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "retrochimera_component_r_smiles_selected_full_view",
+          "dataset_id": "uspto_full_test",
+          "dataset_ref": "uspto_full_gln",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "tokenized_parallel_txt",
+          "native_source_paths": [
+            "benchmarks/native/src-test.txt",
+            "benchmarks/native/tgt-test.txt"
+          ],
+          "notes": [
+            "Benchmark companion for RetroChimera component checkpoint on USPTO-FULL via RetroChimera\u2192R-SMILES\u2192GLN compatibility view.",
+            "These RetroChimera packages contain component checkpoints rather than the full end-to-end ensemble, so the bundled benchmark is best interpreted as a dataset compatibility view.",
+            "The underlying test split is preserved from the local R-SMILES/GLN-compatible tokenized benchmark files already tracked by this repository.",
+            "The source files preserve rooted augmentation with factor 5; benchmarks/default_test.jsonl collapses each contiguous augmentation block back to one benchmark case per underlying reaction.",
+            "The native source files are tokenized parallel text; benchmarks/default_test.jsonl stores the corresponding de-tokenized product/reactant pairs.",
+            "RetroChimera notes that multiple USPTO-FULL variants exist and selects the R-SMILES/Zhong et al. version for fair comparison; the R-SMILES release points that data lineage back to GLN."
+          ],
+          "num_examples": 96023,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "uspto_full",
+      "input_schema": null,
+      "model_id": "retrochimera_transformer",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "uspto_full",
+      "output_schema": null,
+      "package_dir": "checkpoints/retrochimera_transformer/uspto_full",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "uspto_full",
+          "model_id": "retrochimera_transformer"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_full_gln"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "retrochimera_smiles_transformer"
+    },
+    {
+      "auxiliary_files": [
+        "aux/configs/base.yml",
+        "aux/configs/rxngpt_llama1B.json",
+        "aux/deepspeed.yaml",
+        "aux/tokenizer/special_tokens_map.json",
+        "aux/tokenizer/tokenizer.json",
+        "aux/tokenizer/tokenizer_config.json",
+        "aux/vocab.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/USPTO-50k_test_tidy.txt"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "class_unknown_test_aug_best_effort_underlying_test_examples",
+          "dataset_id": "finetune_50k_aug20_pth_test",
+          "dataset_ref": "uspto_50k_rsgpt",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "reaction_smiles_txt",
+          "native_source_paths": [
+            "benchmarks/native/USPTO-50k_test_tidy.txt"
+          ],
+          "notes": [
+            "The exact upstream test-time augmentation prefix file test_50k_prefixs20.txt is not present in the local manual artifacts or pinned source snapshot.",
+            "This bundle therefore preserves the underlying standard USPTO-50K test reactions as the normalized benchmark truth set, while the checkpoint metadata still records the intended aug20 evaluation variant."
+          ],
+          "num_examples": 5007,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "finetune_50k_aug20_pth",
+      "input_schema": null,
+      "model_id": "rsgpt",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "finetune_50k_aug20_pth",
+      "output_schema": null,
+      "package_dir": "checkpoints/rsgpt/finetune_50k_aug20_pth",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "finetune_50k_aug20_pth",
+          "model_id": "rsgpt"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_50k_rsgpt",
+        "synthetic_10b_rsgpt"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "rsgpt_llama_decoder"
+    },
+    {
+      "auxiliary_files": [
+        "aux/configs/base.yml",
+        "aux/configs/rxngpt_llama1B.json",
+        "aux/deepspeed.yaml",
+        "aux/tokenizer/special_tokens_map.json",
+        "aux/tokenizer/tokenizer.json",
+        "aux/tokenizer/tokenizer_config.json",
+        "aux/vocab.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/USPTO-50k_test_tidy_class1.csv"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "class_known",
+          "dataset_id": "finetune_50k_pth_test",
+          "dataset_ref": "uspto_50k_rsgpt",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "class_csv_reaction_smiles",
+          "native_source_paths": [
+            "benchmarks/native/USPTO-50k_test_tidy_class1.csv"
+          ],
+          "notes": [
+            "This is the class-labeled USPTO-50K benchmark companion file corresponding to the best-effort class-known mapping for finetune_50k_pth."
