experiment_id: E1_core
splits:
- test
families:
- synthetic
- exec
lengths:
- 2000
- 8000
- 16000
limit_docs_per_family: 40
max_questions_per_doc: 5
model_pairs:
- sender: llama3b
  receiver: llama3b
generation:
  max_tokens: 48
  per_family:
    exec: 320
seeds:
- 0
conditions:
- method: A8
  budget_tokens: 512
  access: blind
note: 'Retry of A8 (LLMLingua-2) for the Llama pair on synthetic+exec: the M3U96a
  environment lacked torch (implementation_error); rerun on the laptop, new attempts
  appended to E1_core'
protocol_hash: 404c52b34d964693da72ef459a036ffc42fbf5f6ba8748cc9de0c554ebebddd6