Discover QM9 and GEOM epoch 100; order model tables QM9, GEOM-Drugs, ChEMBL

README.md

@@ -4,16 +4,17 @@ tags:
   - chemistry
   - molecular-generation
   - flow-matching
-  - chembl3d
+  - qm9
   - geom-drugs
+  - chembl3d
   - 3d
   - arxiv:2606.07239
 ---
 
 # Morph: 3D molecular generation
 
-Unconditional molecular generation checkpoints trained on **ChEMBL3D** and
-**GEOM-Drugs**, with the configurations, vocabularies and training distributions
+Unconditional molecular generation checkpoints trained on **QM9**, **GEOM-Drugs**
+and **ChEMBL3D**, with the configurations, vocabularies and training distributions
 needed for inference. No training dataset is required to load a model or generate
 molecules.
 
@@ -23,9 +24,10 @@ See our [paper](https://arxiv.org/abs/2606.07239), *Generative Molecular Morphin
 
 | Dataset | Model name | Hydrogen representation | Completed epochs | Files |
 | --- | --- | --- | --- | --- |
+| QM9 | `qm9-explicit-h` | Explicit H atoms | 1,000, 1,500 | [Bundle](https://huggingface.co/maltefranke/morph/tree/main/qm9/explicit_h) |
+| GEOM-Drugs | `geom-explicit-h` | Explicit H atoms | 50, 100 | [Bundle](https://huggingface.co/maltefranke/morph/tree/main/geom/explicit_h) |
 | ChEMBL3D | `chembl3d-explicit-h` | Explicit H atoms | 200, 300, 400 | [Bundle](https://huggingface.co/maltefranke/morph/tree/main/chembl3d/explicit_h) |
 | ChEMBL3D | `chembl3d-implicit-h` | Heavy atoms; implicit H | 100, 200, 300 | [Bundle](https://huggingface.co/maltefranke/morph/tree/main/chembl3d/implicit_h) |
-| GEOM-Drugs | `geom-explicit-h` | Explicit H atoms | 50 | [Bundle](https://huggingface.co/maltefranke/morph/tree/main/geom/explicit_h) |
 
 Epoch numbers count **completed training epochs**: `epoch_050.ckpt` means 50
 completed epochs and stores the zero-based Lightning counter `epoch=49`.
@@ -64,7 +66,7 @@ generation inside a compute allocation.
 ```python
 from morph.inference import load_pretrained
 
-model = load_pretrained("geom-explicit-h", epoch=50, device="cuda")
+model = load_pretrained("geom-explicit-h", epoch=100, device="cuda")
 
 trajectories, validity, vocab, coordinate_std = model.generate(
     n_mols=64,
@@ -74,9 +76,11 @@ trajectories, validity, vocab, coordinate_std = model.generate(
 )
 ```
 
-Use `device="cpu"` for CPU inference. The same interface loads ChEMBL3D:
+Use `device="cpu"` for CPU inference. The same interface loads QM9 and ChEMBL3D:
 
 ```python
+model = load_pretrained("qm9-explicit-h", epoch=1500, device="cuda")
+# Use epoch=1000 for the earlier QM9 checkpoint.
 model = load_pretrained("chembl3d-explicit-h", epoch=400, device="cuda")
 # Or choose the heavy-atom model:
 model = load_pretrained("chembl3d-implicit-h", epoch=300, device="cuda")
@@ -102,8 +106,8 @@ multiply them by `coordinate_std` again.
 Generate GEOM-Drugs molecules and save valid structures to SDF:
 
 ```bash
-morph generate --model geom-explicit-h --epoch 50 \
-  --device cuda --batch 64 --batch-size 16 --output geom-e50.sdf
+morph generate --model geom-explicit-h --epoch 100 \
+  --device cuda --batch 64 --batch-size 16 --output geom-e100.sdf
 ```
 
 `--batch` is the total number of generation attempts. Add `--save-trajs` to save
@@ -113,7 +117,7 @@ weights, and `--steps` and `--seed` control sampling.
 Download a portable bundle without loading the model:
 