+          ],
+          "num_examples": 5007,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "finetune_50k_pth",
+      "input_schema": null,
+      "model_id": "rsgpt",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "finetune_50k_pth",
+      "output_schema": null,
+      "package_dir": "checkpoints/rsgpt/finetune_50k_pth",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "finetune_50k_pth",
+          "model_id": "rsgpt"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_50k_rsgpt",
+        "synthetic_10b_rsgpt"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "rsgpt_llama_decoder"
+    },
+    {
+      "auxiliary_files": [
+        "aux/configs/base.yml",
+        "aux/configs/rxngpt_llama1B.json",
+        "aux/deepspeed.yaml",
+        "aux/tokenizer/special_tokens_map.json",
+        "aux/tokenizer/tokenizer.json",
+        "aux/tokenizer/tokenizer_config.json",
+        "aux/vocab.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/USPTO-FULL_test_tidy.txt"
+      ],
+      "benchmark_datasets": [
+        {
+          "dataset_id": "finetune_full_pth_test",
+          "dataset_ref": "uspto_full_rsgpt",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "reaction_smiles_txt",
+          "native_source_paths": [
+            "benchmarks/native/USPTO-FULL_test_tidy.txt"
+          ],
+          "notes": [
+            "Native RSGPT benchmark source extracted from USPTO_data/USPTO-FULL_test_tidy.txt."
+          ],
+          "num_examples": 95389,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "finetune_full_pth",
+      "input_schema": null,
+      "model_id": "rsgpt",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "finetune_full_pth",
+      "output_schema": null,
+      "package_dir": "checkpoints/rsgpt/finetune_full_pth",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "finetune_full_pth",
+          "model_id": "rsgpt"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_full_rsgpt",
+        "synthetic_10b_rsgpt"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "rsgpt_llama_decoder"
+    },
+    {
+      "auxiliary_files": [
+        "aux/configs/base.yml",
+        "aux/configs/rxngpt_llama1B.json",
+        "aux/deepspeed.yaml",
+        "aux/tokenizer/special_tokens_map.json",
+        "aux/tokenizer/tokenizer.json",
+        "aux/tokenizer/tokenizer_config.json",
+        "aux/vocab.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/USPTO-MIT_test_tidy.txt"
+      ],
+      "benchmark_datasets": [
+        {
+          "dataset_id": "finetune_mit_pth_test",
+          "dataset_ref": "uspto_mit_rsgpt",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "reaction_smiles_txt",
+          "native_source_paths": [
+            "benchmarks/native/USPTO-MIT_test_tidy.txt"
+          ],
+          "notes": [
+            "Native RSGPT benchmark source extracted from USPTO_data/USPTO-MIT_test_tidy.txt."
+          ],
+          "num_examples": 40000,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "finetune_mit_pth",
+      "input_schema": null,
+      "model_id": "rsgpt",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "finetune_mit_pth",
+      "output_schema": null,
+      "package_dir": "checkpoints/rsgpt/finetune_mit_pth",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "finetune_mit_pth",
+          "model_id": "rsgpt"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_mit_rsgpt",
+        "synthetic_10b_rsgpt"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "rsgpt_llama_decoder"
+    },
+    {
+      "auxiliary_files": [
+        "aux/configs/base.yml",
+        "aux/configs/rxngpt_llama1B.json",
+        "aux/deepspeed.yaml",
+        "aux/tokenizer/special_tokens_map.json",
+        "aux/tokenizer/tokenizer.json",
+        "aux/tokenizer/tokenizer_config.json",
+        "aux/vocab.json",
+        "benchmarks/default_test.jsonl",
+        "benchmarks/metadata.json",
+        "benchmarks/native/USPTO-50k_test_tidy.txt"
+      ],
+      "benchmark_datasets": [
+        {
+          "benchmark_variant": "class_unknown_no_pretraining_ablation",
+          "dataset_id": "nopretrain_50k_pth_test",
+          "dataset_ref": "uspto_50k_rsgpt",
+          "default": true,
+          "format": "jsonl",
+          "metadata_path": "benchmarks/metadata.json",
+          "native_input_format": "reaction_smiles_txt",
+          "native_source_paths": [
+            "benchmarks/native/USPTO-50k_test_tidy.txt"
+          ],
+          "notes": [
+            "Native RSGPT benchmark source extracted from USPTO_data/USPTO-50k_test_tidy.txt."
+          ],
+          "num_examples": 5007,
+          "path": "benchmarks/default_test.jsonl",
+          "split": "test"
+        }
+      ],
+      "checkpoint_id": "nopretrain_50k_pth",
+      "input_schema": null,
+      "model_id": "rsgpt",
+      "model_role": "backward",
+      "normalized_checkpoint_id": "nopretrain_50k_pth",
+      "output_schema": null,
+      "package_dir": "checkpoints/rsgpt/nopretrain_50k_pth",
+      "requires_atom_mapping": null,
+      "score_direction": null,
+      "source_registry_refs": [
+        {
+          "checkpoint_id": "nopretrain_50k_pth",
+          "model_id": "rsgpt"
+        }
+      ],
+      "task_direction": "backward",
+      "taxonomy": null,
+      "taxonomy_version": null,
+      "training_dataset_refs": [
+        "uspto_50k_rsgpt"
+      ],
+      "weights_path": "model.safetensors",
+      "wrapper_family": "rsgpt_llama_decoder"
+    }
+  ],
+  "generated_at": "2026-07-20T13:51:14.589861+00:00"
+}