Spaces:

NYSERDA-CRE-Working-Group
/

Updated_code_complaince

Sleeping

Upload 70 files

e1ced8e verified 8 months ago

1.38 kB

import os
from dotenv import load_dotenv

# Explicit path so .env is found regardless of the working directory
_THIS_DIR = os.path.dirname(os.path.abspath(__file__))
load_dotenv(os.path.join(_THIS_DIR, ".env"))

# --- API Keys ---
GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY", "")
OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY", "")

# --- Model Names ---
PLANNER_MODEL = "gemini-3-flash-preview"
CROPPER_MODEL = "gemini-3-flash-preview"
ANNOTATOR_MODEL = "gemini-2.5-flash-image"
ANALYZER_MODEL = "gemini-3-flash-preview"
CODE_LOOKUP_MODEL = "gpt-5-mini"
DELIBERATION_MODEL = "gpt-5.2"
VERDICT_MODEL = "gemini-3-flash-preview"
METADATA_MODEL = "gemini-3-flash-preview"

# --- ChromaDB ---
CHROMA_DB_PATH = os.path.join(os.path.dirname(__file__), "data", "nyc_code_db")
CHROMA_COLLECTION_NAME = "nyc_building_codes"
EMBEDDING_MODEL_NAME = "BAAI/bge-large-en-v1.5"

# --- Processing Constants ---
PDF_RENDER_DPI = 100
MAX_INVESTIGATION_ROUNDS = 3
CROP_PADDING_PX = 40

# --- Code Lookup Budgets (per individual query, queries run in parallel) ---
MAX_DISCOVER_CALLS = 1
MAX_FETCH_CALLS = 1
MAX_CODE_LOOKUP_TURNS = 5

# --- Search Defaults ---
DISCOVER_N_RESULTS = 50    # Retrieve more candidates for re-ranking
RERANK_TOP_K = 20          # Return top-K after re-ranking
FETCH_MAX_SECTIONS = 30    # Max sections per chapter fetch

	import os
	from dotenv import load_dotenv

	# Explicit path so .env is found regardless of the working directory
	_THIS_DIR = os.path.dirname(os.path.abspath(__file__))
	load_dotenv(os.path.join(_THIS_DIR, ".env"))

	# --- API Keys ---
	GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY", "")
	OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY", "")

	# --- Model Names ---
	PLANNER_MODEL = "gemini-3-flash-preview"
	CROPPER_MODEL = "gemini-3-flash-preview"
	ANNOTATOR_MODEL = "gemini-2.5-flash-image"
	ANALYZER_MODEL = "gemini-3-flash-preview"
	CODE_LOOKUP_MODEL = "gpt-5-mini"
	DELIBERATION_MODEL = "gpt-5.2"
	VERDICT_MODEL = "gemini-3-flash-preview"
	METADATA_MODEL = "gemini-3-flash-preview"

	# --- ChromaDB ---
	CHROMA_DB_PATH = os.path.join(os.path.dirname(__file__), "data", "nyc_code_db")
	CHROMA_COLLECTION_NAME = "nyc_building_codes"
	EMBEDDING_MODEL_NAME = "BAAI/bge-large-en-v1.5"

	# --- Processing Constants ---
	PDF_RENDER_DPI = 100
	MAX_INVESTIGATION_ROUNDS = 3
	CROP_PADDING_PX = 40

	# --- Code Lookup Budgets (per individual query, queries run in parallel) ---
	MAX_DISCOVER_CALLS = 1
	MAX_FETCH_CALLS = 1
	MAX_CODE_LOOKUP_TURNS = 5

	# --- Search Defaults ---
	DISCOVER_N_RESULTS = 50 # Retrieve more candidates for re-ranking
	RERANK_TOP_K = 20 # Return top-K after re-ranking
	FETCH_MAX_SECTIONS = 30 # Max sections per chapter fetch