"""Compression configuration for LLMLingua-2."""
from dataclasses import dataclass, field
from typing import List
@dataclass
class CompressionConfig:
"""Configuration for context compression."""
# Threshold: Only compress if context exceeds this token count
compression_threshold: int = 4000
# Default compression rate (0.33 = 3x compression)
default_rate: float = 0.33
# Model selection: large (better quality) or small (less memory)
model_name: str = "microsoft/llmlingua-2-xlm-roberta-large-meetingbank"
# Device: "cuda", "cpu", or "auto"
device: str = "auto"
# Base tokens to always preserve (punctuation, structure)
base_force_tokens: List[str] = field(default_factory=lambda: [
".", "\n", "?", "!", ",", ":", # Punctuation
"(", ")", "{", "}", "[", "]", # Brackets
])
# Whether to drop consecutive duplicate tokens
drop_consecutive: bool = True
