Better error message for incorrect input in LanguageModel

[Butanium]

Currently this snippet:

from nnsight import LanguageModel

model  = LanguageModel("gpt2")
toks = model.tokenizer("Hello, world!", return_tensors="pt")
input_ids = toks["input_ids"]
attention_mask = toks["attention_mask"]
with model.trace(
            input_ids=input_ids, attention_mask=attention_mask
):
    x = model.transformer.h[0].output

Fails with this very uninformative message:

---------------------------------------------------------------------------
NNsightException                          Traceback (most recent call last)
Cell In[6], line 5
      2 from nnsight import LanguageModel
      4 model  = LanguageModel("gpt2")
----> 5 with model.trace(
      6             input_ids=input_ids, attention_mask=attention_mask
      7 ):
      8     x = model.transformer.h[0].output

File ~/projects/diffing-toolkit/.venv/lib/python3.12/site-packages/nnsight/intervention/tracing/base.py:665, in Tracer.__exit__(self, exc_type, exc_val, exc_tb)
    662 if self.asynchronous:
    663     return self.backend(self)
--> 665 self.backend(self)
    667 return True

File ~/projects/diffing-toolkit/.venv/lib/python3.12/site-packages/nnsight/intervention/backends/execution.py:24, in ExecutionBackend.__call__(self, tracer)
     21     return tracer.execute(fn)
     22 except Exception as e:
---> 24     raise wrap_exception(e, tracer.info) from None
     25 finally:
     26     Globals.exit()

NNsightException: 

Traceback (most recent call last):
  File "<ipython-input-6-77795bbc4ead>", line 8, in <module>
    x = model.transformer.h[0].output

ValueError: Cannot return output of Envoy that is not interleaving nor has a fake output set.

I guess this is a LanguageModel specific issue and not a general NNsight issue, but this can be confusing for people used to the hf interfact where this works:

from transformers import AutoModelForCausalLM
hf_model = AutoModelForCausalLM.from_pretrained("gpt2")
hf_model(input_ids=input_ids, attention_mask=attention_mask)

if there are some limitations that do not allow to support this synthax a nicer error message would be nice