Unable to initiate OCPCalculator using the eqV2_dens_31M_mp.pt checkpoint

[jinlhr542]

Python version

3.11

fairchem-core version

1.3.0

pytorch version

2.5.1

cuda version

used cpu

Operating system version

No response

Minimal example

No response

Current behavior

from fairchem.core import OCPCalculator
from pymatgen.core.structure import Structure
lattice = [[5.0, 0.0, 0.0], [0.0, 5.0, 0.0], [0.0, 0.0, 5.0]]
atoms = [("Si", [0, 0, 0]), ("Si", [1.5, 1.5, 1.5])]
atoms = Structure(lattice, [atom[0] for atom in atoms], [atom[1] for atom in atoms]).to_ase_atoms()
calc = OCPCalculator(checkpoint_path="eqV2_dens_31M_mp.pt", local_cache="pretrained_models", cpu = True)

---

TypeError Traceback (most recent call last)
Cell In[4], line 1
----> 1 calc = OCPCalculator(checkpoint_path="eqV2_dens_31M_mp.pt", local_cache="pretrained_models", cpu = True)
2 #atoms.calc = calc

File /opt/anaconda3/envs/3.11/lib/python3.11/site-packages/fairchem/core/common/relaxation/ase_utils.py:212, in OCPCalculator.init(self, config_yml, checkpoint_path, model_name, local_cache, trainer, cpu, seed)
209 self.config["checkpoint"] = str(checkpoint_path)
210 del config["dataset"]["src"]
--> 212 self.trainer = registry.get_trainer_class(config["trainer"])(
213 task=config.get("task", {}),
214 model=config["model"],
215 dataset=[config["dataset"]],
216 outputs=config["outputs"],
217 loss_functions=config["loss_functions"],
218 evaluation_metrics=config["evaluation_metrics"],
219 optimizer=config["optim"],
220 identifier="",
221 slurm=config.get("slurm", {}),
222 local_rank=config.get("local_rank", 0),
223 is_debug=config.get("is_debug", True),
224 cpu=cpu,
225 amp=config.get("amp", False),
226 inference_only=True,
227 )
229 if checkpoint_path is not None:
230 self.load_checkpoint(checkpoint_path=checkpoint_path, checkpoint=checkpoint)

File /opt/anaconda3/envs/3.11/lib/python3.11/site-packages/fairchem/core/trainers/ocp_trainer.py:109, in OCPTrainer.init(self, task, model, outputs, dataset, optimizer, loss_functions, evaluation_metrics, identifier, local_rank, timestamp_id, run_dir, is_debug, print_every, seed, logger, amp, cpu, name, slurm, gp_gpus, inference_only)
107 if slurm is None:
108 slurm = {}
--> 109 super().init(
110 task=task,
111 model=model,
112 outputs=outputs,
113 dataset=dataset,
114 optimizer=optimizer,
115 loss_functions=loss_functions,
116 evaluation_metrics=evaluation_metrics,
117 identifier=identifier,
118 local_rank=local_rank,
119 timestamp_id=timestamp_id,
120 run_dir=run_dir,
121 is_debug=is_debug,
122 print_every=print_every,
123 seed=seed,
124 logger=logger,
125 amp=amp,
126 cpu=cpu,
127 slurm=slurm,
128 name=name,
129 gp_gpus=gp_gpus,
130 inference_only=inference_only,
131 )

File /opt/anaconda3/envs/3.11/lib/python3.11/site-packages/fairchem/core/trainers/base_trainer.py:220, in BaseTrainer.init(self, task, model, outputs, dataset, optimizer, loss_functions, evaluation_metrics, identifier, local_rank, timestamp_id, run_dir, is_debug, print_every, seed, logger, amp, cpu, name, slurm, gp_gpus, inference_only)
217 self.primary_metric = None
218 self.ema = None
--> 220 self.load(inference_only)

File /opt/anaconda3/envs/3.11/lib/python3.11/site-packages/fairchem/core/trainers/base_trainer.py:243, in BaseTrainer.load(self, inference_only)
241 self.load_logger()
242 self.load_task()
--> 243 self.load_model()
245 if inference_only is False:
246 self.load_datasets()

File /opt/anaconda3/envs/3.11/lib/python3.11/site-packages/fairchem/core/trainers/base_trainer.py:558, in BaseTrainer.load_model(self)
556 model_config_copy = copy.deepcopy(self.config["model"])
557 model_name = model_config_copy.pop("name")
--> 558 self.model = registry.get_model_class(model_name)(
559 **model_config_copy,
560 ).to(self.device)
562 num_params = sum(p.numel() for p in self.model.parameters())
564 if distutils.is_master():

File /opt/anaconda3/envs/3.11/lib/python3.11/site-packages/fairchem/core/models/base.py:267, in HydraModel.init(self, backbone, heads, finetune_config, otf_graph, pass_through_head_outputs, freeze_backbone)
265 backbone = copy.deepcopy(backbone)
266 backbone_model_name = backbone.pop("model")
--> 267 self.backbone: BackboneInterface = registry.get_model_class(
268 backbone_model_name
269 )(
270 **backbone,
271 )
272 elif starting_model is not None:
273 self.backbone = starting_model.backbone

TypeError: EqV2DeNSBackbone.init() got an unexpected keyword argument 'use_denoising_energy'

Expected Behavior

https://huggingface.co/fairchem/OMAT24 suggests to use fairchem-core 1.2.1 version. However this has been decrepted.

