Fix: revert google url to proper openai extension

Fix: remove trailing slash from google genai proxy base_url

Fix: brutally enforce base_url for Google GenAI in all OpenAI wrappers

Fix: allow token counting logic to intercept gemini model gracefully

Feat: Set Google Gemini GenerativeLanguage API endpoint as default

Fix: Strip empty `stop` parameter for Google Gemini fallback

Fix: Strip unsupported payload parameters for Google Gemini OpenAI-compatible proxy

Feat: Fallback to Google gemini-3-flash-preview on primary LLM proxy error

Fix: combine system messages to satisfy strict LLM chatml parser

Fix: allow token counting logic to intercept alias models gracefully

Fix: enforce system message to be the first in openai_utils requests

Upload folder using huggingface_hub

Fix: Replace deprecated alias-fast with alias-large in fallback cascade

Fix: map base_url explicitely to Helmholtz endpoint in OpenAI client constructor

Fix: Explicitly provide dummy token to OpenAI client constructor if environment vars not mapped early enough

Upload folder using huggingface_hub

Upload folder using huggingface_hub

Upload folder using huggingface_hub