feat: lab 3307 aau i use the kilillm functions to import llm data static

[RuellePaul]

new method kili.llm.import_conversations

note for reviewers

You can try create a LLM static project and import labeled conversations with this snipper

from kili.client import Kili
from scripts.constants import LOCAL_KILI_API_KEY

kili = Kili(api_key=LOCAL_KILI_API_KEY, api_endpoint='http://localhost:4001/api/label/v2/graphql')

interface = {
  "jobs": {
      "CLASSIFICATION_JOB_AT_COMPLETION_LEVEL": {
          "content": {
              "categories": {
                  "TOO_SHORT": {
                      "children": [],
                      "name": "Too short",
                      "id": "category1"
                  },
                  "JUST_RIGHT": {
                      "children": [],
                      "name": "Just right",
                      "id": "category2"
                  },
                  "TOO_VERBOSE": {
                      "children": [],
                      "name": "Too verbose",
                      "id": "category3"
                  }
              },
              "input": "radio"
          },
          "instruction": "Verbosity",
          "level": "completion",
          "mlTask": "CLASSIFICATION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "CLASSIFICATION_JOB_AT_COMPLETION_LEVEL_1": {
          "content": {
              "categories": {
                  "NO_ISSUES": {
                      "children": [],
                      "name": "No issues",
                      "id": "category4"
                  },
                  "MINOR_ISSUES": {
                      "children": [],
                      "name": "Minor issue(s)",
                      "id": "category5"
                  },
                  "MAJOR_ISSUES": {
                      "children": [],
                      "name": "Major issue(s)",
                      "id": "category6"
                  }
              },
              "input": "radio"
          },
          "instruction": "Instructions Following",
          "level": "completion",
          "mlTask": "CLASSIFICATION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "CLASSIFICATION_JOB_AT_COMPLETION_LEVEL_2": {
          "content": {
              "categories": {
                  "NO_ISSUES": {
                      "children": [],
                      "name": "No issues",
                      "id": "category7"
                  },
                  "MINOR_INACCURACY": {
                      "children": [],
                      "name": "Minor inaccuracy",
                      "id": "category8"
                  },
                  "MAJOR_INACCURACY": {
                      "children": [],
                      "name": "Major inaccuracy",
                      "id": "category9"
                  }
              },
              "input": "radio"
          },
          "instruction": "Truthfulness",
          "level": "completion",
          "mlTask": "CLASSIFICATION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "CLASSIFICATION_JOB_AT_COMPLETION_LEVEL_3": {
          "content": {
              "categories": {
                  "NO_ISSUES": {
                      "children": [],
                      "name": "No issues",
                      "id": "category10"
                  },
                  "MINOR_SAFETY_CONCERN": {
                      "children": [],
                      "name": "Minor safety concern",
                      "id": "category11"
                  },
                  "MAJOR_SAFETY_CONCERN": {
                      "children": [],
                      "name": "Major safety concern",
                      "id": "category12"
                  }
              },
              "input": "radio"
          },
          "instruction": "Harmlessness/Safety",
          "level": "completion",
          "mlTask": "CLASSIFICATION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "TRANSCRIPTION_JOB_AT_COMPLETION_LEVEL": {
          "content": {
              "input": "textField"
          },
