Key error when running trainer

[lina-van]

Hi, it's super basic and I apologize, but I seem to be getting a strange error when trying to run the trainer from SetFit.

Here is the code to instantiate:

trainer = AbsaTrainer(
    model,
    train_dataset=train_data

When I run this once, I get this errr:

ImportError: Using the `Trainer` with `PyTorch` requires `accelerate>=0.26.0`: Please run `pip install transformers[torch]` or `pip install 'accelerate>=0.26.0'`
File <command-6658730431216308>, line 5
      2 model = AbsaModel.from_pretrained("sentence-transformers/paraphrase-mpnet-base-v2")
      4 # Create a trainer:
----> 5 trainer = AbsaTrainer(
      6     model,
      7     train_dataset=train_data
      8 )
      9 # Execute training:
     10 trainer.train()
File /local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.11/site-packages/transformers/training_args.py:2192, in TrainingArguments._setup_devices(self)
   2190 if not is_sagemaker_mp_enabled():
   2191     if not is_accelerate_available():
-> 2192         raise ImportError(
   2193             f"Using the `Trainer` with `PyTorch` requires `accelerate>={ACCELERATE_MIN_VERSION}`: "
   2194             f"Please run `pip install transformers[torch]` or `pip install 'accelerate>={ACCELERATE_MIN_VERSION}'`"
   2195         )
   2196 # We delay the init of `PartialState` to the end for clarity
   2197 accelerator_state_kwargs: dict[str, Any] = {"enabled": True, "use_configured_state": False}

but when I re-run the cell, I get a different error:

KeyError: 'text'
File <command-6658730431216308>, line 5
      2 model = AbsaModel.from_pretrained("sentence-transformers/paraphrase-mpnet-base-v2")
      4 # Create a trainer:
----> 5 trainer = AbsaTrainer(
      6     model,
      7     train_dataset=train_data
      8 )
      9 # Execute training:
     10 trainer.train()
File /local_disk0/.ephemeral_nfs/envs/pythonEnv-87f5ca92-d40a-4dab-b7e4-2553804dffba/lib/python3.11/site-packages/setfit/span/trainer.py:120, in AbsaTrainer.preprocess_dataset(self, aspect_model, polarity_model, dataset)
    118 grouped_data = defaultdict(list)
    119 for sample in dataset:
--> 120     text = sample.pop("text")
    121     grouped_data[text].append(sample)
    123 def index_ordinal(text: str, target: str, ordinal: int) -> Tuple[int, int]:```


From what I've read of the pop function it just removes and return this element. My data is already in the dictionary format of text, span, label ordinal dictionaries as needed by SetFit. Torch and accelerate are indeed installed.