 ```bash
-morph models download geom-explicit-h --epoch 50 --local-dir morph-models
+morph models download geom-explicit-h --epoch 100 --local-dir morph-models
 ```
 
 Load that local bundle without network access or catalog lookup:
@@ -121,12 +125,12 @@ Load that local bundle without network access or catalog lookup:
 ```python
 model = load_pretrained(
     weights_dir="morph-models/geom/explicit_h",
-    epoch=50,
+    epoch=100,
     device="cuda",
 )
 ```
 
-Use another model name from the table for ChEMBL3D. The interface also accepts
+Use another model name from the table for QM9 or ChEMBL3D. The interface also accepts
 direct Hub repository IDs with `subfolder` and `revision`. Pass `cache_dir` to
 choose a cache location, or `local_files_only=True` to load a previously cached
 release without network access. The corresponding CLI options are `--cache-dir`
@@ -134,7 +138,7 @@ and `--offline`.
 
 The root `sample.py` remains available as the archived ChEMBL3D sampling entry
 point with its corresponding bundled runtimes. The shared interface above
-supports all three published model variants.
+supports all four published model variants.
 
 ## Reproducibility
 
@@ -144,10 +148,10 @@ The bundles live alongside the corresponding model files on Hugging Face:
 
 | Model | Configuration and source |
 | --- | --- |
+| QM9, explicit H | [Training bundle](https://huggingface.co/maltefranke/morph/resolve/main/qm9/explicit_h/training_bundle.tar.gz) |
+| GEOM-Drugs, explicit H | [Training bundle](https://huggingface.co/maltefranke/morph/resolve/main/geom/explicit_h/training_bundle.tar.gz) |
 | ChEMBL3D, explicit H | [Training bundle](https://huggingface.co/maltefranke/morph/resolve/main/chembl3d/explicit_h/training_bundle.tar.gz) |
 | ChEMBL3D, implicit H | [Training bundle](https://huggingface.co/maltefranke/morph/resolve/main/chembl3d/implicit_h/training_bundle.tar.gz) |
-| GEOM-Drugs, explicit H | [Training bundle](https://huggingface.co/maltefranke/morph/resolve/main/geom/explicit_h/training_bundle.tar.gz) |
-| QM9, explicit H (no published weights yet) | [Training bundle](https://huggingface.co/maltefranke/morph/resolve/main/qm9/explicit_h/training_bundle.tar.gz) |
 
 For example, download and unpack the GEOM training bundle:
 
@@ -165,8 +169,9 @@ are recorded in the release metadata and bundle manifests.
 
 | Dataset | Training filter | Representation |
 | --- | --- | --- |
-| ChEMBL3D | At most 100 total atoms, including H before hydrogen removal | Explicit-H and implicit-H variants |
+| QM9 | Original QM9 split | Explicit-H variant |
 | GEOM-Drugs | At most 72 total atoms, including H | Explicit-H variant |
+| ChEMBL3D | At most 100 total atoms, including H before hydrogen removal | Explicit-H and implicit-H variants |
 
 Both ChEMBL3D variants use the same split membership: 1,474,008 training,
 184,295 validation and 184,242 test molecules. The implicit-H model predicts
@@ -174,8 +179,11 @@ heavy atoms, bonds and formal charges; RDKit infers hydrogens during conversion
 and validity checking. Its geometry statistics and coordinate normalization were
 recomputed for the heavy-atom representation.
 
-The GEOM-Drugs epoch 50 checkpoint was obtained by resuming the best checkpoint
-after 48 completed epochs. The resumed run validates every epoch and retains
+QM9 uses 100,000 training, 20,000 validation and 12,947 test molecules. Its run
+started from scratch, validates every epoch and retains checkpoints every 25 epochs.
+
+The GEOM-Drugs epoch 50 and 100 checkpoints come from the run resumed from the
+best checkpoint after 48 completed epochs. The resumed run validates every epoch and retains
 checkpoints every fifth epoch. Only the epochs listed in the catalog have been
 published here.
 
@@ -204,11 +212,13 @@ Each model directory contains:
 
 The published bundles were checked for strict current/EMA weight loading and
 end-to-end sampling. These are functionality checks, not validity benchmarks.
-The GEOM release also passed remote file-size and hash verification.
+The GEOM and QM9 releases also passed remote file-size and hash verification.
 
 The [ChEMBL3D release manifest](https://huggingface.co/maltefranke/morph/blob/main/manifest.json)
-describes the original ChEMBL3D release, including its historical model card. GEOM has its own
-[bundle manifest](https://huggingface.co/maltefranke/morph/blob/main/geom/explicit_h/manifest.json).
+describes the original ChEMBL3D release, including its historical model card.
+[QM9](https://huggingface.co/maltefranke/morph/blob/main/qm9/explicit_h/manifest.json) and
+[GEOM-Drugs](https://huggingface.co/maltefranke/morph/blob/main/geom/explicit_h/manifest.json)
+have their own bundle manifests.
 Original training configs contain historical cluster paths; use `config.yaml`
 through the inference loader for portable generation.
 

catalog.json

@@ -2,6 +2,30 @@
   "schema_version": 1,
   "default_model": "chembl3d-explicit-h",
   "models": [
+    {
+      "name": "qm9-explicit-h",
+      "repo_id": "maltefranke/morph",
+      "subfolder": "qm9/explicit_h",
+      "revision": "cd010661227ece54e197343854330d2871138e18",
+      "dataset": "qm9",
+      "variant": "explicit_h",
+      "epochs": [
+        1000,
+        1500
+      ]
+    },
+    {
+      "name": "geom-explicit-h",
+      "repo_id": "maltefranke/morph",
+      "subfolder": "geom/explicit_h",
+      "revision": "cd010661227ece54e197343854330d2871138e18",
+      "dataset": "geom",
+      "variant": "explicit_h",
+      "epochs": [
+        50,
+        100
+      ]
+    },
     {
       "name": "chembl3d-explicit-h",
       "repo_id": "maltefranke/morph",
@@ -27,17 +51,6 @@
         200,
         300
       ]
-    },
-    {
-      "name": "geom-explicit-h",
-      "repo_id": "maltefranke/morph",
-      "subfolder": "geom/explicit_h",
-      "revision": "b839af108eaade757a5d2d668d05498e8c733c1e",
-      "dataset": "geom",
-      "variant": "explicit_h",
-      "epochs": [
-        50
-      ]
     }
   ]
 }

geom/explicit_h/training_bundle.json

@@ -2,7 +2,8 @@
   "model": "geom-explicit-h",
   "source_snapshot": "geom_resume_e48_20261008",
   "published_completed_epochs": [
-    50
+    50,
+    100
   ],
   "original_resume_completed_epoch": 48,
   "original_validation_every_epochs": 1,
@@ -36,9 +37,10 @@
   "checkpoint_release": {
     "repo_id": "maltefranke/morph",
     "subfolder": "geom/explicit_h",
-    "revision": "b839af108eaade757a5d2d668d05498e8c733c1e",
+    "revision": "cd010661227ece54e197343854330d2871138e18",
     "epochs": [
-      50
+      50,
+      100
     ]
   },
   "files": {

qm9/explicit_h/training_bundle.json

@@ -1,7 +1,10 @@
 {
   "model": "qm9-explicit-h",
   "source_snapshot": "qm9_launch_20261008",
-  "published_completed_epochs": [],
+  "published_completed_epochs": [
+    1000,
+    1500
+  ],
   "original_resume_completed_epoch": null,
   "original_validation_every_epochs": 1,
   "reproduction_validation_every_epochs": 1,
@@ -31,7 +34,15 @@
     "revision": "d0332f1c022345956470bb948ada79e5702ed8dc",
     "filename": "reproduction/qm9-explicit-h.tar.gz"
   },
-  "checkpoint_release": null,
+  "checkpoint_release": {
+    "repo_id": "maltefranke/morph",
+    "subfolder": "qm9/explicit_h",
+    "revision": "cd010661227ece54e197343854330d2871138e18",
+    "epochs": [
+      1000,
+      1500
+    ]
+  },
   "files": {
     "original_config": "reproduction/original_training_config.yaml",
     "portable_config": "configs/reproduce.yaml",

reproduction/README.md

@@ -49,10 +49,10 @@ is the model name, also recorded as `directory` in `training_bundle.json`.
 
