Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/GingerlyData247/SOTeam4-P2/llms.txt

Use this file to discover all available pages before exploring further.

​
Endpoint

GET /artifact/model/{id}/lineage
Returns a directed graph representing the lineage relationships for a model artifact, including parent models (base models that this model was fine-tuned from) and metadata about how these relationships were discovered.

​
Path Parameters

​
id
string
required
The unique identifier of the model artifact to retrieve lineage for

​
Response

Returns a lineage graph with nodes and edges.
​
nodes
array
Array of model nodes in the lineage graphNode Properties:
  • artifact_id (string): Unique identifier for this node. For models in the registry, this is the model ID. For external models, this is prefixed with “external:”
  • name (string): Human-readable name of the model (usually the Hugging Face model ID)
  • source (string): How this parent relationship was discovered (typically “config_json”)
  • metadata (object): Additional metadata. External models include {"external": true}
​
edges
array
Array of directed edges representing parent-child relationshipsEdge Properties:
  • from_node_artifact_id (string): The artifact ID of the parent model
  • to_node_artifact_id (string): The artifact ID of the child model (derived from the parent)
  • relationship (string): Type of relationship, typically “base_model”

​
Parent Model Detection

Parent models are identified from structured metadata stored in the model’s config.json file. The system checks multiple fields:
  • base_model
  • teacher_model
  • parent_model
  • source_model
  • original_model
  • pretrained_model_name_or_path
If a parent model exists in the registry, it is linked directly. If the parent model is not in the registry (external dependency), it is included as an external node with a placeholder ID.

​
Tree Score Calculation

The tree_score metric (available via the /artifact/model/{id}/rate endpoint) uses lineage information to compute a quality score that incorporates ancestor model metrics:
  1. Compute the model’s own aggregate metric score
  2. Recursively walk the lineage tree to collect ancestor scores
  3. Combine the model’s score with the average ancestor score: (own_score + avg_ancestor_score) / 2
This approach rewards models with high-quality lineage while still prioritizing the model’s own metrics.

​
Example Request

curl -X GET https://api.example.com/artifact/model/42/lineage

​
Example Response

​
Model with Registered Parent

{
  "nodes": [
    {
      "artifact_id": "42",
      "name": "my-org/my-finetuned-model",
      "source": "config_json",
      "metadata": {}
    },
    {
      "artifact_id": "15",
      "name": "facebook/wav2vec2-base",
      "source": "config_json",
      "metadata": {}
    }
  ],
  "edges": [
    {
      "from_node_artifact_id": "15",
      "to_node_artifact_id": "42",
      "relationship": "base_model"
    }
  ]
}

​
Model with External Parent

{
  "nodes": [
    {
      "artifact_id": "42",
      "name": "my-org/custom-bert",
      "source": "config_json",
      "metadata": {}
    },
    {
      "artifact_id": "external:bert-base-uncased",
      "name": "bert-base-uncased",
      "source": "config_json",
      "metadata": {
        "external": true
      }
    }
  ],
  "edges": [
    {
      "from_node_artifact_id": "external:bert-base-uncased",
      "to_node_artifact_id": "42",
      "relationship": "base_model"
    }
  ]
}

​
Model with No Parents

{
  "nodes": [
    {
      "artifact_id": "42",
      "name": "my-org/standalone-model",
      "source": "config_json",
      "metadata": {}
    }
  ],
  "edges": []
}

​
Error Responses

​
404
error
Artifact not found - the specified model ID does not exist in the registry

​
Lineage Graph Structure

The lineage graph follows these principles:
  • Root node first: The requested model is always the first node in the array
  • Directed edges: Edges point from parent (base model) to child (fine-tuned model)
  • External dependencies: Models not in the registry are marked with artifact_id prefix “external:” and metadata.external = true
  • Single-level depth: Currently returns immediate parents only (not recursive ancestors)
  • Deterministic: Same model ID always returns the same graph structure

​
Use Cases

  • Model provenance: Track which base models were used to create fine-tuned models
  • Quality assessment: Use lineage in combination with tree_score to evaluate model trustworthiness
  • Dependency analysis: Identify external dependencies that may affect reproducibility
  • Visualization: Build model family trees and lineage diagrams

Build docs developers (and LLMs) love

Get started for freeTalk to us