Relevant files to reproduce this bug

No response

[lbluque]

Hi @jinlhr542,

Thanks for reporting this! The checkpoint should work with the omat24 branch, in case you want to use it now.

I will push updated checkpoints that work with the newest version (1.3.0) of FAIRChem too.

[jinlhr542]

Hi @jinlhr542,

Thanks for reporting this! The checkpoint should work with the omat24 branch, in case you want to use it now.

I will push updated checkpoints that work with the newest version (1.3.0) of FAIRChem too.

Thank you

[jinlhr542]

Hi @jinlhr542,

Thanks for reporting this! The checkpoint should work with the omat24 branch, in case you want to use it now.

I will push updated checkpoints that work with the newest version (1.3.0) of FAIRChem too.

Dear Luis

I failed to create the conda environment using the env.cpu.yml file in the omat24 branch:

(3.12) jason@YaoshuXiedeMacBook-Pro packages % conda env create -f env.cpu.yml
Channels:

pytorch
conda-forge
defaults
Platform: osx-arm64
Collecting package metadata (repodata.json): done
Solving environment: done
Downloading and Extracting Packages:

Preparing transaction: done
Verifying transaction: done
Executing transaction: done
Installing pip dependencies: - Ran pip subprocess with arguments:
['/opt/anaconda3/envs/fair-chem/bin/python', '-m', 'pip', 'install', '-U', '-r', '/Users/jason/Documents/GitHub/mlipdockers/docker_image_build/eqV2/fairchem-omat24/packages/condaenv.k5z_6me_.requirements.txt', '--exists-action=b']
Pip subprocess output:
Looking in links: https://data.pyg.org/whl/torch-2.4.0+cu121.html

Pip subprocess error:
ERROR: Could not find a version that satisfies the requirement torch_cluster==1.6.3+pt24cpu (from versions: 0.1.1, 0.2.3, 0.2.4, 1.0.1, 1.0.3, 1.1.1, 1.1.2, 1.1.3, 1.1.4, 1.1.5, 1.2.1, 1.2.2, 1.2.3, 1.2.4, 1.3.0, 1.4.0, 1.4.1, 1.4.2, 1.4.3a1, 1.4.3, 1.4.4, 1.4.5, 1.5.2, 1.5.3, 1.5.4, 1.5.5, 1.5.6, 1.5.7, 1.5.8, 1.5.9, 1.6.0, 1.6.1, 1.6.2, 1.6.3)
ERROR: No matching distribution found for torch_cluster==1.6.3+pt24cpu

failed

CondaEnvException: Pip failed

[lbluque]

Hi @jinlhr542,

Did you run into this issue before? or only when trying to install a version to use the omat24 branch?

In any case, I have just pushed checkpoints that are compatible with fairchem-core >= 1.3.0, so there is no need to use the omat24 branch anymore. Could you try re-dowloading the checkpoint and running your original code snippet?

[jinlhr542]

Hi @jinlhr542,

Did you run into this issue before? or only when trying to install a version to use the omat24 branch?

In any case, I have just pushed checkpoints that are compatible with fairchem-core >= 1.3.0, so there is no need to use the omat24 branch anymore. Could you try re-dowloading the checkpoint and running your original code snippet?

I have tried to install the current 1.3.0 fairchem-core following the instruction:
wget https://raw.githubusercontent.com/FAIR-Chem/fairchem/main/packages/env.cpu.yml
conda env create -f env.cpu.yml,
and get the error:

Channels:
 - pytorch
 - conda-forge
 - defaults
Platform: osx-arm64
Collecting package metadata (repodata.json): done
Solving environment: done

Downloading and Extracting Packages:
                                                                                
Preparing transaction: done                                                     
Verifying transaction: done                                                     
Executing transaction: done                                                     
Installing pip dependencies: \ Ran pip subprocess with arguments:               
['/opt/anaconda3/envs/fair-chem/bin/python', '-m', 'pip', 'install', '-U', '-r', '/Users/jason/condaenv.j_3tkbpv.requirements.txt', '--exists-action=b']        
Pip subprocess output:
Looking in links: https://data.pyg.org/whl/torch-2.4.0+cu121.html

Pip subprocess error:
ERROR: Could not find a version that satisfies the requirement torch_cluster==1.6.3+pt24cpu (from versions: 0.1.1, 0.2.3, 0.2.4, 1.0.1, 1.0.3, 1.1.1, 1.1.2, 1.1.3, 1.1.4, 1.1.5, 1.2.1, 1.2.2, 1.2.3, 1.2.4, 1.3.0, 1.4.0, 1.4.1, 1.4.2, 1.4.3a1, 1.4.3, 1.4.4, 1.4.5, 1.5.2, 1.5.3, 1.5.4, 1.5.5, 1.5.6, 1.5.7, 1.5.8, 1.5.9, 1.6.0, 1.6.1, 1.6.2, 1.6.3)
ERROR: No matching distribution found for torch_cluster==1.6.3+pt24cpu

failed

CondaEnvException: Pip failed

I am afraid that there are some errors in the env.cpu.yml. Could you please check it?

[jinlhr542]

Hi @jinlhr542,

Did you run into this issue before? or only when trying to install a version to use the omat24 branch?

In any case, I have just pushed checkpoints that are compatible with fairchem-core >= 1.3.0, so there is no need to use the omat24 branch anymore. Could you try re-dowloading the checkpoint and running your original code snippet?

Dear Luis

I have successfully installed the 1.3.0 version using a docker container. As you mentioned that 1.3.0 should support eqV2 Dense model, I tried OCPCalculator(checkpoint_path="eqV2_dens_31M_mp.pt") and get the error:

2024-12-20 15:09:20 Traceback (most recent call last):
2024-12-20 15:09:20   File "/opt/conda/lib/python3.11/site-packages/flask/app.py", line 1511, in wsgi_app
2024-12-20 15:09:20     response = self.full_dispatch_request()
2024-12-20 15:09:20                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-12-20 15:09:20   File "/opt/conda/lib/python3.11/site-packages/flask/app.py", line 919, in full_dispatch_request
2024-12-20 15:09:20     rv = self.handle_user_exception(e)
2024-12-20 15:09:20          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-12-20 15:09:20   File "/opt/conda/lib/python3.11/site-packages/flask/app.py", line 917, in full_dispatch_request
2024-12-20 15:09:20     rv = self.dispatch_request()
2024-12-20 15:09:20          ^^^^^^^^^^^^^^^^^^^^^^^
2024-12-20 15:09:20   File "/opt/conda/lib/python3.11/site-packages/flask/app.py", line 902, in dispatch_request
2024-12-20 15:09:20     return self.ensure_sync(self.view_functions[rule.endpoint])(**view_args)  # type: ignore[no-any-return]
2024-12-20 15:09:20            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-12-20 15:09:20   File "/app/server.py", line 12, in predict_energy
2024-12-20 15:09:20     calc = OCPCalculator(checkpoint_path="eqV2_dens_31M_mp.pt")
2024-12-20 15:09:20            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-12-20 15:09:20   File "/opt/conda/lib/python3.11/site-packages/fairchem/core/common/relaxation/ase_utils.py", line 227, in __init__
2024-12-20 15:09:20     self.trainer = registry.get_trainer_class(config["trainer"])(
2024-12-20 15:09:20                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-12-20 15:09:20   File "/opt/conda/lib/python3.11/site-packages/fairchem/core/trainers/ocp_trainer.py", line 109, in __init__
2024-12-20 15:09:20     super().__init__(
2024-12-20 15:09:20   File "/opt/conda/lib/python3.11/site-packages/fairchem/core/trainers/base_trainer.py", line 220, in __init__
2024-12-20 15:09:20     self.load(inference_only)
2024-12-20 15:09:20   File "/opt/conda/lib/python3.11/site-packages/fairchem/core/trainers/base_trainer.py", line 243, in load
2024-12-20 15:09:20     self.load_model()
2024-12-20 15:09:20   File "/opt/conda/lib/python3.11/site-packages/fairchem/core/trainers/base_trainer.py", line 561, in load_model
2024-12-20 15:09:20     self.model = registry.get_model_class(model_name)(
2024-12-20 15:09:20                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2024-12-20 15:09:20   File "/opt/conda/lib/python3.11/site-packages/fairchem/core/models/base.py", line 267, in __init__
2024-12-20 15:09:20     self.backbone: BackboneInterface = registry.get_model_class(
2024-12-20 15:09:20                                        ^^^^^^^^^^^^^^^^^^^^^^^^^
2024-12-20 15:09:20 TypeError: EqV2DeNSBackbone.__init__() got an unexpected keyword argument 'use_denoising_energy'

I have checked the source code of the EqV2DeNSBackbone class, indeed this class init function does not have a 'use_denoising_energy' argument.

Then, I used another checkpoint eqV2_31M_omat_mp_salex.pt, and it suceeded without getting error. Therefore, I am afraid that there are some arguments settings for the eqV2_dens_31M_mp.pt eqV2_dens_86M_mp.pt eqV2_dens_153M_mp.pt which are not compatible with the EqV2DeNSBackbone.init() function.

[lbluque]

Hi @jinlhr542 did you download the checkpoints recently? I uploaded compatible checkpoints earlier this week.

I double checked on my end and can load the DeNS checkpoints without issue.