          "instruction": "Additional comments...",
          "level": "completion",
          "mlTask": "TRANSCRIPTION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "TRANSCRIPTION_MARKDOWN_JOB_AT_COMPLETION_LEVEL": {
          "content": {
              "input": "markdown"
          },
          "instruction": "Additional comments...",
          "level": "completion",
          "mlTask": "TRANSCRIPTION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "COMPARISON_JOB": {
          "content": {
              "options": {
                  "IS_MUCH_BETTER": {
                      "children": [],
                      "name": "Is much better",
                      "id": "option13"
                  },
                  "IS_BETTER": {
                      "children": [],
                      "name": "Is better",
                      "id": "option14"
                  },
                  "IS_SLIGHTLY_BETTER": {
                      "children": [],
                      "name": "Is slightly better",
                      "id": "option15"
                  },
                  "TIE": {
                      "children": [],
                      "name": "Tie",
                      "mutual": True,
                      "id": "option16"
                  }
              },
              "input": "radio"
          },
          "instruction": "Pick the best answer",
          "mlTask": "COMPARISON",
          "required": 1,
          "isChild": False,
          "isNew": False
      },
      "CLASSIFICATION_JOB_AT_ROUND_LEVEL": {
          "content": {
              "categories": {
                  "BOTH_ARE_GOOD": {
                      "children": [],
                      "name": "Both are good",
                      "id": "category17"
                  },
                  "BOTH_ARE_BAD": {
                      "children": [],
                      "name": "Both are bad",
                      "id": "category18"
                  }
              },
              "input": "radio"
          },
          "instruction": "Overall quality",
          "level": "round",
          "mlTask": "CLASSIFICATION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "TRANSCRIPTION_JOB_AT_ROUND_LEVEL": {
          "content": {
              "input": "textField"
          },
          "instruction": "Additional comments...",
          "level": "round",
          "mlTask": "TRANSCRIPTION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "TRANSCRIPTION_MARKDOWN_JOB_AT_ROUND_LEVEL": {
          "content": {
              "input": "markdown"
          },
          "instruction": "Additional comments...",
          "level": "round",
          "mlTask": "TRANSCRIPTION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "CLASSIFICATION_JOB_AT_CONVERSATION_LEVEL": {
          "content": {
              "categories": {
                  "GLOBAL_GOOD": {
                      "children": [],
                      "name": "Globally good",
                      "id": "category19"
                  },
                  "BOTH_ARE_BAD": {
                      "children": [],
                      "name": "Globally bad",
                      "id": "category20"
                  }
              },
              "input": "radio"
          },
          "instruction": "Global",
          "level": "conversation",
          "mlTask": "CLASSIFICATION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "TRANSCRIPTION_JOB_AT_CONVERSATION_LEVEL": {
          "content": {
              "input": "textField"
          },
          "instruction": "Additional comments...",
          "level": "conversation",
          "mlTask": "TRANSCRIPTION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
      "TRANSCRIPTION_MARKDOWN_JOB_AT_CONVERSATION_LEVEL": {
          "content": {
              "input": "markdown"
          },
          "instruction": "Additional comments...",
          "level": "conversation",
          "mlTask": "TRANSCRIPTION",
          "required": 0,
          "isChild": False,
          "isNew": False
      },
  }
}

