| import os |
| import time |
| from urllib.parse import urlsplit |
|
|
| import requests |
|
|
| from ._vectors import validate_vectors |
|
|
|
|
| def api_settings(): |
| endpoint = os.getenv('EMBEDDING_BASE_URL', '').rstrip('/') |
| model = os.getenv('EMBEDDING_MODEL', '') |
| if not endpoint or not model: |
| raise ValueError('Set EMBEDDING_BASE_URL and EMBEDDING_MODEL before calling the API.') |
| if not endpoint.endswith('/embeddings'): |
| endpoint += '/embeddings' |
| parsed = urlsplit(endpoint) |
| if parsed.scheme not in {'http', 'https'} or not parsed.hostname or parsed.username or parsed.password: |
| raise ValueError('Use an HTTP(S) endpoint without credentials in the URL.') |
| dimension = int(os.getenv('EMBEDDING_DIMENSIONS', '0')) |
| max_chars = int(os.getenv('EMBEDDING_MAX_CHARS', '8000')) |
| if dimension < 0 or max_chars < 1: |
| raise ValueError('Dimensions must be nonnegative and max chars positive.') |
| return {'endpoint': endpoint, 'model': model, 'dimensions': dimension, 'max_chars': max_chars, |
| 'prefix': os.getenv('EMBEDDING_DOCUMENT_PREFIX', ''), 'normalize': True} |
|
|
|
|
| def response_vectors(payload, count, dimension): |
| items = payload.get('data') |
| if not isinstance(items, list) or len(items) != count: |
| raise ValueError('Embedding response count does not match the request.') |
| indexes = [item.get('index') for item in items] |
| if any(type(index) is not int for index in indexes) or sorted(indexes) != list(range(count)): |
| raise ValueError('Embedding response indexes are missing, duplicated or out of range.') |
| ordered = sorted(items, key=lambda item: item['index']) |
| return validate_vectors([item['embedding'] for item in ordered], dimension, normalize=True) |
|
|
|
|
| def embed(settings, texts): |
| payload = {'model': settings['model'], 'input': [settings['prefix'] + text[:settings['max_chars']] for text in texts], 'encoding_format': 'float'} |
| if settings['dimensions']: |
| payload['dimensions'] = settings['dimensions'] |
| headers = {'Content-Type': 'application/json'} |
| if os.getenv('EMBEDDING_API_KEY'): |
| headers['Authorization'] = 'Bearer ' + os.environ['EMBEDDING_API_KEY'] |
| for attempt in range(5): |
| try: |
| response = requests.post(settings['endpoint'], json=payload, headers=headers, timeout=(15, 120)) |
| except (requests.ConnectionError, requests.Timeout): |
| if attempt == 4: |
| raise RuntimeError('Embedding endpoint connection failed after five attempts.') from None |
| else: |
| if response.status_code == 200: |
| return response_vectors(response.json(), len(texts), settings['dimensions']) |
| if response.status_code not in {408, 429, 500, 502, 503, 504} or attempt == 4: |
| raise RuntimeError(f'Embedding endpoint returned HTTP {response.status_code}; check model, credentials and limits.') |
| time.sleep(min(2 ** attempt, 16)) |
| raise RuntimeError('Embedding retries exhausted.') |
|
|
