isagellm-core 0.3.0.10__py2.py3-none-any.whl → 0.4.0.0__py2.py3-none-any.whl

{isagellm_core-0.3.0.10.dist-info → isagellm_core-0.4.0.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: isagellm-core
-Version: 0.3.0.10
+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.dist-info/RECORD

@@ -0,0 +1,109 @@
+sagellm_core/__init__.py,sha256=xw3aVDgBsbemqEsfM2dLd9bqDBi1edXdKrz2jIS44UE,6615
+sagellm_core/__init__.pyc,sha256=vwQCwAbKpbKiofl2DFQWLVEhY9sO4vAuRo2ne-LrLi0,4414
+sagellm_core/__main__.pyc,sha256=48Ej1ycqV-z87qawGOTNBKNgY3EcbffUOOXrDecOR3g,384
+sagellm_core/config.pyc,sha256=cCMLvMVb_s3HzPYSfqlbRWodQ4Uby8U962l1vmunT5Y,8453
+sagellm_core/demo.pyc,sha256=0hZkAJch6ETjSDsPq8XkDLNcd8qAys_hqOH1qi0xBzg,6028
+sagellm_core/engine.pyc,sha256=o5yzFfLhpRx5lEYC2E75q9b04UqsDY63fC4nZdkHyeU,12258
+sagellm_core/engine_factory.pyc,sha256=cNbEImM0MNcCgVeC0zViAT1AZkPAt2WARrVd1U__tS4,14886
+sagellm_core/engine_server.pyc,sha256=ldiQFB0nYDX6jSGtAduDqdUitBIoMwwnWsblPauQ5qI,19120
+sagellm_core/factory.pyc,sha256=B0s_JSOjVPwE95Oj2moP_vVA3vCrWsi2753UrwF8-eM,2737
+sagellm_core/health.pyc,sha256=c4A60b1ZANbAqck59AV-kczK2b355EPi9JxoYHRlhl8,684
+sagellm_core/llm_engine.pyc,sha256=rc8esb3UDLmwM6LtYDH2olHYWQtqpiuEUshI58XNW-U,22251
+sagellm_core/pd_executor.pyc,sha256=FqwEBT_ooJzJUpFoVr9_aoJmVK1gXVZhhkoZEHABRSQ,17970
+sagellm_core/plugins.pyc,sha256=g6eke_yH65_7Bu6yfT7v19cZ7gjIAWbkyG7gv1JSXy0,2617
+sagellm_core/py.typed,sha256=ixa8YukDZ3kLo0WsFJRGohLMyHzbMur1ALmmASML2cs,64
+sagellm_core/runner.pyc,sha256=AqOu9y2avlD4SNb89YBGV_WgNdC2CTjJ8WFDEtrpdkM,14180
+sagellm_core/runtime.pyc,sha256=RFPMN7Vu9M5O1ZYgFQ9REz4wkMFznRPVPt2ra3u2vns,10018
+sagellm_core/workload.pyc,sha256=G9yMm2L4ZyZpGsCxuEOQ83t78Y_lCQmbg5p6OHl5rn0,4127
+sagellm_core/__pycache__/__init__.cpython-311.pyc,sha256=nBqzd2ITLq8p3n22-0ntfsziD1qImxCYVFUGYIuay1U,4444
+sagellm_core/__pycache__/base_engine.cpython-311.pyc,sha256=XiynMTxG3srYsUwzIoEbwEJOmpjBBB6UNEoAXDxDgWM,9563
+sagellm_core/__pycache__/config.cpython-311.pyc,sha256=6gUdzUs3m31zGQVa2OzlPcPbX1it2Ziag0cQbCaXvIU,8483
+sagellm_core/__pycache__/demo.cpython-311.pyc,sha256=PN6yZ3PgPWjiZ4XqpqRT-s5YYWitMwR_muYFvPmXdWA,6058
+sagellm_core/__pycache__/engine.cpython-311.pyc,sha256=B6ynwwETTD_kjZ2CvUao_QRjM0EluEWg60Be-nRiVeE,12288
+sagellm_core/__pycache__/engine_factory.cpython-311.pyc,sha256=Qte3LXKSpcOWGhu-Hns7xPvTdkyYry3v_EeYQbfVp4w,14916
+sagellm_core/__pycache__/engine_server.cpython-311.pyc,sha256=Hg9T6xwjqulMcyB8MaIdbP9S9ci7CwDmYZ9E9zPS4mc,19150
+sagellm_core/__pycache__/factory.cpython-311.pyc,sha256=R4MbGyreD3U2iWmwQ0r14R347GAY9RM_I1xOiWovozo,2767
+sagellm_core/__pycache__/health.cpython-311.pyc,sha256=Z2mo-4iGINZp0w9AmCJomhHGRSzvbwxCfmdMAjg5vc4,714
+sagellm_core/__pycache__/llm_engine.cpython-311.pyc,sha256=nBLF5BI0rRHRl_PE_1R_XPI-d5tjvczQ3tMi5k2eB0Y,22281
+sagellm_core/__pycache__/mock_engine.cpython-311.pyc,sha256=zCHNncwIzEgqZ0lc7kiWkS-zOHf7CQ73UhCXztR6mFM,581
+sagellm_core/__pycache__/pd_executor.cpython-311.pyc,sha256=zzZ2Cbitv9BsWrqWMqKxE2DbtKkkW3UFacP4WaK2FZA,18000
+sagellm_core/__pycache__/plugins.cpython-311.pyc,sha256=gaoaHApylGSrmMxXVwFs2v4idiZJnisttawlXTbXBWY,2647
+sagellm_core/__pycache__/runner.cpython-311.pyc,sha256=9BLsarR-ouiHIpPpKmJYlo51zBE7PcA5ePvD4PGjbak,14210
+sagellm_core/__pycache__/runtime.cpython-311.pyc,sha256=STMbXptWWqislgsqEk4kIXmUYXdFzGDihSeXGWLy_6k,10048
+sagellm_core/__pycache__/workload.cpython-311.pyc,sha256=Y01UUCDE85D8GY5kF_UsHyNCBbMkmsoB3vOUJV-vulA,4157
+sagellm_core/distributed/__init__.py,sha256=8WTrKkZmYGrV7pj3L083496LskHAv3BOULEIZflgBTc,328
+sagellm_core/distributed/__init__.pyc,sha256=Ur1qc3njwxEPY7PweXoIHTgo1zMblp7HUX84hT0tRAo,498
+sagellm_core/distributed/strategies.pyc,sha256=caK8josx0icKJ0vDwo1UQFeVLzbZf4KB6iUq30ClVnk,6061
+sagellm_core/distributed/__pycache__/__init__.cpython-311.pyc,sha256=4Fv2xNuHYN0Sf98_c9KfU0SbafKhWANL8s4YeTLxiVk,528
+sagellm_core/distributed/__pycache__/strategies.cpython-311.pyc,sha256=MOa7p3j41XvmkJ2tr4OnuHID6gZm6vF6tDWRPe3Uzb8,6091
+sagellm_core/engine_core/__init__.py,sha256=wG6ONHCRWk4ScU6WwW8f-qL3xITF5OcqbxBHGKAAwd4,314
+sagellm_core/engine_core/__init__.pyc,sha256=tnc2J-6ys0KEntuHngGj2vMa_dr6O2NMwkNB2X_BhLQ,502
+sagellm_core/engine_core/engine_core.pyc,sha256=IH5gVbWk3H4Bd2FHqOdb2NVwqhYWcO3qzPKFmLaSfXE,8790
+sagellm_core/engine_core/__pycache__/__init__.cpython-311.pyc,sha256=z_4RxeRts-lRdyGW8z7GVqpcJUNnqK79ep-8GozJGgs,532
+sagellm_core/engine_core/__pycache__/engine_core.cpython-311.pyc,sha256=GnFylOovdrLKBY9wAYymsr_BHMfHpezgEHYTCr9cVI8,8820
+sagellm_core/engine_core/scheduler/__init__.py,sha256=V-XPvuplqlFusYZE7ta2NjLivYiN1ALDLkH7H6BbegU,448
+sagellm_core/engine_core/scheduler/__init__.pyc,sha256=OGnLLe6IHOa0LNy0cO2ojDE8e_vRNsSlEq3za6r5SLU,614
+sagellm_core/engine_core/scheduler/scheduler.pyc,sha256=hrlGzabvyv1mXbB2H17aVIrAPkAdP9GpRekXmiALzzI,8974
+sagellm_core/engine_core/scheduler/__pycache__/__init__.cpython-311.pyc,sha256=no6FHo-7naNZChZLX4Bx6EV5fThrKUXPKjzeT6e7HkM,644
+sagellm_core/engine_core/scheduler/__pycache__/scheduler.cpython-311.pyc,sha256=C_MblVBp8AwnDeE_35J9OS3IMC1HZhZ3-LRp5J6G89o,9004
+sagellm_core/engines/__init__.py,sha256=eRxLCXebcl0DqhMh4J3yPcpmoSMtZzjWjCXNku0hcek,865
+sagellm_core/engines/__init__.pyc,sha256=sOw8q-14ii7rYPyRgXMbyYmDArYQ7fXclqZ86eWhmkE,934
+sagellm_core/engines/embedding.pyc,sha256=bdCOYYasewAb2q5aeAksOEFKGG7VfBjNGt4Ms1_uHcc,12189
+sagellm_core/engines/__pycache__/__init__.cpython-311.pyc,sha256=lDgW-nA9PGC1xxQOYI-2xNMlj5mi5DoB7Re26E334_k,964
+sagellm_core/engines/__pycache__/ascend.cpython-311.pyc,sha256=A3F0gZAhTeUDM7Fpur1skw1DaWBX-W88oXCs9L-0qm0,34653
+sagellm_core/engines/__pycache__/cpu.cpython-311.pyc,sha256=lxuj9LGXjsg7WwB0GXQUsYBphevFENmeLAjKYiswNxY,27863
+sagellm_core/engines/__pycache__/embedding.cpython-311.pyc,sha256=OOCYxcrrEQjzLw7pHCkw_jIYv3t5ihYZ-e9DbEbiRrU,12219
+sagellm_core/engines/__pycache__/hf_cuda.cpython-311.pyc,sha256=WcWhj1ktaGONooxWJZKRv-os5Hw-InlfV2ZfZ6myrLY,61133
+sagellm_core/engines/__pycache__/mock.cpython-311.pyc,sha256=1g9YyKfo9yv-6VTjzzfacv-9ZICHjsVXjyJ_IajRnQo,14641
+sagellm_core/engines/__pycache__/pytorch.cpython-311.pyc,sha256=MRSQN2ZBJmuIxFIA_yflaxarFhslfT0SG0Lw7-EFyDQ,18438
+sagellm_core/engines/__pycache__/pytorch_engine.cpython-311.pyc,sha256=dLBcBODxEqgmH25nCBPWhwy3uu3-dqzyzOwQy-izN7A,11675
+sagellm_core/executor/__init__.py,sha256=Acm1uxVigM823XnTaJ2UC_mSgTIlvtHb6uX076tMSN0,507
+sagellm_core/executor/__init__.pyc,sha256=shUq-xmHO7s6GZGQgUH5XCTJaowF_Pu4KfhcKAbO8KU,705
+sagellm_core/executor/executor_base.pyc,sha256=8ayzIqog994GKozxRQfHHpm5di9HTlcQ-QSAaQit08w,4759
+sagellm_core/executor/uniproc_executor.pyc,sha256=5JQOhTzMFK5onf19qbnzNXV1S_j3xw4p2Vgd8VWmhEY,4940
+sagellm_core/executor/__pycache__/__init__.cpython-311.pyc,sha256=QLugLvDhoO3nbdW-f_3g-HQXKIK437_QkgcLcPdQabA,735
+sagellm_core/executor/__pycache__/executor_base.cpython-311.pyc,sha256=soUj35YdORDrUQG99uxMBlL0MDXwowPAPRR38TtQ7N4,4789
+sagellm_core/executor/__pycache__/uniproc_executor.cpython-311.pyc,sha256=4UEx0RnvX1AoVHvv22OasO0o70KApSNopdSih5l9iDk,4970
+sagellm_core/inputs/__init__.py,sha256=t_DP0n7Cx-L9BiniDcKHBcg6IK6EVCs8J7GRbAwPbLk,301
+sagellm_core/inputs/__init__.pyc,sha256=HcZwXTd9lUm2sOjAz6EP9vLXq9CyQap9GJyRHxeTJMc,513
+sagellm_core/inputs/processor.pyc,sha256=1V5pK5K7_fH6Il93uUoa6317SdiV_6zrZ75U-RFZ_BU,3641
+sagellm_core/inputs/tokenizer_utils.pyc,sha256=6US7QanjlsJhNl5wy9jFQPt3gA8ik2Um3QvFO2_9-gk,2871
+sagellm_core/inputs/__pycache__/__init__.cpython-311.pyc,sha256=1VPyuhLyj5pwRrAcZvoEVIzNiASV38linDfG3RsZw7w,543
+sagellm_core/inputs/__pycache__/processor.cpython-311.pyc,sha256=-6LVrSIfXnca6BK3aji1hcAbk1yuB0jIjp1J9_WQ8y4,3671
+sagellm_core/inputs/__pycache__/tokenizer_utils.cpython-311.pyc,sha256=sOa5f9-SDvPIbLnVgfrYyM4TP7pNuiNbpOJkaFQn7Ds,2901
+sagellm_core/model/__init__.py,sha256=WKrGGLrz5yhK5R1-YapCiV8eBec7kH-NvBOEIm6VD90,329
+sagellm_core/model/__init__.pyc,sha256=yqCreHdJZqQ32Mgg-syjtxuDSZcJXRX5vBYjPkjPopA,542
+sagellm_core/model/model_loader.pyc,sha256=g3Xtz87w22DCdPVbV_eOkd7W_iH7E6Bbc6O23EAYMQI,5713
+sagellm_core/model/weight_utils.pyc,sha256=QHflYnXneKpU4UHGb3RP51_Bx5UiHtYKVsaH_MfvcCg,1830
+sagellm_core/model/__pycache__/__init__.cpython-311.pyc,sha256=o1Raqg_7W-OQzqIDEB_tA-ekmbbIZF9b9TNRM_sRs-k,572
+sagellm_core/model/__pycache__/model_loader.cpython-311.pyc,sha256=fPtDeWbF0fbc0lFxcoCI3RFC84wNjlASypdEniuYMMU,5743
+sagellm_core/model/__pycache__/weight_utils.cpython-311.pyc,sha256=dBd4nMqBX9fyHMFjLoL45-oQ3lQfP1eqeZm47jgz6-M,1860
+sagellm_core/observability/__init__.py,sha256=lT3qt-69vwH3KlxqN0H_hjknmaXl8Yx2Q3PrWifNvd8,378
+sagellm_core/observability/__init__.pyc,sha256=1K3GCd5Gi7770Bx6tLW665_C6KNK3-j447miFkM8GVg,608
+sagellm_core/observability/logger.pyc,sha256=Ng9CYnn9Xr-rJUhethU6HN3oSy5lalmupOtbxk6-g84,1910
+sagellm_core/observability/metrics.pyc,sha256=DZ-bXLTQfnt54BN17PRFtdMZkvIBB4btoEmBzzYmSng,6314
+sagellm_core/observability/__pycache__/__init__.cpython-311.pyc,sha256=xgniiPnO5ImAeCfmvTW6w78ezs8c7aB6vW0XHWn8o7c,638
+sagellm_core/observability/__pycache__/logger.cpython-311.pyc,sha256=7mFsPuy8eHu1nRKi4M5_d4iRSPEPf9fa_jqM8aEDZFY,1940
+sagellm_core/observability/__pycache__/metrics.cpython-311.pyc,sha256=FdoJItSrs_g5_6GTJk7vqjTrpoqc7UthuAuq2m40uw0,6344
+sagellm_core/sampling/__init__.py,sha256=H8b1UnwiYfMP2rODrjDJXvoI8fRTyfCoDImHRstIDGg,330
+sagellm_core/sampling/__init__.pyc,sha256=36Xya9y4YCKQQx8a7EzKJ6htImljpW_dE18TLZMX_OU,556