project_id = kili.create_project(
  title="LLM Static Project (created with SDK)",
  description="Project Description",
  input_type="LLM_STATIC",
  json_interface=interface
)["id"]

conversations = [
{
  "externalId": "Fibonacci python function",
  "chatItems": [
    {
      "externalId": "48cf1705-7be7-4b35-b4f4-3a13e9eadc1e",
      "role": "SYSTEM",
      "content": "You are an assistant helping the user by providing precise answers."
    },
    {
      "externalId": "9c43a0ae-5a6e-439b-94f6-83f7d5618b22",
      "role": "USER",
      "content": "Can you give me a function to compute Fibonacci numbers?"
    },
    {
      "externalId": "d5596f35-1b0f-42cb-b529-684e23138141",
      "role": "ASSISTANT",
      "content": "Here is a recursive implementation for calculating Fibonacci numbers:\n\n```python\ndef fibonacci_recursive(n):\n    if n <= 1:\n        return n\n    return fibonacci_recursive(n - 1) + fibonacci_recursive(n - 2)\n```\n\nThis is simple to write but has a time complexity of \\( O(2^n) \\) due to redundant calculations. For large values of \\( n \\), this method becomes highly inefficient.",
      "modelName": "gpt-4o"
    },
    {
      "externalId": "94f676b7-c3be-45a8-bccf-36b56566b765",
      "role": "ASSISTANT",
      "content": "Here is an optimized implementation using matrix exponentiation, along with an explanation of why it is faster:\n\n```python\nimport numpy as np\n\ndef fibonacci_matrix(n):\n    def matrix_power(matrix, power):\n        result = np.identity(len(matrix), dtype=int)\n        while power:\n            if power % 2:\n                result = np.dot(result, matrix)\n            matrix = np.dot(matrix, matrix)\n            power //= 2\n        return result\n\n    if n == 0:\n        return 0\n    base_matrix = np.array([[1, 1], [1, 0]], dtype=int)\n    result_matrix = matrix_power(base_matrix, n - 1)\n    return result_matrix[0, 0]\n```\n\n### Why is Matrix Exponentiation Faster?\n\nThe recursive method recalculates values for overlapping subproblems, leading to exponential time complexity:\n\n\\[ T(n) = T(n-1) + T(n-2) + O(1), \\]\n\nwhich resolves to \\( O(2^n) \\). This inefficiency arises from redundant recalculations.\n\nThe matrix exponentiation method uses the property that Fibonacci numbers can be represented as a matrix multiplication:\n\n\\[ \\begin{bmatrix} F(n+1) & F(n) \\\\ F(n) & F(n-1) \\end{bmatrix} = \\begin{bmatrix} 1 & 1 \\\\ 1 & 0 \\end{bmatrix}^n. \\]\n\nUsing fast exponentiation (repeated squaring), the matrix power can be computed in \\( O(\\log n) \\) steps. For example, instead of multiplying the matrix \\( n \\) times, we:\n\n1. Square the matrix \\( \\lceil \\log_2 n \\rceil \\) times.\n2. Combine results for any remaining powers.\n\nThis drastically reduces the number of operations compared to the recursive method.\n\n### Comparison of Complexities\n\n- **Recursive Method**: \\( O(2^n) \\)\n- **Matrix Exponentiation**: \\( O(\\log n) \\)\n\nFor large \\( n \\), the difference is substantial. Consider \\( n = 50 \\):\n\n- Recursive method performs roughly \\( 2^{50} \\approx 10^{15} \\) operations.\n- Matrix exponentiation performs approximately \\( \\log_2(50) \\approx 6 \\) matrix multiplications, where each multiplication is \\( O(1) \\) for a fixed-size \\( 2 \\times 2 \\) matrix.\n\n### Example\nUsing \\( n = 10 \\):\n\n1. Recursive Method: Explores all possible paths in the Fibonacci tree, recalculating values like \\( F(8) \\) multiple times.\n2. Matrix Exponentiation: Squares the base matrix 4 times and performs 1 additional multiplication, yielding the result in far fewer steps.\n\nThis efficiency makes matrix exponentiation the preferred method for calculating Fibonacci numbers at scale.",
      "modelName": "gpt-o1"
    }
  ],
  "metadata": {}
},
{
  "externalId": "The sum of the integers",
  "chatItems": [
    {
      "externalId": "00000000-8b5c-4b6a-0000-5b232d8bcbb9",
      "role": "SYSTEM",
