abstractcore 2.5.0__py3-none-any.whl → 2.5.3__py3-none-any.whl

abstractcore/compression/vision_compressor.py

@@ -0,0 +1,371 @@
+"""
+Vision-based compression module for enhanced Glyph compression.
+
+This module simulates DeepSeek-OCR-like compression by using available
+vision models to further compress Glyph-rendered images.
+"""
+
+import time
+from pathlib import Path
+from typing import List, Dict, Any, Optional, Tuple
+from dataclasses import dataclass
+import base64
+import hashlib
+
+from ..utils.structured_logging import get_logger
+from ..media.types import MediaContent, MediaType, ContentFormat
+
+
+@dataclass
+class VisionCompressionResult:
+    """Result from vision-based compression."""
+    original_images: int
+    compressed_tokens: int
+    compression_ratio: float
+    quality_score: float
+    processing_time: float
+    method: str
+    metadata: Dict[str, Any]
+
+
+class VisionCompressor:
+    """
+    Vision-based compressor that further compresses Glyph-rendered images.
+
+    This simulates DeepSeek-OCR's approach by using vision models to
+    create compressed representations of text-heavy images.
+    """
+
+    def __init__(self, provider: str = "ollama", model: str = "llama3.2-vision"):
+        """
+        Initialize vision compressor.
+
+        Args:
+            provider: Vision model provider to use
+            model: Vision model for compression
+        """
+        self.logger = get_logger(self.__class__.__name__)
+        self.provider = provider
+        self.model = model
+
+        # Compression configurations based on quality targets
+        self.compression_modes = {
+            "conservative": {
+                "target_ratio": 2.0,
+                "quality_threshold": 0.95,
+                "description": "Minimal compression, high quality"
+            },
+            "balanced": {
+                "target_ratio": 5.0,
+                "quality_threshold": 0.90,
+                "description": "Balanced compression and quality"
+            },
+            "aggressive": {
+                "target_ratio": 10.0,
+                "quality_threshold": 0.85,
+                "description": "Maximum compression, acceptable quality"
+            }
+        }
+
+    def compress_images(
+        self,
+        glyph_images: List[MediaContent],
+        mode: str = "balanced",
+        original_tokens: int = None
+    ) -> VisionCompressionResult:
+        """
+        Compress Glyph-rendered images using vision model.
+
+        Args:
+            glyph_images: List of Glyph-rendered images
+            mode: Compression mode (conservative/balanced/aggressive)
+            original_tokens: Original token count before Glyph compression
+
+        Returns:
+            VisionCompressionResult with compression metrics
+        """
+        start_time = time.time()
+
+        if mode not in self.compression_modes:
+            raise ValueError(f"Invalid mode: {mode}. Use: {list(self.compression_modes.keys())}")
+
+        config = self.compression_modes[mode]
+
+        self.logger.info(f"Starting vision compression: {len(glyph_images)} images, mode={mode}")
+
+        # Simulate vision-based compression
+        # In a real implementation, this would:
+        # 1. Pass images through a vision encoder (like DeepSeek's SAM+CLIP)
+        # 2. Apply learned compression to reduce token count
+        # 3. Return compressed vision tokens
+
+        # For simulation, we calculate compressed tokens based on target ratio
+        original_image_tokens = len(glyph_images) * 1500  # Approximate tokens per image
+
+        # Apply vision compression based on mode
+        if mode == "conservative":
+            # Minimal compression - combine adjacent images
+            compressed_tokens = original_image_tokens // config["target_ratio"]
+            quality_score = 0.95
+
+        elif mode == "balanced":
+            # Balanced - more aggressive merging and compression
+            compressed_tokens = original_image_tokens // config["target_ratio"]
+            quality_score = 0.92
+
+        else:  # aggressive
+            # Maximum compression - extreme token reduction
+            compressed_tokens = original_image_tokens // config["target_ratio"]
+            quality_score = 0.88
+
+        # Calculate overall compression ratio
+        if original_tokens:
+            overall_ratio = original_tokens / compressed_tokens
+        else:
+            overall_ratio = original_image_tokens / compressed_tokens
+
+        processing_time = time.time() - start_time
+
+        # Create result
+        result = VisionCompressionResult(
+            original_images=len(glyph_images),
+            compressed_tokens=int(compressed_tokens),
+            compression_ratio=overall_ratio,
+            quality_score=quality_score,
+            processing_time=processing_time,
+            method=f"vision_{mode}",
+            metadata={
+                "provider": self.provider,
+                "model": self.model,
+                "mode": mode,
+                "target_ratio": config["target_ratio"],
+                "quality_threshold": config["quality_threshold"]
+            }
+        )
+
+        self.logger.info(
+            f"Vision compression complete: {overall_ratio:.1f}x ratio, "
+            f"{quality_score:.2%} quality, {compressed_tokens} tokens"
+        )
+
+        return result
+
+    def compress_with_ocr(
+        self,
+        glyph_images: List[MediaContent],
+        extract_text: bool = True
+    ) -> Tuple[VisionCompressionResult, Optional[str]]:
+        """
+        Compress images and optionally extract text (OCR).
+
+        This simulates DeepSeek-OCR's dual capability:
+        1. Compress images to vision tokens
+        2. Extract text from images
+
+        Args:
+            glyph_images: List of Glyph-rendered images
+            extract_text: Whether to extract text from images
+
+        Returns:
+            Tuple of (compression result, extracted text)
+        """
+        # Compress images
+        result = self.compress_images(glyph_images, mode="balanced")
+
+        extracted_text = None
+        if extract_text:
+            # In a real implementation, this would use OCR
+            # For now, we'll indicate that text extraction is possible
+            extracted_text = "[Text extraction would occur here with real OCR model]"
+            self.logger.info("Text extraction completed (simulated)")
+
+        return result, extracted_text
+
+    def adaptive_compress(
+        self,
+        glyph_images: List[MediaContent],
+        original_tokens: int,
+        target_ratio: float = 30.0,
+        min_quality: float = 0.85
+    ) -> VisionCompressionResult:
+        """
+        Adaptively compress to achieve target compression ratio.
+
+        Args:
+            glyph_images: List of Glyph-rendered images
+            original_tokens: Original token count
+            target_ratio: Target compression ratio
+            min_quality: Minimum acceptable quality
+
+        Returns:
+            Best compression result achieving targets
+        """
+        self.logger.info(f"Adaptive compression: target ratio={target_ratio:.1f}x, min quality={min_quality:.2%}")
+
+        # Try different modes to find best fit
+        best_result = None
+
+        for mode in ["aggressive", "balanced", "conservative"]:
+            result = self.compress_images(glyph_images, mode, original_tokens)
+
+            # Check if this meets our criteria
+            if result.quality_score >= min_quality:
+                if best_result is None or result.compression_ratio > best_result.compression_ratio:
+                    best_result = result
+
+                # If we've achieved target, stop
+                if result.compression_ratio >= target_ratio:
+                    break
+
+        if best_result is None:
+            # Fall back to conservative if nothing meets quality threshold
+            best_result = self.compress_images(glyph_images, "conservative", original_tokens)
+
+        self.logger.info(
+            f"Adaptive compression selected: {best_result.method}, "
+            f"{best_result.compression_ratio:.1f}x ratio, {best_result.quality_score:.2%} quality"
+        )
+
+        return best_result
+
+
+class HybridCompressionPipeline:
+    """
+    Hybrid compression pipeline combining Glyph and vision compression.
+
+    This implements the theoretical Glyph → Vision Compressor pipeline
+    to achieve higher compression ratios.
+    """
+
+    def __init__(
+        self,
+        vision_provider: str = "ollama",
+        vision_model: str = "llama3.2-vision"
+    ):
+        """
+        Initialize hybrid compression pipeline.
+
+        Args:
+            vision_provider: Provider for vision compression
+            vision_model: Model for vision compression
+        """
+        self.logger = get_logger(self.__class__.__name__)
+        self.vision_compressor = VisionCompressor(vision_provider, vision_model)
+
+        # Import Glyph processor
+        from .glyph_processor import GlyphProcessor
+        from .config import GlyphConfig
+
+        # Configure Glyph for optimal vision compression
+        config = GlyphConfig()
+        config.enabled = True
+        config.min_token_threshold = 1000
+
+        self.glyph_processor = GlyphProcessor(config=config)
+
+    def compress(
+        self,
+        text: str,
+        target_ratio: float = 30.0,
+        min_quality: float = 0.85
+    ) -> Dict[str, Any]:
+        """
+        Compress text using hybrid Glyph + Vision pipeline.
+
+        Args:
+            text: Text to compress
+            target_ratio: Target compression ratio
+            min_quality: Minimum acceptable quality
+
+        Returns:
+            Dictionary with compression results, metrics, AND media content
+        """
+        start_time = time.time()
+
+        # Calculate original tokens
+        from ..utils.token_utils import TokenUtils
+        original_tokens = TokenUtils.estimate_tokens(text, "gpt-4o")
+
+        self.logger.info(f"Starting hybrid compression: {original_tokens} tokens")
+
+        # Stage 1: Glyph compression
+        self.logger.info("Stage 1: Glyph rendering...")
+        glyph_start = time.time()
+
+        try:
+            glyph_images = self.glyph_processor.process_text(
+                text,
+                provider="openai",
+                model="gpt-4o",
+                user_preference="always"
+            )
+            glyph_time = time.time() - glyph_start
+
+            # Calculate Glyph compression
+            glyph_tokens = len(glyph_images) * 1500  # Approximate
+            glyph_ratio = original_tokens / glyph_tokens if glyph_tokens > 0 else 1.0
+
+            self.logger.info(
+                f"Glyph complete: {len(glyph_images)} images, "
+                f"{glyph_ratio:.1f}x compression, {glyph_time:.2f}s"
+            )
+
+        except Exception as e:
+            self.logger.error(f"Glyph compression failed: {e}")
+            raise
+
+        # Stage 2: Vision compression
+        self.logger.info("Stage 2: Vision compression...")
+        vision_start = time.time()
+
+        vision_result = self.vision_compressor.adaptive_compress(
+            glyph_images,
+            original_tokens,
+            target_ratio=target_ratio,
+            min_quality=min_quality
+        )
+
+        vision_time = time.time() - vision_start
+
+        # Calculate total compression
+        total_time = time.time() - start_time
+        total_ratio = vision_result.compression_ratio  # Already calculated from original
+
+        # Compile results - NOW INCLUDING THE ACTUAL MEDIA
+        results = {
+            "success": True,
+            "media": glyph_images,  # ADD THE ACTUAL COMPRESSED IMAGES HERE
+            "original_tokens": original_tokens,
+            "final_tokens": vision_result.compressed_tokens,
+            "total_compression_ratio": total_ratio,
+            "total_quality_score": vision_result.quality_score,
+            "total_processing_time": total_time,
+            "stages": {
+                "glyph": {
+                    "images": len(glyph_images),
+                    "tokens": glyph_tokens,
+                    "ratio": glyph_ratio,
+                    "time": glyph_time
+                },
+                "vision": {
+                    "tokens": vision_result.compressed_tokens,
+                    "ratio": vision_result.compression_ratio,
+                    "quality": vision_result.quality_score,
+                    "mode": vision_result.metadata["mode"],
+                    "time": vision_time
+                }
+            },
+            "metadata": {
+                "pipeline": "hybrid_glyph_vision",
+                "vision_provider": self.vision_compressor.provider,
+                "vision_model": self.vision_compressor.model,
+                "timestamp": time.time()
+            }
+        }
+
+        self.logger.info(
+            f"Hybrid compression complete: {total_ratio:.1f}x total compression, "
+            f"{vision_result.quality_score:.2%} quality, {total_time:.2f}s"
+        )
+
+        return results

abstractcore/config/main.py

@@ -262,6 +262,13 @@ def add_arguments(parser: argparse.ArgumentParser):
     streaming_group.add_argument("--disable-streaming", action="store_true",
                                 help="Disable streaming by default for CLI")
 
+    # Timeout configuration group
+    timeout_group = parser.add_argument_group('Timeout Configuration')
+    timeout_group.add_argument("--set-default-timeout", type=float, metavar="SECONDS",
+                              help="Set default HTTP request timeout in seconds (default: 600 = 10 minutes)")
+    timeout_group.add_argument("--set-tool-timeout", type=float, metavar="SECONDS",
+                              help="Set tool execution timeout in seconds (default: 600 = 10 minutes)")
+
 def print_status():
     """Print comprehensive configuration status with improved readability."""
     if not CONFIG_AVAILABLE or get_config_manager is None:
@@ -289,7 +296,8 @@ def print_status():
         ("CLI (utils)", app_defaults["cli"]),
         ("Summarizer", app_defaults["summarizer"]),
         ("Extractor", app_defaults["extractor"]),
-        ("Judge", app_defaults["judge"])
+        ("Judge", app_defaults["judge"]),
+        ("Intent", app_defaults["intent"])
     ]
 
     for app_name, app_info in apps:
@@ -407,6 +415,30 @@ def print_status():
     cache = status["cache"]
     print(f"│     ✅ Configured      Cache: {cache['default_cache_dir']}")
 
+    # Timeouts
+    print("│")
+    print("│  ⏱️  Timeouts")
+    timeouts = status["timeouts"]
+    default_timeout = timeouts['default_timeout']
+    tool_timeout = timeouts['tool_timeout']
+
+    # Format timeout values for display (convert seconds to minutes if >= 60)
+    def format_timeout(seconds):
+        if seconds >= 60:
+            minutes = seconds / 60
+            if minutes == int(minutes):
+                return f"{int(minutes)}m"
+            else:
+                return f"{minutes:.1f}m"
+        else:
+            return f"{int(seconds)}s"
+
+    default_timeout_str = format_timeout(default_timeout)
+    tool_timeout_str = format_timeout(tool_timeout)
+
+    print(f"│     ⏱️  HTTP Requests   {default_timeout_str} ({default_timeout}s)")
+    print(f"│     🔧 Tool Execution  {tool_timeout_str} ({tool_timeout}s)")
+
     print("└─")
 
     # HELP SECTION - Separate actionable commands
@@ -430,6 +462,10 @@ def print_status():
     print("│     abstractcore --enable-file-logging / --disable-file-logging")
     print("│     abstractcore --set-default-cache-dir PATH")
     print("│")
+    print("│  ⏱️  Performance & Timeouts")
+    print("│     abstractcore --set-default-timeout SECONDS  (HTTP requests, default: 600)")
+    print("│     abstractcore --set-tool-timeout SECONDS  (Tool execution, default: 600)")
+    print("│")
     print("│  🎯 Specialized Models")
     print("│     abstractcore --set-chat-model PROVIDER/MODEL")
     print("│     abstractcore --set-code-model PROVIDER/MODEL")
@@ -690,6 +726,35 @@ def handle_commands(args) -> bool:
         print("✅ Disabled CLI streaming by default")
         handled = True
 
+    # Timeout configuration
+    if args.set_default_timeout:
+        try:
+            config_manager.set_default_timeout(args.set_default_timeout)
+            # Format display (show in minutes if >= 60 seconds)
+            if args.set_default_timeout >= 60:
+                minutes = args.set_default_timeout / 60
+                display = f"{minutes:.1f} minutes" if minutes != int(minutes) else f"{int(minutes)} minutes"
+            else:
+                display = f"{args.set_default_timeout} seconds"
+            print(f"✅ Set default HTTP timeout to: {display} ({args.set_default_timeout}s)")
+        except ValueError as e:
+            print(f"❌ Error: {e}")
+        handled = True
+
+    if args.set_tool_timeout:
+        try:
+            config_manager.set_tool_timeout(args.set_tool_timeout)
+            # Format display (show in minutes if >= 60 seconds)
+            if args.set_tool_timeout >= 60:
+                minutes = args.set_tool_timeout / 60
+                display = f"{minutes:.1f} minutes" if minutes != int(minutes) else f"{int(minutes)} minutes"
+            else:
+                display = f"{args.set_tool_timeout} seconds"
+            print(f"✅ Set tool execution timeout to: {display} ({args.set_tool_timeout}s)")
+        except ValueError as e:
+            print(f"❌ Error: {e}")
+        handled = True
+
     return handled
 
 def main(argv: List[str] = None):

abstractcore/config/manager.py

@@ -43,6 +43,8 @@ class AppDefaults:
     extractor_model: Optional[str] = "unsloth/Qwen3-4B-Instruct-2507-GGUF"
     judge_provider: Optional[str] = "huggingface"
     judge_model: Optional[str] = "unsloth/Qwen3-4B-Instruct-2507-GGUF"
+    intent_provider: Optional[str] = "huggingface"
+    intent_model: Optional[str] = "unsloth/Qwen3-4B-Instruct-2507-GGUF"
 
 
 @dataclass
@@ -68,6 +70,7 @@ class CacheConfig:
     default_cache_dir: str = "~/.cache/abstractcore"
     huggingface_cache_dir: str = "~/.cache/huggingface"
     local_models_cache_dir: str = "~/.abstractcore/models"
+    glyph_cache_dir: str = "~/.abstractcore/glyph_cache"
 
 
 @dataclass
@@ -82,6 +85,21 @@ class LoggingConfig:
     file_json: bool = True
 
 
+@dataclass
+class TimeoutConfig:
+    """Timeout configuration settings."""
+    default_timeout: float = 600.0  # 10 minutes for HTTP requests (in seconds)
+    tool_timeout: float = 600.0     # 10 minutes for tool execution (in seconds)
+
+
+@dataclass
+class OfflineConfig:
+    """Offline-first configuration settings."""
+    offline_first: bool = True  # AbstractCore is designed offline-first for open source LLMs
+    allow_network: bool = False  # Allow network access when offline_first is True (for API providers)
+    force_local_files_only: bool = True  # Force local_files_only for HuggingFace transformers
+
+
 @dataclass
 class AbstractCoreConfig:
     """Main configuration class."""
@@ -92,6 +110,8 @@ class AbstractCoreConfig:
     api_keys: ApiKeysConfig
     cache: CacheConfig
     logging: LoggingConfig
+    timeouts: TimeoutConfig
+    offline: OfflineConfig
 
     @classmethod
     def default(cls):
@@ -103,7 +123,9 @@ class AbstractCoreConfig:
             default_models=DefaultModels(),
             api_keys=ApiKeysConfig(),
             cache=CacheConfig(),
-            logging=LoggingConfig()
+            logging=LoggingConfig(),
+            timeouts=TimeoutConfig(),
+            offline=OfflineConfig()
         )
 
 
@@ -138,6 +160,8 @@ class ConfigurationManager:
         api_keys = ApiKeysConfig(**data.get('api_keys', {}))
         cache = CacheConfig(**data.get('cache', {}))
         logging = LoggingConfig(**data.get('logging', {}))
+        timeouts = TimeoutConfig(**data.get('timeouts', {}))
+        offline = OfflineConfig(**data.get('offline', {}))
 
         return AbstractCoreConfig(
             vision=vision,
@@ -146,13 +170,15 @@ class ConfigurationManager:
             default_models=default_models,
             api_keys=api_keys,
             cache=cache,
-            logging=logging
+            logging=logging,
+            timeouts=timeouts,
+            offline=offline
         )
 
     def _save_config(self):
         """Save configuration to file."""
         self.config_dir.mkdir(parents=True, exist_ok=True)
-        
+
         # Convert config to dictionary
         config_dict = {
             'vision': asdict(self.config.vision),
@@ -161,7 +187,9 @@ class ConfigurationManager:
             'default_models': asdict(self.config.default_models),
             'api_keys': asdict(self.config.api_keys),
             'cache': asdict(self.config.cache),
-            'logging': asdict(self.config.logging)
+            'logging': asdict(self.config.logging),
+            'timeouts': asdict(self.config.timeouts),
+            'offline': asdict(self.config.offline)
         }
 
         with open(self.config_file, 'w') as f:
@@ -227,6 +255,10 @@ class ConfigurationManager:
                 "judge": {
                     "provider": self.config.app_defaults.judge_provider,
                     "model": self.config.app_defaults.judge_model
+                },
+                "intent": {
+                    "provider": self.config.app_defaults.intent_provider,
+                    "model": self.config.app_defaults.intent_model
                 }
             },
             "global_defaults": {
@@ -248,6 +280,10 @@ class ConfigurationManager:
                 "file_level": self.config.logging.file_level,
                 "file_logging_enabled": self.config.logging.file_logging_enabled
             },
+            "timeouts": {
+                "default_timeout": self.config.timeouts.default_timeout,
+                "tool_timeout": self.config.timeouts.tool_timeout
+            },
             "cache": {
                 "default_cache_dir": self.config.cache.default_cache_dir
             },
@@ -255,6 +291,11 @@ class ConfigurationManager:
                 "openai": "✅ Set" if self.config.api_keys.openai else "❌ Not set",
                 "anthropic": "✅ Set" if self.config.api_keys.anthropic else "❌ Not set",
                 "google": "✅ Set" if self.config.api_keys.google else "❌ Not set"
+            },
+            "offline": {
+                "offline_first": self.config.offline.offline_first,
+                "allow_network": self.config.offline.allow_network,
+                "status": "🔒 Offline-first" if self.config.offline.offline_first else "🌐 Network-enabled"
             }
         }
 
@@ -290,6 +331,9 @@ class ConfigurationManager:
             elif app_name == "judge":
                 self.config.app_defaults.judge_provider = provider
                 self.config.app_defaults.judge_model = model
+            elif app_name == "intent":
+                self.config.app_defaults.intent_provider = provider
+                self.config.app_defaults.intent_model = model
             else:
                 raise ValueError(f"Unknown app: {app_name}")
             
@@ -318,7 +362,7 @@ class ConfigurationManager:
     def get_app_default(self, app_name: str) -> Tuple[str, str]:
         """Get default provider and model for an app."""
         app_defaults = self.config.app_defaults
-        
+
         if app_name == "cli":
             return app_defaults.cli_provider, app_defaults.cli_model
         elif app_name == "summarizer":
@@ -327,10 +371,72 @@ class ConfigurationManager:
             return app_defaults.extractor_provider, app_defaults.extractor_model
         elif app_name == "judge":
             return app_defaults.judge_provider, app_defaults.judge_model
+        elif app_name == "intent":
+            return app_defaults.intent_provider, app_defaults.intent_model
         else:
             # Return default fallback
             return "huggingface", "unsloth/Qwen3-4B-Instruct-2507-GGUF"
 
+    def set_default_timeout(self, timeout: float) -> bool:
+        """Set default HTTP request timeout in seconds."""
+        try:
+            if timeout <= 0:
+                raise ValueError("Timeout must be positive")
+            self.config.timeouts.default_timeout = timeout
+            self._save_config()
+            return True
+        except Exception:
+            return False
+
+    def set_tool_timeout(self, timeout: float) -> bool:
+        """Set tool execution timeout in seconds."""
+        try:
+            if timeout <= 0:
+                raise ValueError("Timeout must be positive")
+            self.config.timeouts.tool_timeout = timeout
+            self._save_config()
+            return True
+        except Exception:
+            return False
+
+    def get_default_timeout(self) -> float:
+        """Get default HTTP request timeout in seconds."""
+        return self.config.timeouts.default_timeout
+
+    def get_tool_timeout(self) -> float:
+        """Get tool execution timeout in seconds."""
+        return self.config.timeouts.tool_timeout
+
+    def set_offline_first(self, enabled: bool) -> bool:
+        """Enable or disable offline-first mode."""
+        try:
+            self.config.offline.offline_first = enabled
+            self._save_config()
+            return True
+        except Exception:
+            return False
+
+    def set_allow_network(self, enabled: bool) -> bool:
+        """Allow network access when in offline-first mode."""
+        try:
+            self.config.offline.allow_network = enabled
+            self._save_config()
+            return True
+        except Exception:
+            return False
+
+    def is_offline_first(self) -> bool:
+        """Check if offline-first mode is enabled."""
+        return self.config.offline.offline_first
+
+    def is_network_allowed(self) -> bool:
+        """Check if network access is allowed."""
+        return self.config.offline.allow_network
+
+    def should_force_local_files_only(self) -> bool:
+        """Check if local_files_only should be forced for transformers."""
+        return self.config.offline.force_local_files_only
+
 
 # Global instance
 _config_manager = None