Geometry of Truth tutorial uses pandas with remote models, but pandas is not whitelisted

[marthaflinderslewis]

Running the Geometry of truth tutorial up to the following cell gives the error that pandas is not whitelisted. I can rewrite to not use pandas, but is there any plan to allow the use of pandas?

import torch

PROMPT_TEMPLATE = """The city of Tokyo is in Japan. This statement is: TRUE
The city of Hanoi is in Poland. This statement is: FALSE
{statement} This statement is:"""

LAYER = 20
punctuation_token_id = model.tokenizer('.').input_ids[1] # extract activation over "."

true_activations = []
false_activations = []
with model.session(remote=True) as session:
  for i in range(df.shape[0]):
    row = df.iloc[i]
    prompt = PROMPT_TEMPLATE.format(statement=row.statement)
    prompt_token_ids = model.tokenizer(prompt).input_ids
    punctuation_index = rindex(prompt_token_ids, punctuation_token_id) - 1

    with model.trace(prompt):
      activation = model.model.layers[LAYER].output[0][:, punctuation_index, :].save()
    if row.label == 0:
      false_activations.append(activation)
    else:
      true_activations.append(activation)

clear_output()

true_activations = torch.cat(true_activations)
false_activations = torch.cat(false_activations)

true_activations.shape, false_activations.shape

gives the following output and error:

[2025-10-05 17:18:43] [67dc0c6a-65ed-4755-962a-4675f97730c6] RECEIVED   : Your job has been received and is waiting to be queued.
INFO:nnsight_remote:[67dc0c6a-65ed-4755-962a-4675f97730c6] RECEIVED   : Your job has been received and is waiting to be queued.
[2025-10-05 17:18:43] [67dc0c6a-65ed-4755-962a-4675f97730c6] QUEUED     : Your job has been recieved by the coordinator and is waiting to be queued.
INFO:nnsight_remote:[67dc0c6a-65ed-4755-962a-4675f97730c6] QUEUED     : Your job has been recieved by the coordinator and is waiting to be queued.
[2025-10-05 17:18:43] [67dc0c6a-65ed-4755-962a-4675f97730c6] QUEUED     : Task 67dc0c6a-65ed-4755-962a-4675f97730c6 has been added to the queue. Currently at position 1
INFO:nnsight_remote:[67dc0c6a-65ed-4755-962a-4675f97730c6] QUEUED     : Task 67dc0c6a-65ed-4755-962a-4675f97730c6 has been added to the queue. Currently at position 1
[2025-10-05 17:18:43] [67dc0c6a-65ed-4755-962a-4675f97730c6] DISPATCHED : Dispatching task...
INFO:nnsight_remote:[67dc0c6a-65ed-4755-962a-4675f97730c6] DISPATCHED : Dispatching task...

---------------------------------------------------------------------------

ExitTracingException                      Traceback (most recent call last)

[/tmp/ipython-input-1817750371.py](https://localhost:8080/#) in <cell line: 0>()
     12 with model.session(remote=True) as session:
---> 13   for i in range(df.shape[0]):
     14     row = df.iloc[i]

6 frames

ExitTracingException: 


During handling of the above exception, another exception occurred:

RemoteException                           Traceback (most recent call last)

[/usr/local/lib/python3.12/dist-packages/nnsight/intervention/backends/remote.py](https://localhost:8080/#) in handle_response(self, response, tracer)
    123 
    124         if response.status == ResponseModel.JobStatus.ERROR:
--> 125             raise RemoteException(f"{response.description}\nRemote exception.")
    126 
    127         # Log response for user

RemoteException: Traceback (most recent call last):
  File "/u/svcndifuser/ndif-deployment/repos/prod/ndif/src/services/ray/src/ray/deployments/modeling/base.py", line 191, in __call__
    inputs = self.pre()
             ^^^^^^^^^^
  File "/u/svcndifuser/miniconda3/envs/service/lib/python3.12/site-packages/ray/util/tracing/tracing_helper.py", line 463, in _resume_span
    return method(self, *_args, **_kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/u/svcndifuser/ndif-deployment/repos/prod/ndif/src/services/ray/src/ray/deployments/modeling/base.py", line 242, in pre
    request = self.request.deserialize(self.model)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/u/svcndifuser/ndif-deployment/repos/prod/ndif/src/services/ray/src/schema/request.py", line 64, in deserialize
    return RequestModel.deserialize(model, request, self.zlib)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/u/svcndifuser/miniconda3/envs/service/lib/python3.12/site-packages/nnsight/schema/request.py", line 44, in deserialize
    request:RequestModel = load(data.read(), model)
                           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/u/svcndifuser/miniconda3/envs/service/lib/python3.12/site-packages/nnsight/intervention/serialization.py", line 72, in load
    return CustomCloudUnpickler(io.BytesIO(data), model, frame).load()
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/u/svcndifuser/ndif-deployment/repos/prod/ndif/src/services/ray/src/ray/nn/security/protected_objects.py", line 93, in load
    return pickle.Unpickler.load(self)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/u/svcndifuser/ndif-deployment/repos/prod/ndif/src/services/ray/src/ray/nn/security/protected_environment.py", line 168, in __call__
    raise ImportError(f"Module {name} is not whitelisted")
ImportError: Module pandas.core.frame is not whitelisted

Module pandas.core.frame is not whitelisted
Remote exception.

[Butanium]

to make the code more readable you can use "py" instead of "", this will enable to syntax highlighting