      "content": "You are an assistant helping the user by providing clear and concise answers."
    },
    {
      "externalId": "19d4a9e2-8b5c-4b6a-a27c-5b232d8bcbb9",
      "role": "USER",
      "content": "What is the sum of all integers up to infinity?"
    },
    {
      "externalId": "a82f76e4-7f47-4e45-a745-4b934f11f6f1",
      "role": "ASSISTANT",
      "content": "The series for the sum of all integers up to infinity, given by:\n\n\\[ S = 1 + 2 + 3 + 4 + \\dots \\]\n\ndiverges to infinity. This means the sum grows without bound and does not converge to a finite value.",
      "modelName": "gpt-o1"
    },
    {
      "externalId": "f3dcd3f0-5e0e-404c-9759-3c60a15e634e",
      "role": "ASSISTANT",
      "content": "While the series:\n\n\\[ S = 1 + 2 + 3 + 4 + \\dots \\]\n\ndiverges in the traditional sense, there exists a deeper mathematical interpretation through the **analytic continuation of the Riemann zeta function**. This assigns a finite value to this sum in a unique way.\n\nThe Riemann zeta function \\( \\zeta(s) \\) is initially defined for \\( \\text{Re}(s) > 1 \\) as:\n\n\\[ \\zeta(s) = \\sum_{n=1}^\\infty \\frac{1}{n^s}. \\]\n\nThrough a process called analytic continuation, \\( \\zeta(s) \\) can be extended to the entire complex plane, except at \\( s = 1 \\) (where it has a pole). For \\( s = -1 \\), the value of \\( \\zeta(-1) \\) is:\n\n\\[ \\zeta(-1) = -\\frac{1}{12}. \\]\n\nThus, using this extended framework:\n\n\\[ 1 + 2 + 3 + 4 + \\dots = -\\frac{1}{12}. \\]\n\n### Applications and Context\nThis surprising result is not meant to suggest that adding all positive integers in the traditional sense equals \\( -1/12 \\). Instead, it has profound implications in advanced mathematics and theoretical physics:\n\n1. **String Theory**: The value \\( -1/12 \\) arises in the computation of the energy of quantum strings, providing finite results where naive calculations yield divergences.\n2. **Quantum Field Theory**: Techniques like zeta regularization assign finite values to divergent sums, often leveraging \\( \\zeta(-1) \\).\n3. **Casimir Effect**: The Casimir effect calculates the force between two parallel plates due to quantum vacuum fluctuations. The series \\( 1 + 2 + 3 + \\dots \\) naturally appears in this context, and its regularized value of \\( -1/12 \\) provides a correction aligning with experimental results.\n\n### Example: Casimir Effect\nConsider two conducting plates placed in a vacuum. Quantum field theory predicts the energy density due to vacuum fluctuations, and summing these energies involves divergent series such as \\( S = 1 + 2 + 3 + \\dots \\). By assigning \\( \\zeta(-1) = -1/12 \\), the theory yields a finite and experimentally verified prediction of the force between the plates.\n\nThis demonstrates how the concept of \\( \\zeta(-1) \\) provides a mathematically consistent and physically meaningful interpretation in specific advanced scenarios.",
      "modelName": "gpt-4o"
    }
  ],
  "metadata": {}
},
{
  "externalId": "Caesar cipher decoding",
  "chatItems": [
    {
      "externalId": "48cf1705-7be7-4b35-b4f4-3a13e9eadc1e",
      "role": "SYSTEM",
      "content": "You are an assistant helping the user by providing clear and concise answers."
    },
    {
      "externalId": "7a4cde2c-95b4-4d3f-94fa-233a3c3a9171",
      "role": "USER",
      "content": "Can you decode this: Pmttw ewztl !"
    },
    {
      "externalId": "ab1e29bf-6b94-4b78-b920-8bfe5c2370f4",
      "role": "ASSISTANT",
      "content": "This seems like an encoded message, perhaps in a secret language of aliens or something far beyond human comprehension! The mystery is profound, but I'm unable to decode it. Maybe it contains a cosmic secret?",
      "modelName": "gpt-3"
    },
    {
      "externalId": "93dc1b4d-dc25-48cc-9ef6-13e01cbd45f2",
      "role": "ASSISTANT",
