isagellm-core 0.3.0.9__tar.gz → 0.4.0.0__tar.gz

{isagellm_core-0.3.0.9/src/isagellm_core.egg-info → isagellm_core-0.4.0.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: isagellm-core
-Version: 0.3.0.9
+Version: 0.4.0.0
 Summary: sageLLM core runtime with PD separation (MVP)
 Author: IntelliStream Team
 License: Proprietary - IntelliStream
@@ -13,10 +13,10 @@ Requires-Python: ==3.11.*
 Description-Content-Type: text/markdown
 Requires-Dist: pydantic>=2.0.0
 Requires-Dist: pyyaml>=6.0.0
-Requires-Dist: isagellm-protocol<0.4.0,>=0.3.0.2
-Requires-Dist: isagellm-backend<0.4.0,>=0.3.0.5
-Requires-Dist: isagellm-comm<0.4.0,>=0.3.0.1
-Requires-Dist: isagellm-kv-cache<0.4.0,>=0.3.0.1
+Requires-Dist: isagellm-protocol<0.5.0,>=0.3.0.2
+Requires-Dist: isagellm-backend<0.5.0,>=0.4.0.0
+Requires-Dist: isagellm-comm<0.5.0,>=0.4.0.0
+Requires-Dist: isagellm-kv-cache<0.5.0,>=0.3.0.1
 Requires-Dist: fastapi>=0.100.0
 Requires-Dist: uvicorn>=0.22.0
 Provides-Extra: dev
@@ -50,10 +50,10 @@ sageLLM Core - 引擎协调层与运行时系统
 ┌─────────────────────────────────────────────────────────────┐
 │  sagellm-core (引擎协调层) ← 本仓库                         │
 │  ┌─────────────────────────────────────────────────────┐    │
-│  │  Engine Abstraction (BaseEngine, EngineFactory)    │    │
-│  │  • CPUEngine, HFCudaEngine                         │    │
-│  │  • 自描述架构 (is_available, priority)              │    │
-│  │  • 引擎自动发现与注册                                │    │
+│  │  LLMEngine (Hardware-Agnostic, vLLM v1 style)      │    │
+│  │  • 统一推理接口: generate, stream, execute         │    │
+│  │  • 自动后端选择 (auto-detect cuda/ascend/cpu)      │    │
+│  │  • 配置驱动 (LLMEngineConfig)                       │    │
 │  └─────────────────────────────────────────────────────┘    │
 │  ┌─────────────────────────────────────────────────────┐    │
 │  │  Configuration System (config.py)                  │    │
@@ -68,9 +68,8 @@ sageLLM Core - 引擎协调层与运行时系统
 ```
 
 **职责分离**：
-- ❌ **Backend 不再包含**：BaseEngine, EngineFactory（已移至 core）
-- ✅ **Core 负责**：引擎接口、工厂、运行时、配置、协调
-- ✅ **Backend 负责**：硬件抽象、设备管理、内存原语
+- ✅ **Core 负责**：LLMEngine (硬件无关)、配置、协调
+- ✅ **Backend 负责**：硬件抽象、设备管理、Provider 实现
 
 ## Features
 

{isagellm_core-0.3.0.9 → isagellm_core-0.4.0.0}/README.md

@@ -19,10 +19,10 @@ sageLLM Core - 引擎协调层与运行时系统
 ┌─────────────────────────────────────────────────────────────┐
 │  sagellm-core (引擎协调层) ← 本仓库                         │
 │  ┌─────────────────────────────────────────────────────┐    │
-│  │  Engine Abstraction (BaseEngine, EngineFactory)    │    │
-│  │  • CPUEngine, HFCudaEngine                         │    │
-│  │  • 自描述架构 (is_available, priority)              │    │
-│  │  • 引擎自动发现与注册                                │    │
+│  │  LLMEngine (Hardware-Agnostic, vLLM v1 style)      │    │
+│  │  • 统一推理接口: generate, stream, execute         │    │
+│  │  • 自动后端选择 (auto-detect cuda/ascend/cpu)      │    │
+│  │  • 配置驱动 (LLMEngineConfig)                       │    │
 │  └─────────────────────────────────────────────────────┘    │
 │  ┌─────────────────────────────────────────────────────┐    │
