Context Compression Techniques

123456789101112
from llmlingua import PromptCompressor

# Initialize compressor with a small model for speed
compressor = PromptCompressor(
    model_name="microsoft/llmlingua-2-bert-base-multilingual-cased-meetingbank",
    use_llmlingua2=True,
    device_map="cuda"  # GPU acceleration recommended
)

def compress_context(context: str, target_ratio: float = 0.5) -> dict:
    """
    Compress context while preserving semantic meaning.

123456789101112
from dataclasses import dataclass
from typing import List, Tuple
import numpy as np
from sentence_transformers import SentenceTransformer

@dataclass
class ContextChunk:
    content: str
    source: str
    timestamp: float
    token_count: int
    

123456789101112
from typing import Dict, List, Optional
import hashlib
from dataclasses import dataclass

@dataclass
class SummaryNode:
    content: str
    level: int  # 0 = root, higher = more detailed
    token_count: int
    children: List['SummaryNode']
    source_hash: str
    embedding: Optional[List[float]] = None

123456789101112
import numpy as np
from typing import List, Tuple
from sentence_transformers import SentenceTransformer

class SelectiveContextCompressor:
    def __init__(self, model_name: str = 'all-MiniLM-L6-v2'):
        self.encoder = SentenceTransformer(model_name)
        self.relevance_cache = {}
    
    def compress(self, 
                 query: str, 
                 context_chunks: List[str],

123456789101112
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from typing import List, Tuple

class TokenImportanceCompressor:
    def __init__(self, model_name: str = 'gpt2'):
        self.tokenizer = AutoTokenizer.from_pretrained(model_name)
        self.model = AutoModelForCausalLM.from_pretrained(model_name)
        self.model.eval()
    
    def calculate_token_importance(self, text: str) -> List[Tuple[str, float]]:
        """

123456789101112
from typing import List, Dict, Optional
import asyncio
from dataclasses import dataclass

@dataclass
class SummaryResult:
    text: str
    level: int
    source_tokens: int
    output_tokens: int
    quality_score: float
    metadata: Dict