isagellm-core 0.4.0.17__py2.py3-none-any.whl → 0.4.0.19__py2.py3-none-any.whl

sagellm_core/model/__init__.py

@@ -1,13 +1,139 @@
-"""Model loading utilities for sageLLM."""
+"""Model definitions using backend kernels.
 
-from __future__ import annotations
+This module provides model architectures that use sagellm-core layers
+instead of direct HuggingFace implementations.
 
+Key models:
+- BaseLLMModel: Abstract base for all LLM models
+- GPT2Model: GPT-2 architecture (for testing)
+- LlamaModel: LLaMA/LLaMA-2/LLaMA-3 architecture
+- Qwen2Model: Qwen2/Qwen2.5 architecture
+- MixtralModel: Mixtral Mixture of Experts architecture
+- create_model_from_config: Factory function to create models
+
+Registry system:
+- MODEL_REGISTRY: Central registry of model types
+- register_model: Decorator to register new models
+- get_model_class: Look up model class by type
+
+Weight loading:
+- WeightLoaderProtocol: Protocol for weight loaders
+- WEIGHT_LOADERS: Registry of weight formats
+- SafetensorsLoader, PytorchLoader: Built-in loaders
+- QuantizedWeightLoader: Quantized weight loading (INT8/AWQ/GPTQ)
+
+Quantization:
+- QuantConfig: Quantization configuration
+- detect_quantization: Auto-detect quantization format
+"""
+
+# Registry (import first to ensure registration works)
+from sagellm_core.model.registry import (
+    MODEL_REGISTRY,
+    get_model_class,
+    is_model_registered,
+    list_registered_models,
+    register_model,
+)
+
+# Base class
+from sagellm_core.model.base import BaseLLMModel
+
+# Factory
+from sagellm_core.model.factory import create_model_from_config, load_model_weights
+
+# Models (import triggers registration)
+from sagellm_core.model.gpt2 import GPT2Model
+from sagellm_core.model.llama import (
+    LlamaModel,
+    LlamaAttention,
+    LlamaMLP,
+    LlamaDecoderLayer,
+    LlamaRotaryEmbedding,
+)
+from sagellm_core.model.qwen2 import (
+    Qwen2Model,
+    Qwen2Attention,
+    Qwen2MLP,
+    Qwen2DecoderLayer,
+)
+from sagellm_core.model.mixtral import (
+    MixtralModel,
+    MixtralSparseMoE,
+    MixtralMLP,
+    MixtralDecoderLayer,
+)
+
+# Quantization
+from sagellm_core.model.quantization import (
+    QuantConfig,
+    QuantizationType,
+    detect_quantization,
+)
+
+# Weight loaders
+from sagellm_core.model.weight_loader import (
+    WEIGHT_LOADERS,
+    WeightLoaderProtocol,
+    WeightNameMapper,
+    SafetensorsLoader,
+    PytorchLoader,
+    detect_weight_format,
+    get_weight_loader,
+    list_weight_loaders,
+    load_weights_into_model,
+    register_weight_loader,
+)
+from sagellm_core.model.weight_loader.quantized import QuantizedWeightLoader
+
+# Legacy compatibility
 from sagellm_core.model.model_loader import ModelLoader, load_model
-from sagellm_core.model.weight_utils import WeightLoader, QuantizedWeightLoader
 
 __all__ = [
+    # Base
+    "BaseLLMModel",
+    # Models
+    "GPT2Model",
+    "LlamaModel",
+    "LlamaAttention",
+    "LlamaMLP",
+    "LlamaDecoderLayer",
+    "LlamaRotaryEmbedding",
+    "Qwen2Model",
+    "Qwen2Attention",
+    "Qwen2MLP",
+    "Qwen2DecoderLayer",
+    "MixtralModel",
+    "MixtralSparseMoE",
+    "MixtralMLP",
+    "MixtralDecoderLayer",
+    # Factory
+    "create_model_from_config",
+    "load_model_weights",
+    # Registry
+    "MODEL_REGISTRY",
+    "register_model",
+    "get_model_class",
+    "list_registered_models",
+    "is_model_registered",
+    # Quantization
+    "QuantConfig",
+    "QuantizationType",
+    "detect_quantization",
+    # Weight loaders
+    "WEIGHT_LOADERS",
+    "WeightLoaderProtocol",
+    "WeightNameMapper",
+    "SafetensorsLoader",
+    "PytorchLoader",
+    "QuantizedWeightLoader",
+    "detect_weight_format",
+    "get_weight_loader",
+    "list_weight_loaders",
+    "load_weights_into_model",
+    "register_weight_loader",
+    # Legacy
     "ModelLoader",
     "load_model",
-    "WeightLoader",
-    "QuantizedWeightLoader",
 ]
+

sagellm_core/model/weight_loader/__init__.py

@@ -0,0 +1,54 @@
+"""Weight loader module.
+
+This module provides utilities for loading model weights from various formats:
+- safetensors: Fast, secure tensor format (recommended)
+- pytorch: Standard PyTorch .bin/.pt checkpoints
+
+Usage:
+    from sagellm_core.model.weight_loader import (
+        get_weight_loader,
+        detect_weight_format,
+        load_weights_into_model,
+    )
+
+    # Auto-detect format and load
+    load_weights_into_model(model, "path/to/model")
+
+    # Or manually:
+    loader = get_weight_loader("safetensors")
+    for name, tensor in loader.load_weights("path/to/model"):
+        print(f"{name}: {tensor.shape}")
+"""
+
+from sagellm_core.model.weight_loader.base import (
+    WEIGHT_LOADERS,
+    WeightLoaderProtocol,
+    WeightNameMapper,
+    detect_weight_format,
+    get_weight_loader,
+    is_weight_loader_registered,
+    list_weight_loaders,
+    load_weights_into_model,
+    register_weight_loader,
+)
+
+# Import loaders to trigger registration
+from sagellm_core.model.weight_loader.safetensors import SafetensorsLoader
+from sagellm_core.model.weight_loader.pytorch import PytorchLoader
+
+__all__ = [
+    # Protocol and registry
+    "WeightLoaderProtocol",
+    "WEIGHT_LOADERS",
+    "register_weight_loader",
+    "get_weight_loader",
+    "list_weight_loaders",
+    "is_weight_loader_registered",
+    # Utilities
+    "detect_weight_format",
+    "WeightNameMapper",
+    "load_weights_into_model",
+    # Concrete loaders
+    "SafetensorsLoader",
+    "PytorchLoader",
+]

isagellm_core-0.4.0.17.dist-info/METADATA

@@ -1,308 +0,0 @@
-Metadata-Version: 2.4
-Name: isagellm-core
-Version: 0.4.0.17
-Summary: sageLLM core runtime with PD separation (MVP)
-Author: IntelliStream Team
-License: Proprietary - IntelliStream
-Classifier: Development Status :: 3 - Alpha
-Classifier: Programming Language :: Python :: 3
-Classifier: Programming Language :: Python :: 3.10
-Classifier: Programming Language :: Python :: 3.11
-Classifier: Programming Language :: Python :: 3.12
-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<0.5.0,>=0.4.0.0
-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,>=0.4.0.0
-Requires-Dist: fastapi>=0.100.0
-Requires-Dist: uvicorn>=0.22.0
-Requires-Dist: torch>=2.0.0
-Requires-Dist: transformers>=4.35.0
-Requires-Dist: accelerate>=0.26.0
-Provides-Extra: dev
-Requires-Dist: pytest>=7.0.0; extra == "dev"
-Requires-Dist: pytest-cov>=4.0.0; extra == "dev"
-Requires-Dist: pytest-timeout>=2.0.0; extra == "dev"
-Requires-Dist: ruff>=0.8.0; extra == "dev"
-Requires-Dist: mypy>=1.0.0; extra == "dev"
-Requires-Dist: types-PyYAML>=6.0.0; extra == "dev"
-Requires-Dist: pre-commit>=3.0.0; extra == "dev"
-Requires-Dist: isage-pypi-publisher>=0.2.0; extra == "dev"
-
-# sagellm-core
-
-## Protocol Compliance (Mandatory)
-
-- MUST follow Protocol v0.1: https://github.com/intellistream/sagellm-docs/blob/main/docs/specs/protocol_v0.1.md
-- Any globally shared definitions (fields, error codes, metrics, IDs, schemas) MUST be added to Protocol first.
-
-[![CI](https://github.com/intellistream/sagellm-core/actions/workflows/ci.yml/badge.svg)](https://github.com/intellistream/sagellm-core/actions/workflows/ci.yml)
-[![PyPI](https://img.shields.io/pypi/v/isagellm-core.svg)](https://pypi.org/project/isagellm-core/)
-[![Python](https://img.shields.io/pypi/pyversions/isagellm-core.svg)](https://pypi.org/project/isagellm-core/)
-[![License](https://img.shields.io/github/license/intellistream/sagellm-core.svg)](https://github.com/intellistream/sagellm-core/blob/main/LICENSE)
-[![Code style: ruff](https://img.shields.io/badge/code%20style-ruff-000000.svg)](https://github.com/astral-sh/ruff)
-
-sageLLM Core - 引擎协调层与运行时系统
-
-## 架构定位
-
-```
-┌─────────────────────────────────────────────────────────────┐
-│  sagellm-core (引擎协调层) ← 本仓库                         │
-│  ┌─────────────────────────────────────────────────────┐    │
-│  │  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)                  │    │
-│  │  • YAML/JSON 配置解析                               │    │
-│  │  • Pydantic v2 类型验证                            │    │
-│  └─────────────────────────────────────────────────────┘    │
-├─────────────────────────────────────────────────────────────┤
-│  sagellm-backend (硬件抽象层)                               │
-│  • BackendProvider (CPU/CUDA/Ascend)                       │
-│  • Stream/Event/KVBlock 管理                              │
-└─────────────────────────────────────────────────────────────┘
-```
-
-**职责分离**：
-- ✅ **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
-
-- 🔧 引擎抽象层与自描述架构
-- 🏭 EngineFactory - 支持自动发现与优先级选择
-- 🎯 内置引擎实现（CPU/CUDA/Embedding）
-- 🔌 插件系统 - 扩展引擎与后端
-- ⚙️ 配置系统 - YAML/JSON + Pydantic v2
-- ✅ CPU-First - 无 GPU 测试支持
-
-## MoE（规划中）
-
-- MoE 路由/调度/执行图由 **sagellm-core** 主责
-- 依赖 **sagellm-comm**（all-to-all/拓扑通信）与 **sagellm-backend**（专家算子/内核）
-- 需先在 Protocol 中补充全局字段（如 router 元数据、专家负载指标）
-
-## 安装
-
-```bash
-# 从 PyPI 安装（自动安装 protocol + backend 依赖）
-pip install isagellm-core
-```
-
-## 🚀 开发者快速开始
-
-```bash
-git clone git@github.com:intellistream/sagellm-core.git
-cd sagellm-core
-./quickstart.sh   # 一键安装开发环境（含依赖）
-
-# 或手动安装
-pip install -e ".[dev]"
-```
-
-运行测试：
-```bash
-pytest tests/ -v
-```
-
-> 💡 **提示**：`isagellm-protocol` 和 `isagellm-backend` 会自动从 PyPI 安装。
-> 如需本地联调：
-> ```bash
-> pip install -e ../sagellm-protocol
-> pip install -e ../sagellm-backend
-> ```
-
-## Configuration System
-
-### 使用方法
-
-```python
-from sagellm_core import load_config, create_backend, create_engine
-
-# 从 YAML/JSON 加载配置
-config = load_config("config.yaml")
-
-# 创建 backend 和 engine（通过插件发现）
-backend = create_backend(config.backend)
-engine = create_engine(config.engine, backend)
-```
-
-### Configuration Structure
-
-Main configuration components:
-- `BackendConfig`: Device and backend settings
-- `EngineConfig`: Inference engine configuration
-- `WorkloadConfig`: Workload parameters
-- `OutputConfig`: Output paths and logging
-
-### 配置示例
-
-#### 示例配置文件
-
-- [config_cpu.yaml](examples/config_cpu.yaml) - CPU 模式（CI/开发）
-- [config_cuda.yaml](examples/config_cuda.yaml) - CUDA 生产模式
-- [config_ascend.yaml](examples/config_ascend.yaml) - 昇腾生产模式
-- [config_minimal.json](examples/config_minimal.json) - 最小 JSON 配置
-
-更多信息参见 [examples/README.md](examples/README.md)
-
-### 配置格式
-
-支持 YAML（推荐）和 JSON 格式：
-
-```yaml
-backend:
-  kind: cpu
-engine:
-  kind: cpu
-  model: sshleifer/tiny-gpt2
-workload:
-  segments: [short, long, stress]
-  concurrency: 4
-output:
-  metrics_path: ./metrics.json
-```
-
-### 插件解析
-
-当配置指定的 backend/engine kind 未安装时，会抛出 `PluginResolutionError`：
-
-```python
-from sagellm_core import create_backend, BackendConfig, PluginResolutionError
-
-try:
-    backend = create_backend(BackendConfig(kind="ascend_cann", device="npu:0"))
-except PluginResolutionError as e:
-    print(f"错误: {e}")
-    # 输出: No implementation found for sagellm.backends kind='ascend_cann'.
-    #       Install hint: pip install isagellm-backend-ascend_cann
-```
-
-## Development Guide
-
-### 快速开始
-
-```bash
-# 克隆仓库
-git clone https://github.com/intellistream/sagellm-core.git
-cd sagellm-core
-
-# 安装开发依赖
-pip install -e ".[dev]"
-
-# 安装 pre-commit hooks
-pre-commit install
-
-# 验证环境
-pytest tests/ -v
-```
-
-### Testing and Quality
-
-#### Pre-commit Hooks
-
-安装后，每次 `git commit` 会自动运行：
-- **Ruff**: 代码格式化 + Lint 检查
-- **Mypy**: 静态类型检查
-- **YAML/JSON**: 配置文件验证
-
-```bash
-# 手动运行所有 hooks
-pre-commit run --all-files
-
-# 绕过 hooks（紧急情况）
-git commit --no-verify
-```
-
-#### Running Tests
-
-```bash
-# Run all tests
-pytest tests/ -v
-
-# Run specific test module
-pytest tests/unit/test_demo.py -v
-
-# Generate coverage report
-pytest tests/ --cov=sagellm_core --cov-report=html
-```
-
-#### Continuous Integration
-
-GitHub Actions automatically runs on each PR:
-- Code linting and formatting checks
-- Tests across Python 3.10, 3.11, 3.12
-- Package build verification
-
-### Code Style
-
-This project uses:
-- **Ruff** for formatting and linting
-- **Mypy** for type checking
-- **Type hints** are required for all functions
-
-For detailed guidelines, see [CONTRIBUTING.md](CONTRIBUTING.md)
-
-### 代码检查
-
-```bash
-# 格式化代码
-ruff format .
-
-# Lint 检查
-ruff check .
-
-# 类型检查
-mypy src/sagellm_core
-
-# 一键检查所有
-pre-commit run --all-files
-```
-
-## 依赖
-
-- `pydantic>=2.0.0`: 配置校验
-- `pyyaml>=6.0.0`: YAML 配置支持
-- `isagellm-protocol>=0.1.0,<0.2.0`: 协议定义（Level 0）
-- `isagellm-backend>=0.1.0,<0.2.0`: 后端抽象（Level 1）
-
-## Related Packages
-
-- `isagellm-protocol` - Protocol definitions
-- `isagellm-backend` - Backend abstraction layer
-- `isagellm` - Main package with CLI
-
-For more packages, see the [sageLLM ecosystem](https://github.com/intellistream/sagellm)
-
-## 🔄 贡献指南
-
-请遵循以下工作流程：
-
-1. **创建 Issue** - 描述问题/需求
-   ```bash
-   gh issue create --title "[Bug] 描述" --label "bug,sagellm-core"
-   ```
-
-2. **开发修复** - 在本地 `fix/#123-xxx` 分支解决
-   ```bash
-   git checkout -b fix/#123-xxx origin/main-dev
-   # 开发、测试...
-   pytest -v
-   ruff format . && ruff check . --fix
-   ```
-
-3. **发起 PR** - 提交到 `main-dev` 分支
-   ```bash
-   gh pr create --base main-dev --title "Fix: 描述" --body "Closes #123"
-   ```
-
-4. **合并** - 审批后合并到 `main-dev`
-
-更多详情见 [.github/copilot-instructions.md](.github/copilot-instructions.md)