Issue for getting embedding

[zxshi0102]

Hi there, I am trying to create the embedding layer for my antibody sequence with ABodyBuilder3-LM

use_precomputed = False
if use_precomputed:
    embedding = torch.load("/home/user/package_download/abodybuilder3/data/structures/structures_plm/6yio_H0-L0.pt")[
        "plm_embedding"
    ]
else:
    plm = ProtT5()
    embedding = plm.get_embeddings(
        [
            heavy,
        ],
        [
            light,
        ],
    )

After running this code, it shows the error:

RuntimeError                              Traceback (most recent call last)
Cell In[4], line 8
      4     embedding = torch.load("/home/user/package_download/abodybuilder3/data/structures/structures_plm/6yio_H0-L0.pt")[
      5         "plm_embedding"
      6     ]
      7 else:
----> 8     plm = ProtT5()
      9     embedding = plm.get_embeddings(
     10         [
     11             heavy,
   (...)
     15         ],
     16     )
     18 print(f"{embedding.shape=}")

File <string>:8, in __init__(self, weights_dir, model_type, paired, batch_size, device_map)

File ~/package_download/abodybuilder3/src/abodybuilder3/language/model.py:55, in ProtTrans.__post_init__(self)
     52 elif self.paired and self.model_type == "t5":
     53     self.seperator_token = "</s>"
---> 55 self.trainer = Trainer(num_nodes=1, devices=1)

File ~/package_download/abodybuilder3/.venv/lib/python3.9/site-packages/lightning/pytorch/utilities/argparse.py:70, in _defaults_from_env_vars.<locals>.insert_env_defaults(self, *args, **kwargs)
     67 kwargs = dict(list(env_variables.items()) + list(kwargs.items()))
     69 # all args were already moved to kwargs
---> 70 return fn(self, **kwargs)

File ~/package_download/abodybuilder3/.venv/lib/python3.9/site-packages/lightning/pytorch/trainer/trainer.py:401, in Trainer.__init__(self, accelerator, strategy, devices, num_nodes, precision, logger, callbacks, fast_dev_run, max_epochs, min_epochs, max_steps, min_steps, max_time, limit_train_batches, limit_val_batches, limit_test_batches, limit_predict_batches, overfit_batches, val_check_interval, check_val_every_n_epoch, num_sanity_val_steps, log_every_n_steps, enable_checkpointing, enable_progress_bar, enable_model_summary, accumulate_grad_batches, gradient_clip_val, gradient_clip_algorithm, deterministic, benchmark, inference_mode, use_distributed_sampler, profiler, detect_anomaly, barebones, plugins, sync_batchnorm, reload_dataloaders_every_n_epochs, default_root_dir)
    398 # init connectors
    399 self._data_connector = _DataConnector(self)
--> 401 self._accelerator_connector = _AcceleratorConnector(
    402     devices=devices,
    403     accelerator=accelerator,
    404     strategy=strategy,
    405     num_nodes=num_nodes,
    406     sync_batchnorm=sync_batchnorm,
    407     benchmark=benchmark,
    408     use_distributed_sampler=use_distributed_sampler,
    409     deterministic=deterministic,
    410     precision=precision,
    411     plugins=plugins,
    412 )
    413 self._logger_connector = _LoggerConnector(self)
    414 self._callback_connector = _CallbackConnector(self)

File ~/package_download/abodybuilder3/.venv/lib/python3.9/site-packages/lightning/pytorch/trainer/connectors/accelerator_connector.py:158, in _AcceleratorConnector.__init__(self, devices, num_nodes, accelerator, strategy, plugins, precision, sync_batchnorm, benchmark, use_distributed_sampler, deterministic)
    155 self._set_parallel_devices_and_init_accelerator()
    157 # 3. Instantiate ClusterEnvironment
--> 158 self.cluster_environment: ClusterEnvironment = self._choose_and_init_cluster_environment()
    160 # 4. Instantiate Strategy - Part 1
    161 if self._strategy_flag == "auto":

File ~/package_download/abodybuilder3/.venv/lib/python3.9/site-packages/lightning/pytorch/trainer/connectors/accelerator_connector.py:428, in _AcceleratorConnector._choose_and_init_cluster_environment(self)
    420 for env_type in (
    421     # TorchElastic has the highest priority since it can also be used inside SLURM
    422     TorchElasticEnvironment,
   (...)
    425     MPIEnvironment,
    426 ):
    427     if env_type.detect():
--> 428         return env_type()
    429 if _LIGHTNING_BAGUA_AVAILABLE:
    430     from lightning_bagua import BaguaEnvironment

File ~/package_download/abodybuilder3/.venv/lib/python3.9/site-packages/lightning/fabric/plugins/environments/slurm.py:52, in SLURMEnvironment.__init__(self, auto_requeue, requeue_signal)
     50 self.requeue_signal = requeue_signal
     51 self._validate_srun_used()
---> 52 self._validate_srun_variables()

File ~/package_download/abodybuilder3/.venv/lib/python3.9/site-packages/lightning/fabric/plugins/environments/slurm.py:210, in SLURMEnvironment._validate_srun_variables()
    208 print(os.environ)
    209 if ntasks > 1 and "SLURM_NTASKS_PER_NODE" not in os.environ:
--> 210     raise RuntimeError(
    211         f"You set `--ntasks={ntasks}` in your SLURM bash script, but this variable is not supported."
    212         f" HINT: Use `--ntasks-per-node={ntasks}` instead."
    213     )

RuntimeError: You set `--ntasks=64` in your SLURM bash script, but this variable is not supported. HINT: Use `--ntasks-per-node=64` instead.

And i tried to change all the ntasks to ntasks-per-node in abodybuilder3/.venv/lib/python3.9/site-packages/lightning/fabric/plugins/environments/slurm.py, it would make the package cannot be imported at the beginning. But I cannot find the other ways to fix it, could you give me some hints to fix it? Thanks!

[henrykenlay]

Hi @zxshi0102, the error suggests you are using Slurm to launch a script or an interactive session to run this code, and that you specify --ntasks instead of --ntasks-per-node in the Slurm script. Can you check in your Slurm script for these variables?

To isolate the issue, could you try importing and initialising ProtT5 from abodybuilder3/language/model.py?

[zxshi0102]

Thank you for your prompt response! I am indeed using an interactive session via jupyter notebook, the error shows only when trying to compute the embeddings, when running the prediction with the full model the command works, and I am able to produce the PDB outputs.

I have tried to import and initialize the model as suggested with

sys.path.append("/home/zshi/package_download/abodybuilder3")
from abodybuilder3.language.model import ProtT5

Unfortunately, the same error shows up。Thanks! and let me know if there is something else, I should try, otherwise I will default to using the full model.