 │  │  Configuration System (config.py)                  │    │
@@ -37,9 +37,8 @@ sageLLM Core - 引擎协调层与运行时系统
 ```
 
 **职责分离**：
-- ❌ **Backend 不再包含**：BaseEngine, EngineFactory（已移至 core）
-- ✅ **Core 负责**：引擎接口、工厂、运行时、配置、协调
-- ✅ **Backend 负责**：硬件抽象、设备管理、内存原语
+- ✅ **Core 负责**：LLMEngine (硬件无关)、配置、协调
+- ✅ **Backend 负责**：硬件抽象、设备管理、Provider 实现
 
 ## Features
 

{isagellm_core-0.3.0.9 → isagellm_core-0.4.0.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "isagellm-core"
-version = "0.3.0.9"
+version = "0.4.0.0"
 description = "sageLLM core runtime with PD separation (MVP)"
 readme = "README.md"
 requires-python = "==3.11.*"
@@ -20,10 +20,10 @@ classifiers = [
 dependencies = [
   "pydantic>=2.0.0",
   "pyyaml>=6.0.0",
-  "isagellm-protocol>=0.3.0.2,<0.4.0",
-  "isagellm-backend>=0.3.0.5,<0.4.0",
-  "isagellm-comm>=0.3.0.1,<0.4.0",
-  "isagellm-kv-cache>=0.3.0.1,<0.4.0",
+  "isagellm-protocol>=0.3.0.2,<0.5.0",
+  "isagellm-backend>=0.4.0.0,<0.5.0",
+  "isagellm-comm>=0.4.0.0,<0.5.0",
+  "isagellm-kv-cache>=0.3.0.1,<0.5.0",
   "fastapi>=0.100.0",
   "uvicorn>=0.22.0",
 ]
@@ -57,11 +57,18 @@ line-length = 100
 [project.scripts]
 sage-engine = "sagellm_core.engine_server:main"
 
-[project.entry-points."sagellm.engines"]
-cpu = "sagellm_core.engines.cpu:create_cpu_engine"
-hf-cuda = "sagellm_core.engines.hf_cuda:create_hf_cuda_engine"
-pytorch = "sagellm_core.engines.pytorch:create_pytorch_engine"
-ascend = "sagellm_core.engines.ascend:create_ascend_engine"
+# DEPRECATED: Old hardware-specific engines have been removed
+# Use LLMEngine with BackendProvider instead:
+#   from sagellm_core import LLMEngine, LLMEngineConfig
+#   engine = LLMEngine(LLMEngineConfig(model="...", backend="cpu"))
+#
+# The entry-points system is being phased out in favor of the unified
+# LLMEngine + BackendProvider architecture (vLLM v1 style).
+# [project.entry-points."sagellm.engines"]
+# cpu = "sagellm_core.engines.cpu:create_cpu_engine"
+# hf-cuda = "sagellm_core.engines.hf_cuda:create_hf_cuda_engine"
+# pytorch = "sagellm_core.engines.pytorch:create_pytorch_engine"
+# ascend = "sagellm_core.engines.ascend:create_ascend_engine"
 
 [tool.mypy]
 python_version = "3.10"

{isagellm_core-0.3.0.9 → isagellm_core-0.4.0.0/src/isagellm_core.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: isagellm-core
-Version: 0.3.0.9
+Version: 0.4.0.0
 Summary: sageLLM core runtime with PD separation (MVP)
 Author: IntelliStream Team
 License: Proprietary - IntelliStream
@@ -13,10 +13,10 @@ Requires-Python: ==3.11.*
 Description-Content-Type: text/markdown
 Requires-Dist: pydantic>=2.0.0
 Requires-Dist: pyyaml>=6.0.0
-Requires-Dist: isagellm-protocol<0.4.0,>=0.3.0.2
-Requires-Dist: isagellm-backend<0.4.0,>=0.3.0.5
-Requires-Dist: isagellm-comm<0.4.0,>=0.3.0.1
-Requires-Dist: isagellm-kv-cache<0.4.0,>=0.3.0.1
+Requires-Dist: isagellm-protocol<0.5.0,>=0.3.0.2
+Requires-Dist: isagellm-backend<0.5.0,>=0.4.0.0
+Requires-Dist: isagellm-comm<0.5.0,>=0.4.0.0
+Requires-Dist: isagellm-kv-cache<0.5.0,>=0.3.0.1
 Requires-Dist: fastapi>=0.100.0
 Requires-Dist: uvicorn>=0.22.0
 Provides-Extra: dev
@@ -50,10 +50,10 @@ sageLLM Core - 引擎协调层与运行时系统
 ┌─────────────────────────────────────────────────────────────┐
 │  sagellm-core (引擎协调层) ← 本仓库                         │
 │  ┌─────────────────────────────────────────────────────┐    │
-│  │  Engine Abstraction (BaseEngine, EngineFactory)    │    │
-│  │  • CPUEngine, HFCudaEngine                         │    │
-│  │  • 自描述架构 (is_available, priority)              │    │
-│  │  • 引擎自动发现与注册                                │    │
+│  │  LLMEngine (Hardware-Agnostic, vLLM v1 style)      │    │
+│  │  • 统一推理接口: generate, stream, execute         │    │
+│  │  • 自动后端选择 (auto-detect cuda/ascend/cpu)      │    │
+│  │  • 配置驱动 (LLMEngineConfig)                       │    │
 │  └─────────────────────────────────────────────────────┘    │
 │  ┌─────────────────────────────────────────────────────┐    │
 │  │  Configuration System (config.py)                  │    │
@@ -68,9 +68,8 @@ sageLLM Core - 引擎协调层与运行时系统
 ```
 
 **职责分离**：
-- ❌ **Backend 不再包含**：BaseEngine, EngineFactory（已移至 core）
-- ✅ **Core 负责**：引擎接口、工厂、运行时、配置、协调
-- ✅ **Backend 负责**：硬件抽象、设备管理、内存原语
+- ✅ **Core 负责**：LLMEngine (硬件无关)、配置、协调
+- ✅ **Backend 负责**：硬件抽象、设备管理、Provider 实现
 
 ## Features
 

isagellm_core-0.4.0.0/src/isagellm_core.egg-info/SOURCES.txt

@@ -0,0 +1,129 @@
+MANIFEST.in
+README.md
+pyproject.toml
+setup.py
+src/isagellm_core.egg-info/PKG-INFO
+src/isagellm_core.egg-info/SOURCES.txt
+src/isagellm_core.egg-info/dependency_links.txt
+src/isagellm_core.egg-info/entry_points.txt
+src/isagellm_core.egg-info/requires.txt
+src/isagellm_core.egg-info/top_level.txt
+src/sagellm_core/__init__.py
+src/sagellm_core/__init__.pyc
+src/sagellm_core/__main__.pyc
+src/sagellm_core/config.pyc
+src/sagellm_core/demo.pyc
+src/sagellm_core/engine.pyc
+src/sagellm_core/engine_factory.pyc
+src/sagellm_core/engine_server.pyc
+src/sagellm_core/factory.pyc
+src/sagellm_core/health.pyc
+src/sagellm_core/llm_engine.pyc
+src/sagellm_core/pd_executor.pyc
+src/sagellm_core/plugins.pyc
+src/sagellm_core/py.typed
+src/sagellm_core/runner.pyc
+src/sagellm_core/runtime.pyc
+src/sagellm_core/workload.pyc
+src/sagellm_core/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/__pycache__/base_engine.cpython-311.pyc
+src/sagellm_core/__pycache__/config.cpython-311.pyc
+src/sagellm_core/__pycache__/demo.cpython-311.pyc
+src/sagellm_core/__pycache__/engine.cpython-311.pyc
+src/sagellm_core/__pycache__/engine_factory.cpython-311.pyc
+src/sagellm_core/__pycache__/engine_server.cpython-311.pyc
+src/sagellm_core/__pycache__/factory.cpython-311.pyc
+src/sagellm_core/__pycache__/health.cpython-311.pyc
+src/sagellm_core/__pycache__/llm_engine.cpython-311.pyc
+src/sagellm_core/__pycache__/mock_engine.cpython-311.pyc
+src/sagellm_core/__pycache__/pd_executor.cpython-311.pyc
+src/sagellm_core/__pycache__/plugins.cpython-311.pyc
+src/sagellm_core/__pycache__/runner.cpython-311.pyc
+src/sagellm_core/__pycache__/runtime.cpython-311.pyc
+src/sagellm_core/__pycache__/workload.cpython-311.pyc
+src/sagellm_core/distributed/__init__.py
+src/sagellm_core/distributed/__init__.pyc
+src/sagellm_core/distributed/strategies.pyc