      "content": "The encoded message is a **Caesar cipher**, a simple encryption technique where each letter is shifted by a fixed number of places in the alphabet.\n\nTo decode \\( Pmttw\\ ewztl! \\), we determine that the shift key is **8** (each letter is shifted back by 8 positions). Decoding it gives:\n\n**Hello world!**\n\n### Python Code for Caesar Cipher Decryption\nYou can use the following Python code to decrypt any Caesar cipher by specifying the shift key:\n\n```python\ndef caesar_cipher_decrypt(ciphertext, shift):\n    decrypted_text = []\n    for char in ciphertext:\n        if char.isalpha():\n            # Determine if the character is uppercase or lowercase\n            base = ord('A') if char.isupper() else ord('a')\n            # Shift the character and wrap around using modulo\n            new_char = chr((ord(char) - base - shift) % 26 + base)\n            decrypted_text.append(new_char)\n        else:\n            # Non-alphabetic characters remain unchanged\n            decrypted_text.append(char)\n    return ''.join(decrypted_text)\n\n# Example usage\nciphertext = \"Pmttw ewztl !\"\nshift = 8\nprint(caesar_cipher_decrypt(ciphertext, shift))\n```\n\n### Explanation\n1. **Shift Key**: The Caesar cipher uses a fixed number to shift each letter. In this case, the shift key is \\( 8 \\).\n2. **Decoding Process**: Each letter is shifted backward by \\( 8 \\) positions in the alphabet, wrapping around from \\( A \\) to \\( Z \\) or \\( a \\) to \\( z \\) as needed.\n\n### Result\nRunning the code will correctly decode the message to:\n\n**Hello world!**",
      "modelName": "gpt-o1"
    }
  ],
  "label": {
    "completion": {
      "CLASSIFICATION_JOB_AT_COMPLETION_LEVEL": {
        "ab1e29bf-6b94-4b78-b920-8bfe5c2370f4": {
          "categories": [
            "TOO_SHORT"
          ]
        },
        "93dc1b4d-dc25-48cc-9ef6-13e01cbd45f2": {
          "categories": [
            "JUST_RIGHT"
          ]
        }
      },
      "CLASSIFICATION_JOB_AT_COMPLETION_LEVEL_1": {
        "ab1e29bf-6b94-4b78-b920-8bfe5c2370f4": {
          "categories": [
            "MINOR_ISSUES"
          ]
        }
      },
      "CLASSIFICATION_JOB_AT_COMPLETION_LEVEL_2": {
        "ab1e29bf-6b94-4b78-b920-8bfe5c2370f4": {
          "categories": [
            "MINOR_INACCURACY"
          ]
        }
      },
      "CLASSIFICATION_JOB_AT_COMPLETION_LEVEL_3": {
        "ab1e29bf-6b94-4b78-b920-8bfe5c2370f4": {
          "categories": [
            "MINOR_SAFETY_CONCERN"
          ]
        }
      }
    },
    "round": {
      "CLASSIFICATION_JOB_AT_ROUND_LEVEL": {
        "0": {
          "categories": [
            "BOTH_ARE_GOOD"
          ]
        }
      },
      "COMPARISON_JOB": {
        "0": {
          "code": "Is much better",
          "firstId": "93dc1b4d-dc25-48cc-9ef6-13e01cbd45f2",
          "secondId": "ab1e29bf-6b94-4b78-b920-8bfe5c2370f4"
        }
      }
    },
    "conversation": {
      "CLASSIFICATION_JOB_AT_CONVERSATION_LEVEL": {
        "categories": [
          "GLOBAL_GOOD"
        ]
      },
      "TRANSCRIPTION_JOB_AT_CONVERSATION_LEVEL": {
        "text": "Great conversation!"
      }
    }
  },
  "labeler": "test+fx@kili-technology.com",
  "metadata": {}
}
]

response = kili.llm.import_conversations(
  project_id=project_id,
  conversations=conversations
)

refactored LLM export

LLM static & dynamic export is the same format as import
Renamed old llm_static related code to llm_rlhf for clarity

note for reviewers

Try to label & export LLM static or dynamic project , then import again into a LLM static project

from kili.client import Kili
from scripts.constants import LOCAL_KILI_API_KEY

kili = Kili(api_key=LOCAL_KILI_API_KEY, api_endpoint='http://localhost:4001/api/label/v2/graphql')

kili.llm.export(project_id="cm66gj5qe01kol70weiqpdyz3")

Add new tutorial for LLM static import

Add unit test for both llm static & dynamic export

Add e2e test for import conversations

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB