api-embedding / src /utils /validators.py
fahmiaziz98
[UPDATE]: support openai compatible
155ad69
raw
history blame
4.52 kB
"""
Input validation utilities.
This module provides validation functions for request inputs,
ensuring data quality and preventing abuse.
"""
from typing import List, Dict, Any
from pydantic import BaseModel
from src.core.exceptions import TextTooLongError, BatchTooLargeError, ValidationError
def validate_text(text: str, max_length: int = 8192, allow_empty: bool = False) -> None:
"""
Validate a single text input.
Args:
text: Input text to validate
max_length: Maximum allowed text length
allow_empty: Whether to allow empty strings
Raises:
ValidationError: If text is empty and not allowed
TextTooLongError: If text exceeds max_length
"""
if not allow_empty and not text.strip():
raise ValidationError("text", "Text cannot be empty")
if len(text) > max_length:
raise TextTooLongError(len(text), max_length)
def validate_texts(
texts: List[str],
max_length: int = 8192,
max_batch_size: int = 100,
allow_empty: bool = False,
) -> None:
"""
Validate a list of text inputs.
Args:
texts: List of texts to validate
max_length: Maximum allowed length per text
max_batch_size: Maximum number of texts in batch
allow_empty: Whether to allow empty strings
Raises:
ValidationError: If texts list is empty or contains invalid items
BatchTooLargeError: If batch size exceeds max_batch_size
TextTooLongError: If any text exceeds max_length
"""
if not texts:
raise ValidationError("texts", "Texts list cannot be empty")
if len(texts) > max_batch_size:
raise BatchTooLargeError(len(texts), max_batch_size)
# Validate each text
for idx, text in enumerate(texts):
if not isinstance(text, str):
raise ValidationError(
f"texts[{idx}]", f"Expected string, got {type(text).__name__}"
)
if not allow_empty and not text.strip():
raise ValidationError(f"texts[{idx}]", "Text cannot be empty")
if len(text) > max_length:
raise TextTooLongError(len(text), max_length)
def validate_model_id(model_id: str, available_models: List[str]) -> None:
"""
Validate that a model_id exists in available models.
Args:
model_id: Model identifier to validate
available_models: List of available model IDs
Raises:
ValidationError: If model_id is invalid
"""
if not model_id:
raise ValidationError("model_id", "Model ID cannot be empty")
if model_id not in available_models:
raise ValidationError(
"model_id",
f"Model '{model_id}' not found. Available: {', '.join(available_models)}",
)
def extract_embedding_kwargs(request: BaseModel) -> Dict[str, Any]:
"""
Extract embedding kwargs from a request object.
This function extracts both the 'options' field and any extra fields
passed in the request, combining them into a single kwargs dict.
Args:
request: Pydantic request model (EmbedRequest or BatchEmbedRequest)
Returns:
Dictionary of kwargs to pass to embedding model
Example:
>>> request = EmbedRequest(
... texts=["hello"],
... model_id="qwen3-0.6b",
... options=EmbeddingOptions(normalize_embeddings=True),
... batch_size=32 # Extra field
... )
>>> extract_embedding_kwargs(request)
{'normalize_embeddings': True, 'batch_size': 32}
"""
kwargs = {}
# Extract from 'options' field if present
if hasattr(request, "options") and request.options is not None:
kwargs.update(request.options.to_kwargs())
# Extract extra fields (excluding standard fields)
standard_fields = {
"input",
"model",
"encoding_format",
"dimensions",
"user",
"options",
"query",
"documents",
"top_k",
}
request_dict = request.model_dump()
for key, value in request_dict.items():
if key not in standard_fields and value is not None:
kwargs[key] = value
return kwargs
def estimate_tokens(text: str) -> int:
"""Estimate token count (simple approximation)."""
# Simple heuristic: ~4 characters per token
return max(1, len(text) // 4)
def count_tokens_batch(texts: List[str]) -> int:
"""Count tokens for batch of texts."""
return sum(estimate_tokens(text) for text in texts)