File size: 2,272 Bytes
1e23279 16c2a22 1e23279 16c2a22 9db586c 16c2a22 1e23279 16c2a22 1e23279 16c2a22 1e23279 16c2a22 1e23279 16c2a22 1e23279 16c2a22 1e23279 16c2a22 83ffe61 16c2a22 1e23279 16c2a22 1e23279 16c2a22 67befa7 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 |
"""Application-wide constants and configuration."""
import os
from typing import Final, List
# Model configuration
MODEL_NAME: Final[str] = "DragonLLM/Qwen-Open-Finance-R-8B"
# Cache directory - respect HF_HOME if set, otherwise use default
CACHE_DIR: Final[str] = os.getenv("HF_HOME", "/tmp/huggingface")
# Hugging Face token environment variable priority order
HF_TOKEN_VARS: Final[List[str]] = [
"HF_TOKEN_LC2",
"HF_TOKEN_LC",
"HF_TOKEN",
"HUGGING_FACE_HUB_TOKEN"
]
# French language detection patterns
FRENCH_PHRASES: Final[List[str]] = [
"en français",
"répondez en français",
"réponse française",
"répondez uniquement en français",
"expliquez en français",
]
FRENCH_CHARS: Final[List[str]] = [
"é", "è", "ê", "à", "ç", "ù", "ô", "î", "â", "û", "ë", "ï"
]
FRENCH_PATTERNS: Final[List[str]] = [
"qu'est-ce",
"qu'est",
"expliquez",
"comment",
"pourquoi",
"combien",
"quel",
"quelle",
"quels",
"quelles",
"où",
"quand",
"définissez",
]
FRENCH_SYSTEM_PROMPT: Final[str] = (
"Vous êtes un assistant financier expert. "
"Répondez TOUJOURS en français. "
"Soyez concis et précis dans vos explications. "
"Fournissez des réponses claires et complètes sans développements excessifs."
)
# Qwen3 EOS tokens
EOS_TOKENS: Final[List[int]] = [151645, 151643] # [<|im_end|>, <|endoftext|>]
PAD_TOKEN_ID: Final[int] = 151643 # <|endoftext|>
# Generation defaults
DEFAULT_MAX_TOKENS: Final[int] = 1000 # Increased for complete answers with concise reasoning
DEFAULT_TEMPERATURE: Final[float] = 0.7
DEFAULT_TOP_P: Final[float] = 1.0
DEFAULT_TOP_K: Final[int] = 20
REPETITION_PENALTY: Final[float] = 1.05
# Model initialization constants
MODEL_INIT_TIMEOUT_SECONDS = 300 # 5 minutes timeout for model initialization
MODEL_INIT_WAIT_INTERVAL_SECONDS = 1 # Check interval while waiting for initialization
# Rate limiting constants (for demo/single user)
RATE_LIMIT_REQUESTS_PER_MINUTE = 30 # 30 requests per minute (generous for single user)
RATE_LIMIT_REQUESTS_PER_HOUR = 500 # 500 requests per hour
# Confidence calculation constants
MIN_ANSWER_LENGTH_FOR_HIGH_CONFIDENCE = 50 # Minimum answer length for high confidence score
|