{
  "id": "inference",
  "concept": "Inference & Pipelines",
  "fiction": "The knight learns to call a soul into a vessel, waking dormant models and commanding them to cast their logic upon the world.",
  "questions": [
    {
      "id": "inf_q1",
      "shown": {
        "question": "Explain what 'inference' means in the context of machine learning models.",
        "concept_brief": "Ask the knight to explain what the spirit does once it is awake and answering questions, rather than learning."
      },
      "check": {
        "strategy": "llm_judge",
        "spec": {
          "rubric": [
            "Mentions making predictions or generating outputs",
            "Mentions using an already trained model"
          ]
        }
      },
      "reveal": "Inference is the process of using a trained model to make predictions or generate outputs on new data."
    },
    {
      "id": "inf_q2",
      "shown": {
        "question": "What is the name of the high-level `transformers` function that combines a tokenizer and a model to easily perform tasks without writing boilerplate code?",
        "concept_brief": "What is the name of the grand conduit that combines both the Runesmith's tool and the spirit's vessel into one swift spell?"
      },
      "check": {
        "strategy": "keyword",
        "spec": {
          "correct_any": [
            "pipeline",
            "pipeline()"
          ]
        }
      },
      "reveal": "The `pipeline` function."
    },
    {
      "id": "inf_q3",
      "shown": {
        "question": "Which specific class in the `transformers` library is used to automatically load language models that generate text step-by-step?",
        "concept_brief": "What specific vessel is forged for a spirit that speaks word by word, foretelling the next?"
      },
      "check": {
        "strategy": "keyword",
        "spec": {
          "correct_any": [
            "automodelforcausallm",
            "AutoModelForCausalLM"
          ]
        }
      },
      "reveal": "The `AutoModelForCausalLM` class."
    },
    {
      "id": "inf_q4",
      "shown": {
        "question": "Write one line of Python code to load the \"gpt2\" model using the standard method of the `AutoModel` class.",
        "concept_brief": "The knight must utter the exact incantation to pull the spirit 'gpt2' into a general vessel."
      },
      "check": {
        "strategy": "ast",
        "spec": {
          "must_call_any": [
            "from_pretrained"
          ]
        }
      },
      "reveal": "AutoModel.from_pretrained(\"gpt2\")"
    },
    {
      "id": "inf_q5",
      "shown": {
        "question": "Write one line of Python code to make a loaded language `model` produce new text based on input tensors, using the standard generation method.",
        "concept_brief": "Command the vessel to speak, bringing forth new runes from the seed it was given."
      },
      "check": {
        "strategy": "ast",
        "spec": {
          "must_call_any": [
            "generate"
          ]
        }
      },
      "reveal": "model.generate(**inputs)"
    },
    {
      "id": "inf_q6",
      "shown": {
        "question": "While Hugging Face supports multiple deep learning frameworks, which one is most commonly used by default for model weights on the Hub?",
        "concept_brief": "What is the foundational deep magic, the primary language of power, that most vessels are forged in?"
      },
      "check": {
        "strategy": "keyword",
        "spec": {
          "correct_any": [
            "pytorch",
            "torch"
          ]
        }
      },
      "reveal": "PyTorch."
    },
    {
      "id": "inf_q7",
      "shown": {
        "question": "In PyTorch, what method is used to move a loaded model to a specific hardware accelerator, like a GPU?",
        "concept_brief": "By what command does the knight physically shift the vessel onto the grand altar of computation?"
      },
      "check": {
        "strategy": "keyword",
        "spec": {
          "correct_any": [
            ".to",
            "to()",
            "cuda()"
          ]
        }
      },
      "reveal": "The `.to()` method (e.g., `model.to(\"cuda\")`)."
    },
    {
      "id": "inf_q8",
      "shown": {
        "question": "Running massive models requires a vast amount of what specific hardware memory resource, usually found on GPUs?",
        "concept_brief": "The great vessels require vast arenas of computation to awake. What is this specific heavy resource called?"
      },
      "check": {
        "strategy": "keyword",
        "spec": {
          "correct_any": [
            "vram",
            "gpu memory"
          ]
        }
      },
      "reveal": "VRAM (Video RAM)."
    },
    {
      "id": "inf_q9",
      "shown": {
        "question": "In the `diffusers` library, what is the core class used to load pre-trained image generation models, similar to how AutoModel works for text?",
        "concept_brief": "What is the grand conduit used specifically by the painters' guild to bring a visual spirit to life?"
      },
      "check": {
        "strategy": "keyword",
        "spec": {
          "correct_any": [
            "diffusionpipeline",
            "DiffusionPipeline",
            "stablediffusionpipeline"
          ]
        }
      },
      "reveal": "The `DiffusionPipeline` class."
    },
    {
      "id": "inf_q10",
      "shown": {
        "question": "What object or file defines the specific architecture of a model, such as its number of layers and hidden size?",
        "concept_brief": "What is the blueprint that dictates exactly how the vessel is shaped and how many chambers it has?"
      },
      "check": {
        "strategy": "keyword",
        "spec": {
          "correct_any": [
            "config",
            "configuration",
            "config.json"
          ]
        }
      },
      "reveal": "The configuration (or `config.json`)."
    }
  ]
}