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.

​
About This Endpoint

Based on the current codebase implementation, there is no separate /artifact/ingest endpoint.The ingestion functionality is integrated into the main artifact creation endpoint:POST /artifact/modelWhen you create a model artifact using POST /artifact/model, the system automatically triggers the full ingestion pipeline described below.

​
How Model Ingestion Works

When you create a model artifact via POST /artifact/model, the system automatically performs the following ingestion steps:

​
1. HuggingFace Model Normalization

The provided URL or model ID is normalized to a canonical HuggingFace identifier:
  • https://huggingface.co/google-bert/bert-base-uncased → google-bert/bert-base-uncased
  • google-bert/bert-base-uncased → google-bert/bert-base-uncased

​
2. License Extraction

The system fetches license information from the HuggingFace API: 
GET https://huggingface.co/api/models/{model_id}
Extracts the license field and normalizes it to lowercase SPDX format (e.g., apache-2.0, mit).

​
3. Trustworthiness Metrics Computation

The ingestion pipeline computes comprehensive trust metrics:
​
net_score
float
Overall trustworthiness score (0.0 - 1.0)
​
reviewedness
float
Code review and peer validation scoreCritical: Models with reviewedness < 0.5 are rejected with HTTP 424
​
reproducibility
float
Reproducibility and documentation quality score
​
ramp_up_time
float
Ease of understanding and adoption
​
bus_factor
float
Project sustainability and contributor diversity
​
performance_claims
float
Validation of performance benchmarks
​
dataset_and_code_score
float
Quality of training data and code
​
dataset_quality
float
Training dataset quality assessment
​
code_quality
float
Code quality and maintainability
​
tree_score
float
Dependency tree health score
​
size_score
object
Model size suitability for different hardware platforms
​
raspberry_pi
float
Score for Raspberry Pi deployment
​
jetson_nano
float
Score for Jetson Nano deployment
​
desktop_pc
float
Score for desktop PC deployment
​
aws_server
float
Score for cloud server deployment

​
4. Lineage Extraction

The system extracts parent model dependencies from:
  • Model card metadata
  • HuggingFace API responses
  • Repository configuration files
These dependencies are used to build the lineage graph accessible via /artifact/model/{id}/lineage.

​
5. Reviewedness Gate

Automatic Rejection: If the computed reviewedness score is less than 0.5, the ingestion fails with:
  • HTTP Status: 424 Failed Dependency
  • Error: "Ingest rejected: reviewedness={score} < 0.50"
This ensures only well-reviewed, trustworthy models are accepted into the registry.

​
6. Artifact Storage

After passing validation:
  1. A minimal artifact ZIP is created containing source_url.txt
  2. The ZIP is uploaded to S3 at artifacts/model/{id}.zip
  3. A pre-signed download URL is generated (valid for limited time)
  4. The download URL is stored in artifact metadata

​
Usage

To trigger ingestion, use the standard artifact creation endpoint: 
curl -X POST https://api.example.com/artifact/model \
  -H "Content-Type: application/json" \
  -d '{
    "url": "google-bert/bert-base-uncased",
    "name": "BERT Base Uncased"
  }'

​
Response

Successful ingestion returns HTTP 201 with full artifact details including computed metrics: 
{
  "metadata": {
    "name": "BERT Base Uncased",
    "id": "12345",
    "type": "model"
  },
  "data": {
    "url": "https://huggingface.co/google-bert/bert-base-uncased",
    "download_url": "https://s3.amazonaws.com/bucket/artifacts/model/12345.zip?signature=..."
  }
}
The artifact metadata stored in the registry includes all computed metrics, which can be retrieved via:
  • GET /artifact/model/{id}/rate - View all trust metrics
  • GET /artifacts/model/{id} - View artifact with embedded metadata

​
Error Codes

  • 400 - Invalid HuggingFace model ID or URL
  • 424 - Reviewedness score below threshold (< 0.5)
  • 502 - Unable to fetch HuggingFace model metadata
  • 500 - Internal error during metric computation or storage

​
Implementation Details

The ingestion logic is implemented in /home/daytona/workspace/source/src/api/routers/models.py:322: 
def _ingest_hf_core(source_url: str) -> Dict[str, Any]:
    # 1. Normalize HF ID
    hf_id = _hf_id_from_url_or_id(source_url)
    
    # 2. Fetch license
    hf_license = _fetch_hf_license(hf_id)
    
    # 3. Compute metrics
    metrics = compute_metrics_for_model(base_resource)
    
    # 4. Reviewedness gate
    if reviewedness < 0.5:
        raise HTTPException(status_code=424, ...)
    
    # 5. Extract lineage
    enriched = _build_hf_resource(hf_id)
    parents = extract_parents_from_resource(enriched)
    
    # 6. Create registry entry
    created = _registry.create(mc)
    
    # 7. Upload artifact ZIP
    _storage.put_bytes(key, mem_zip.getvalue())
    
    return reg_item

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us