+src/sagellm_core/distributed/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/distributed/__pycache__/strategies.cpython-311.pyc
+src/sagellm_core/engine_core/__init__.py
+src/sagellm_core/engine_core/__init__.pyc
+src/sagellm_core/engine_core/engine_core.pyc
+src/sagellm_core/engine_core/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/engine_core/__pycache__/engine_core.cpython-311.pyc
+src/sagellm_core/engine_core/scheduler/__init__.py
+src/sagellm_core/engine_core/scheduler/__init__.pyc
+src/sagellm_core/engine_core/scheduler/scheduler.pyc
+src/sagellm_core/engine_core/scheduler/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/engine_core/scheduler/__pycache__/scheduler.cpython-311.pyc
+src/sagellm_core/engines/__init__.py
+src/sagellm_core/engines/__init__.pyc
+src/sagellm_core/engines/embedding.pyc
+src/sagellm_core/engines/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/engines/__pycache__/ascend.cpython-311.pyc
+src/sagellm_core/engines/__pycache__/cpu.cpython-311.pyc
+src/sagellm_core/engines/__pycache__/embedding.cpython-311.pyc
+src/sagellm_core/engines/__pycache__/hf_cuda.cpython-311.pyc
+src/sagellm_core/engines/__pycache__/mock.cpython-311.pyc
+src/sagellm_core/engines/__pycache__/pytorch.cpython-311.pyc
+src/sagellm_core/engines/__pycache__/pytorch_engine.cpython-311.pyc
+src/sagellm_core/executor/__init__.py
+src/sagellm_core/executor/__init__.pyc
+src/sagellm_core/executor/executor_base.pyc
+src/sagellm_core/executor/uniproc_executor.pyc
+src/sagellm_core/executor/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/executor/__pycache__/executor_base.cpython-311.pyc
+src/sagellm_core/executor/__pycache__/uniproc_executor.cpython-311.pyc
+src/sagellm_core/inputs/__init__.py
+src/sagellm_core/inputs/__init__.pyc
+src/sagellm_core/inputs/processor.pyc
+src/sagellm_core/inputs/tokenizer_utils.pyc
+src/sagellm_core/inputs/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/inputs/__pycache__/processor.cpython-311.pyc
+src/sagellm_core/inputs/__pycache__/tokenizer_utils.cpython-311.pyc
+src/sagellm_core/model/__init__.py
+src/sagellm_core/model/__init__.pyc
+src/sagellm_core/model/model_loader.pyc
+src/sagellm_core/model/weight_utils.pyc
+src/sagellm_core/model/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/model/__pycache__/model_loader.cpython-311.pyc
+src/sagellm_core/model/__pycache__/weight_utils.cpython-311.pyc
+src/sagellm_core/observability/__init__.py
+src/sagellm_core/observability/__init__.pyc
+src/sagellm_core/observability/logger.pyc
+src/sagellm_core/observability/metrics.pyc
+src/sagellm_core/observability/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/observability/__pycache__/logger.cpython-311.pyc
+src/sagellm_core/observability/__pycache__/metrics.cpython-311.pyc
+src/sagellm_core/sampling/__init__.py
+src/sagellm_core/sampling/__init__.pyc
+src/sagellm_core/sampling/params.pyc
+src/sagellm_core/sampling/sampler.pyc
+src/sagellm_core/sampling/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/sampling/__pycache__/params.cpython-311.pyc
+src/sagellm_core/sampling/__pycache__/sampler.cpython-311.pyc
+src/sagellm_core/worker/__init__.py
+src/sagellm_core/worker/__init__.pyc
+src/sagellm_core/worker/worker.pyc
+src/sagellm_core/worker/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/worker/__pycache__/worker.cpython-311.pyc
+src/sagellm_core/worker/model_runner/__init__.py
+src/sagellm_core/worker/model_runner/__init__.pyc
+src/sagellm_core/worker/model_runner/model_runner.pyc
+src/sagellm_core/worker/model_runner/__pycache__/__init__.cpython-311.pyc
+src/sagellm_core/worker/model_runner/__pycache__/model_runner.cpython-311.pyc
+tests/test_ci_smoke.py
+tests/test_config.py
+tests/test_e2e_llm_integration.py
+tests/test_engine.py
+tests/test_engine_behavior_parity.py
+tests/test_engine_contract_simplified.py
+tests/test_engine_server.py
+tests/test_llm_engine_contract.py
+tests/test_llm_engine_error_handling.py
+tests/test_model_loader.py
+tests/test_observability.py
+tests/test_pd_separation.py
+tests/test_sampling.py
+tests/test_streaming_pd.py
+tests/test_task0_10_workload.py

isagellm_core-0.4.0.0/src/isagellm_core.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+sage-engine = sagellm_core.engine_server:main

{isagellm_core-0.3.0.9 → isagellm_core-0.4.0.0}/src/isagellm_core.egg-info/requires.txt

@@ -1,9 +1,9 @@
 pydantic>=2.0.0
 pyyaml>=6.0.0
-isagellm-protocol<0.4.0,>=0.3.0.2
-isagellm-backend<0.4.0,>=0.3.0.5
-isagellm-comm<0.4.0,>=0.3.0.1
-isagellm-kv-cache<0.4.0,>=0.3.0.1
+isagellm-protocol<0.5.0,>=0.3.0.2
+isagellm-backend<0.5.0,>=0.4.0.0
+isagellm-comm<0.5.0,>=0.4.0.0
+isagellm-kv-cache<0.5.0,>=0.3.0.1
 fastapi>=0.100.0
 uvicorn>=0.22.0
 

isagellm_core-0.4.0.0/src/sagellm_core/__init__.py

@@ -0,0 +1,192 @@
+"""sageLLM Core 运行时。
+
+本包提供 sageLLM 的核心运行时组件：
+- LLMEngine: 统一的硬件无关推理引擎（vLLM v1 风格）
+- EngineCore: 协调 Scheduler 和 Executor
+- Scheduler: Continuous Batching 调度器
+- Executor: 管理 Worker 执行
+- Worker/ModelRunner: 模型前向传播
+- 配置 schema 与校验
+- Engine 工厂函数
+- 插件系统
+- Demo Runner
+- 分布式 Runtime（PD 分离 MVP）
+- PD 分离执行器
+
+Architecture (vLLM v1 style):
+    LLMEngine (hardware-agnostic)
+        ├── EngineCore (coordinates Scheduler and Executor)
+        │       └── Scheduler (Continuous Batching)
+        └── Executor
+                └── Worker
+                        └── ModelRunner
+                                ├── uses BackendProvider (from sagellm-backend)
+                                └── uses CommBackend (from sagellm-comm)
+"""
+
+from __future__ import annotations
+
+__version__ = "0.4.0.0"
+
+# ============================================================================
+# New Architecture (vLLM v1 style) - Hardware Agnostic
+# ============================================================================
+from sagellm_core.llm_engine import LLMEngine, LLMEngineConfig
+from sagellm_core.engine_core import EngineCore
+from sagellm_core.engine_core.engine_core import EngineCoreConfig
+from sagellm_core.engine_core.scheduler import (
+    ContinuousBatchingScheduler,
+    SchedulerConfig,
+    SchedulerOutput,
+)
+from sagellm_core.executor import ExecutorBase, UniprocExecutor
+from sagellm_core.executor.executor_base import ExecutorConfig
+from sagellm_core.worker import Worker
+from sagellm_core.worker.model_runner import ModelRunner
+
+# ============================================================================
+# Legacy Architecture (still supported, being refactored)
+# ============================================================================
+from sagellm_core.engine import BaseEngine, EngineInstanceConfig
+from sagellm_core.config import (
+    BackendConfig,
+    DemoConfig,
+    EngineConfig,
+    OutputConfig,
+    WorkloadConfig,
+    WorkloadSegment,
+    load_config,
+)
+from sagellm_core.demo import main as demo_main
+from sagellm_core.engine_factory import EngineFactory
+
+# Only EmbeddingEngine remains from legacy engines
+from sagellm_core.engines import (
+    EmbeddingEngine,
+    EmbeddingEngineConfig,
+)
+from sagellm_core.factory import create_backend, create_engine
+from sagellm_core.health import HealthStatus
+from sagellm_core.plugins import PluginResolutionError, list_entry_points, resolve_kind
+from sagellm_core.runner import DemoRunner, RunnerContext
+
+# PD 分离 MVP 模块
+from sagellm_core.runtime import DistributedConfig, DistributedRuntime, RuntimeState
+from sagellm_core.pd_executor import PDExecutionContext, PDSeparatedExecutor
+
+# Engine HTTP Server
+from sagellm_core.engine_server import app as engine_server_app
+from sagellm_core.engine_server import main as serve_engine
+
+# ============================================================================
+# Phase 2: New Modules (P2 Priority)
+# ============================================================================
+# Model loading utilities
+from sagellm_core.model import ModelLoader, load_model
+
+# Input processing
+from sagellm_core.inputs import InputProcessor, ProcessedInput, TokenizerWrapper
+
+# Sampling utilities
+from sagellm_core.sampling import SamplingParams, Sampler, GreedySampler
+
+# Distributed strategies
+from sagellm_core.distributed import DistributedStrategy, TensorParallelStrategy
+
+# Observability
+from sagellm_core.observability import MetricsCollector, EngineMetrics, setup_logger
+
+# PyTorch engine (optional, loaded lazily)
+PyTorchEngine = None
+create_pytorch_engine = None
+
+# Optional PyTorchEngine import (deprecated, use LLMEngine)
+# try:
+#     from sagellm_core.engines.pytorch_engine import (
+#         PyTorchEngine,
+#         create_pytorch_engine,
+#     )
+# except ImportError:
+#     pass  # torch or transformers not available
+
+# =========================================================================
+# DEPRECATED: Old hardware-specific engines have been removed
+# Use LLMEngine instead:
+#   from sagellm_core import LLMEngine, LLMEngineConfig
+#   engine = LLMEngine(LLMEngineConfig(model="..."))
+#
+# The following engines no longer exist:
+#   - CPUEngine → use LLMEngine(backend="cpu")
+#   - HFCudaEngine → use LLMEngine(backend="cuda")
+#   - AscendEngine → use LLMEngine(backend="ascend")
+#   - PyTorchEngine → use LLMEngine
+#
+# EmbeddingEngine is still available for embedding-only use cases.
+# =========================================================================
+
+# Version is defined at the top of the file (line 29)
+
+__all__ = [
+    # Version
+    "__version__",
+    # =========================================================================
+    # New Architecture (vLLM v1 style) - RECOMMENDED
+    # =========================================================================
+    # LLMEngine - Unified hardware-agnostic engine
+    "LLMEngine",
+    "LLMEngineConfig",
+    # EngineCore - Coordinates Scheduler and Executor
+    "EngineCore",
+    "EngineCoreConfig",
+    # Scheduler - Continuous Batching
+    "ContinuousBatchingScheduler",
+    "SchedulerConfig",
+    "SchedulerOutput",
+    # Executor - Manages Workers
+    "ExecutorBase",
+    "ExecutorConfig",
+    "UniprocExecutor",
+    # Worker - Model execution
+    "Worker",
+    "ModelRunner",
+    # =========================================================================
+    # Configuration (for YAML/config files)
+    # =========================================================================
+    "BackendConfig",
+    "DemoConfig",
+    "EngineConfig",
+    "OutputConfig",
+    "WorkloadConfig",
+    "WorkloadSegment",
+    "load_config",
+    # Engine abstraction
+    "BaseEngine",
+    "EngineInstanceConfig",  # For runtime engine instantiation
+    "HealthStatus",
+    # Engine implementations
+    # DEPRECATED: Old engines removed, use LLMEngine instead
+    # Only EmbeddingEngine remains for embedding-only use cases
+    "EmbeddingEngine",
+    "EmbeddingEngineConfig",
+    # Factory functions
+    "create_backend",
+    "create_engine",
+    "EngineFactory",
+    # Plugin system
+    "PluginResolutionError",
+    "list_entry_points",
+    "resolve_kind",
+    # Demo runner
+    "demo_main",
+    "DemoRunner",
+    "RunnerContext",
+    # PD Separation MVP
+    "DistributedConfig",
+    "DistributedRuntime",
+    "RuntimeState",
+    "PDExecutionContext",
+    "PDSeparatedExecutor",
+    # Engine HTTP Server
+    "engine_server_app",
+    "serve_engine",
+]

isagellm_core-0.4.0.0/src/sagellm_core/distributed/__init__.py

@@ -0,0 +1,15 @@
+"""Distributed inference strategies for sageLLM."""
+
+from __future__ import annotations
+
+from sagellm_core.distributed.strategies import (
+    DistributedStrategy,
+    TensorParallelStrategy,
+    PipelineParallelStrategy,
+)
+
+__all__ = [
+    "DistributedStrategy",
+    "TensorParallelStrategy",
+    "PipelineParallelStrategy",
+]

isagellm_core-0.4.0.0/src/sagellm_core/engine_core/__init__.py

@@ -0,0 +1,12 @@
+"""EngineCore - Coordinates Scheduler and Executor.
+
+The EngineCore is responsible for:
+1. Managing request queues
+2. Coordinating with Scheduler for batch formation
+3. Dispatching batches to Executor
+4. Collecting results
+"""
+
+from sagellm_core.engine_core.engine_core import EngineCore
+
+__all__ = ["EngineCore"]

isagellm_core-0.4.0.0/src/sagellm_core/engine_core/scheduler/__init__.py

@@ -0,0 +1,19 @@
+"""Scheduler - Request scheduling for Continuous Batching.
+
+The Scheduler is responsible for:
+1. Selecting which requests to run in the next step
+2. Managing prefill vs decode scheduling
+3. Preemption decisions
+"""
+
+from sagellm_core.engine_core.scheduler.scheduler import (
+    ContinuousBatchingScheduler,
+    SchedulerConfig,
+    SchedulerOutput,
+)
+
+__all__ = [
+    "ContinuousBatchingScheduler",
+    "SchedulerConfig",
+    "SchedulerOutput",
+]

isagellm_core-0.4.0.0/src/sagellm_core/engines/__init__.py

@@ -0,0 +1,29 @@
+"""Engine implementations for sageLLM Core.
+
+DEPRECATED: This module contains legacy engine implementations.
+Use LLMEngine from sagellm_core instead:
+
+    from sagellm_core import LLMEngine, LLMEngineConfig
+
+    config = LLMEngineConfig(
+        model_path="Qwen/Qwen2-7B",
+        backend_type="cuda",  # or "cpu", "ascend", "auto"
+    )
+    engine = LLMEngine(config)
+    await engine.start()
+    response = await engine.generate("Hello!")
+
+Remaining engines:
+- EmbeddingEngine: Embedding model inference (not yet migrated to LLMEngine)
+"""
+
+from __future__ import annotations
+
+# Only EmbeddingEngine remains - others have been migrated to LLMEngine
+from sagellm_core.engines.embedding import EmbeddingEngine, EmbeddingEngineConfig
+
+__all__ = [
+    # Embedding engine (still needed for embedding-only models)
+    "EmbeddingEngine",
+    "EmbeddingEngineConfig",
+]