Fix PERPLEXITY task

[ScottHoang]

---

Fix PERPLEXITY task evaluation crash due to incorrect sampling method detection

Issue

Evaluating perplexity-based tasks (e.g., lighteval|wikitext:103:document_level|0) causes an IndexError crash:

File "lighteval/logging/info_loggers.py", line 298, in aggregate
hash_types: list[str] = list(self.compiled_details.values())[0].hashes.keys()
IndexError: list index out of range

Root Cause

The @cached decorator in cache_management.py incorrectly filters out all PERPLEXITY responses:

1. Incorrect sampling method detection (cache_management.py:203-208):
   - get_sampling_method() cannot distinguish between PERPLEXITY and LOGPROBS responses
   - Both have non-empty logprobs, so it returns SamplingMethod.LOGPROBS for PERPLEXITY tasks
   - The key difference: PERPLEXITY responses have empty text=[], while LOGPROBS typically has text
2. Filtering removes all responses (cache_management.py:421-423):
   - The decorator filters cached results by sampling method
   - Looks for SamplingMethod.PERPLEXITY but finds SamplingMethod.LOGPROBS
   - All 62 valid responses filtered out → returns empty list []
3. Downstream crash (info_loggers.py:298):
   - DetailsLogger.aggregate() assumes at least one task has details
   - Accesses list(self.compiled_details.values())[0] on empty dict
   - Raises IndexError

Fixes

1. Fix sampling method detection (src/lighteval/utils/cache_management.py:203-208):

  def get_sampling_method(self, sample: dict) -> str:
      if len(sample.get("logprobs", [])) > 0:
          # PERPLEXITY tasks have logprobs but empty text
          if len(sample.get("text", [])) == 0:
              return SamplingMethod.PERPLEXITY
          return SamplingMethod.LOGPROBS
      if len(sample.get("text", [])) > 0:
          return SamplingMethod.GENERATIVE
      return None

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[NathanHB]

thanks for the fix ! happy to review once formatted, don't hesitate to ping me for review :)

[ScottHoang]

Hi @NathanHB, thanks for the response! I have reformatted the file back to its original state. Can you review? 🙏

[ScottHoang]

hi @NathanHB , not quite sure why it is failing some of the test. I don't think my fix would affect anything beside ppl tasks....

[NathanHB]

thanks, taking a look !

[NathanHB]

not sure what the issue is, i will take a deeper look this week, if you can, in the meantime run the tests locally ?