 | Model folder | Extracted directory |
 | --- | --- |
+| `qm9/explicit_h` | `qm9-explicit-h` |
+| `geom/explicit_h` | `geom-explicit-h` |
 | `chembl3d/explicit_h` | `chembl3d-explicit-h` |
 | `chembl3d/implicit_h` | `chembl3d-implicit-h` |
-| `geom/explicit_h` | `geom-explicit-h` |
-| `qm9/explicit_h` | `qm9-explicit-h` |
 
 ## Bundle contents
 
@@ -85,10 +85,10 @@ directory for each run, and retain these records alongside checkpoints.
 
 | Model | Batch per GPU | Accumulation | Effective batch | Original validation cadence | Portable validation / save cadence |
 | --- | ---: | ---: | ---: | --- | --- |
+| `qm9-explicit-h` | 1024 | 2 | 4096 | Every epoch | Every 1 / 25 epochs |
+| `geom-explicit-h` | 320 | 1 | 640 | Every epoch | Every 1 / 5 epochs |
 | `chembl3d-explicit-h` | 320 | 1 | 640 | Every 5 epochs | Every 1 / 5 epochs |
 | `chembl3d-implicit-h` | 320 | 1 | 640 | Every epoch | Every 1 / 5 epochs |
-| `geom-explicit-h` | 320 | 1 | 640 | Every epoch | Every 1 / 5 epochs |
-| `qm9-explicit-h` | 1024 | 2 | 4096 | Every epoch | Every 1 / 25 epochs |
 
 All recipes use two RTX 4090 GPUs, seed 42, bf16 mixed precision, SDPA attention,
 compiled networks and 4096 generation attempts per validation epoch. GEOM uses
@@ -98,7 +98,9 @@ Environment flags are explicit in each archive's `reproduction/environment.sh`.
 
 The portable recipes stop at 400 completed epochs for explicit-H ChEMBL3D,
 300 for implicit-H ChEMBL3D, 50 for GEOM, or 3000 for QM9. Override that limit
-with `bash train.sh trainer.trainer.max_epochs=100`.
+with `bash train.sh trainer.trainer.max_epochs=100` to train GEOM through
+100 completed epochs. For QM9, use `trainer.trainer.max_epochs=1000` or
+`trainer.trainer.max_epochs=1500` to match the published epoch counts.
 
 The portable configs preserve the saved model, optimizer, schedule and batch
 settings. They change data/output paths, disable W&B, remove the historical
@@ -112,10 +114,10 @@ Set `MORPH_DATASET_ROOT` to a dataset directory containing `processed/`:
 
 | Archive | Dataset directory | Selection / representation |
 | --- | --- | --- |
+| QM9 explicit H | `qm9` | Original QM9 split, explicit H |
+| GEOM explicit H | `geom` | At most 72 total atoms, explicit H |
 | ChEMBL3D explicit H | `chembl3d_max100` | At most 100 total atoms, explicit H |
 | ChEMBL3D implicit H | `chembl3d_max100_implicit_h` | Same source molecules, then remove H |
-| GEOM explicit H | `geom` | At most 72 total atoms, explicit H |
-| QM9 explicit H | `qm9` | Original QM9 split, explicit H |
 
 The ChEMBL3D cap counts hydrogens **before removal**. Both variants have
 1,474,008 / 184,295 / 184,242 train/validation/test molecules. The implicit-H cache
@@ -140,12 +142,12 @@ a CPU worker if you need a full content manifest for future comparisons.
 
 ## Fresh training versus resuming
 
-The explicit-H ChEMBL3D release resumed after 5 completed epochs. GEOM epoch 50
-resumed from the best checkpoint after 48 completed epochs. Starting these
+The explicit-H ChEMBL3D release resumed after 5 completed epochs. GEOM epochs
+50 and 100 come from the run resumed from the best checkpoint after 48 completed epochs. Starting these
 recipes from seed 42 reproduces their setup but does not replay that history.
 The intermediate epoch-5 and epoch-48 checkpoints are not in the Hub release.
-The implicit-H ChEMBL3D run and the QM9 recipe start fresh. QM9 has no published
-checkpoint in this release.
+The implicit-H ChEMBL3D and QM9 runs started fresh. QM9 checkpoints after
+1,000 and 1,500 completed epochs are published alongside its training bundle.
 
 To continue from an available full training checkpoint, pass its absolute path
 and the desired **total** completed epoch count to the matching archive: