Release v0.0.2 - Phase 6 complete: local models & observability

Features:
- Add LlamaCppProvider for llama.cpp GGUF model support
- Add QuantizationConfig for quantization settings
- Add observability module: logging, metrics, tracing
- Add OpenTelemetry exporter integration
- Fix thread safety in CognitiveContext (threading.local)
- Fix various bugs in runtime, scheduler, validation

Bug fixes:
- determinism.py: fix indentation in deterministic_call
- runtime.py: thread-safe context with stack structure
- scheduler.py: add _current_coro attribute to ScheduledTask
- validation.py: add max_results to FactCheckValidator
- cli.py: add --api-key parameter to repl command
- pyproject.toml: add aiohttp dependency

Tests: 241 passed, 2 skipped

Author: ChidcGithub

codegnipy/__init__.py

@@ -88,6 +88,8 @@
     AnthropicProvider,
     OllamaProvider,
     TransformersProvider,
+    LlamaCppProvider,
+    QuantizationConfig,
     ProviderFactory,
     create_provider
 )
@@ -104,6 +106,22 @@
     verify_claim,
     verify_claim_async
 )
+from .observability import (
+    LogLevel,
+    MetricType,
+    SpanContext,
+    Metric,
+    CognitiveLogger,
+    MetricsCollector,
+    Tracer,
+    OpenTelemetryExporter,
+    ObservabilityManager,
+    traced,
+    logged,
+    metered,
+    get_default_manager,
+    configure_observability,
+)
 
 __all__ = [
     # Core
@@ -173,6 +191,8 @@
     "AnthropicProvider",
     "OllamaProvider",
     "TransformersProvider",
+    "LlamaCppProvider",
+    "QuantizationConfig",
     "ProviderFactory",
     "create_provider",
     # Validation
@@ -187,4 +207,19 @@
     "create_default_validator",
     "verify_claim",
     "verify_claim_async",
+    # Observability
+    "LogLevel",
+    "MetricType",
+    "SpanContext",
+    "Metric",
+    "CognitiveLogger",
+    "MetricsCollector",
+    "Tracer",
+    "OpenTelemetryExporter",
+    "ObservabilityManager",
+    "traced",
+    "logged",
+    "metered",
+    "get_default_manager",
+    "configure_observability",
 ]

codegnipy/cli.py

@@ -43,6 +43,10 @@ def create_parser() -> argparse.ArgumentParser:
         default="gpt-4o-mini",
         help="使用的 LLM 模型"
     )
+    repl_parser.add_argument(
+        "--api-key", "-k",
+        help="API 密钥（也可通过环境变量 OPENAI_API_KEY 设置）"
+    )
 
     # version 命令
     subparsers.add_parser("version", help="显示版本信息")
@@ -86,17 +90,17 @@ def run_file(filepath: str, model: str, api_key: Optional[str] = None):
             sys.exit(1)
 
 
-def start_repl(model: str):
+def start_repl(model: str, api_key: Optional[str] = None):
     """启动交互式 REPL"""
     import code
-    
+
     print("Codegnipy REPL")
     print(f"模型: {model}")
     print("输入 Python 代码，~\"prompt\" 语法将调用 LLM")
     print("输入 exit() 或 Ctrl+D 退出\n")
-    
+
     # 创建上下文
-    ctx = CognitiveContext(model=model)
+    ctx = CognitiveContext(model=model, api_key=api_key)
     ctx.__enter__()
 
     # 准备 REPL 环境
@@ -137,11 +141,11 @@ def main():
     """主入口"""
     parser = create_parser()
     args = parser.parse_args()
-    
+
     if args.command == "run":
         run_file(args.file, args.model, args.api_key)
     elif args.command == "repl":
-        start_repl(args.model)
+        start_repl(args.model, getattr(args, 'api_key', None))
     elif args.command == "version":
         print(f"Codegnipy v{codegnipy.__version__}")
     else:

codegnipy/determinism.py

@@ -558,54 +558,34 @@ def set_external_validator(self, validator: "BaseValidator") -> None:
 # ============ 确定性认知调用 ============
 
 def deterministic_call(
-
     prompt: str,
-
     constraint: TypeConstraint,
-
     context: Optional["CognitiveContext"] = None,
-
     *,
-
     max_attempts: int = 3,
-
     use_reflection: bool = False,
-
     simulator: Optional[Simulator] = None
-
 ) -> ValidationResult:
-
     """
-
     带类型约束的确定性认知调用
 
-
-
     参数:
-
         prompt: 提示文本
-
         constraint: 类型约束
-
         context: 认知上下文
-
         max_attempts: 最大尝试次数
-
         use_reflection: 是否使用反思
-
         simulator: 模拟器（用于测试）
 
     返回:
-
         ValidationResult 对象
-
     """
     from .runtime import cognitive_call
     from .reflection import with_reflection
-    
+
     # 构建带约束的提示
     constrained_prompt = f"{prompt}\n\n约束: {constraint.to_prompt()}\n\n请严格按照约束要求回答。"
-    
+
     for attempt in range(max_attempts):
         # 获取响应
         if simulator and simulator.mode != SimulationMode.OFF:
@@ -616,16 +596,20 @@ def deterministic_call(
                 response = result.corrected_response or result.original_response
             else:
                 response = cognitive_call(constrained_prompt, context)
-        
+
         # 验证响应
         validation = constraint.validate(response)
-        
+
         if validation.status == ValidationStatus.VALID:
             return validation
-        
+
         # 如果验证失败，添加反馈并重试
         if attempt < max_attempts - 1:
             error_feedback = "; ".join(validation.errors)
-            constrained_prompt = f"{prompt}\n\n约束: {constraint.to_prompt()}\n\n上次的回答不符合要求，错误: {error_feedback}\n\n请修正后重新回答。"
-    
+            constrained_prompt = (
+                f"{prompt}\n\n约束: {constraint.to_prompt()}\n\n"
+                f"上次的回答不符合要求，错误: {error_feedback}\n\n"
+                f"请修正后重新回答。"
+            )
+
     return validation