Embeddings and moderation

Embeddings

Embeddings maps text to numeric vectors. Distances between vectors reflect semantic similarity, enabling text retrieval and matching by meaning.

Supported models

• Capability note: The platform provides /v1/embeddings. See the Model Catalog (Models in the console) for available embedding models.

Endpoint and parameters

POST https://api.yomiapi.com/v1/embeddings

Code examples

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.yomiapi.com/v1"
)
# Single text
response = client.embeddings.create(
    model="YOUR_AVAILABLE_EMBEDDING_MODEL_ID",
    input="The food at this restaurant is excellent and the service is attentive"
)
print(len(response.data[0].embedding))

# Batch text
batch = client.embeddings.create(
    model="YOUR_AVAILABLE_EMBEDDING_MODEL_ID",
    input=["Semantic search", "Vector database", "Retrieval-augmented generation"]
)
print(len(batch.data))

Applications

• Semantic search: Match documents by meaning rather than keywords.
• RAG systems: Retrieve knowledge-base content to improve model answers and reduce hallucinations.
• Clustering and classification: Group or label text data.
• Recommendation systems: Recommend similar content based on previous behavior.

Moderations

Moderations checks text for content such as violence, sexual content, hate speech, harassment and self-harm. It is compatible with the OpenAI Moderation format. Results include an overall decision and confidence scores for each category, so you can choose thresholds for your application.

Endpoint address

POST https://api.yomiapi.com/v1/moderations

Code examples

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.yomiapi.com/v1"
)
response = client.moderations.create(
    input="Text to check"
)
result = response.results[0]
print("Flagged:", result.flagged)
if result.flagged:
    print("Categories:", result.categories)
    print("Confidence:", result.category_scores)

Response fields

• flagged(boolean): Whether the content is flagged.
• categories(object): A true/false decision for each category.
• category_scores(object): Confidence scores for each category, from 0.0 to 1.0.

Applications

• UGC filtering: Automatically filter comments, posts and other user-generated content.
• AI output monitoring: Ensure generated content follows safety requirements.
• Realtime chat moderation: Check chat rooms and support conversations in real time.

Recommendations

• Prefer batches: Embeddings accepts arrays. Batch processing is more efficient than separate requests.
• Tune thresholds: Set custom moderation thresholds based on category_scores rather than relying only on flagged, balancing false positives against missed detections.