+sagellm_core/sampling/params.pyc,sha256=LMPOgZL2nkpDAYHC5jG80uMD9vHkBZXM4N3eGFt9PIw,3391
+sagellm_core/sampling/sampler.pyc,sha256=g2tr8FU-71rMvlpWRNNdiY-d6oTAzJ5v-NfRMFpsLYg,6721
+sagellm_core/sampling/__pycache__/__init__.cpython-311.pyc,sha256=aHSszIvPeMHCRRbWg7CSbuXNI5JVBJ3W2tU5Y254wUA,586
+sagellm_core/sampling/__pycache__/params.cpython-311.pyc,sha256=COqJpqpVd1VUFsrZVkblWhTwkBp7A-tmeOh5saIIDoU,3421
+sagellm_core/sampling/__pycache__/sampler.cpython-311.pyc,sha256=-24n6bhHzS3paTzCIrfpbX50XL7mtB4NehbD4jn_ltk,6751
+sagellm_core/worker/__init__.py,sha256=isHAvgRYHX_HBRQsJtB72ziV4xeSDyjR1xFH3h_HEHQ,258
+sagellm_core/worker/__init__.pyc,sha256=mf0kfM381gC3ghpXYQfUOsT9OJGBXy4kdz6Q__qYnIw,445
+sagellm_core/worker/worker.pyc,sha256=yPn1TqZdy7fvPQ_EuImwW5R9-MCnCcCxWwjHeqqf-5k,4772
+sagellm_core/worker/__pycache__/__init__.cpython-311.pyc,sha256=zAe99l1KXH7X1sFabMubGLK0psDOyEv4zRMTF23-fTg,475
+sagellm_core/worker/__pycache__/worker.cpython-311.pyc,sha256=lotaWHeDvpNoyrOCWHKvGjzDdAVBKGwj0h1z4R9frdI,4802
+sagellm_core/worker/model_runner/__init__.py,sha256=w4cq2JoxfHYNUQI3jXO-XhFCoYxNg6YHkioK-S3Ayro,246
+sagellm_core/worker/model_runner/__init__.pyc,sha256=QhC7F-6BIerf9GdwkEPegvF5lQfiepVzhUkMuEstpD8,447
+sagellm_core/worker/model_runner/model_runner.pyc,sha256=wTv8K-F4qtwlQCXpLr1LGIYD-Gae0rzMayp84U5BJWY,16106
+sagellm_core/worker/model_runner/__pycache__/__init__.cpython-311.pyc,sha256=ykIwBnnZYUKPe0foKGwG9rovfzNuSEO6f-4t5DLpMeY,477
+sagellm_core/worker/model_runner/__pycache__/model_runner.cpython-311.pyc,sha256=CLEz8NGJRepPauN_YjT5O4Wrb_FTQ_C8rS1QPbew5TE,16136
+isagellm_core-0.4.0.0.dist-info/METADATA,sha256=T3grOEyLR-G7-hQiCx4U3xGVoVueLAVeAdIC233fxkw,9168
+isagellm_core-0.4.0.0.dist-info/WHEEL,sha256=JNWh1Fm1UdwIQV075glCn4MVuCRs0sotJIq-J6rbxCU,109
+isagellm_core-0.4.0.0.dist-info/entry_points.txt,sha256=w0kgbHe7jVNOuy785wFMEBwqoZUnfU24HCImsQII3y0,64
+isagellm_core-0.4.0.0.dist-info/top_level.txt,sha256=wcgdWrvkaoYYh_dWSFI5Toi8PZsHutVqfhTB2tb0K6g,13
+isagellm_core-0.4.0.0.dist-info/RECORD,,

isagellm_core-0.4.0.0.dist-info/entry_points.txt

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

sagellm_core/__init__.py

@@ -1,19 +1,52 @@
 """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 抽象接口与实现
 - 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__ = "0.3.0.9"
+__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,
@@ -26,15 +59,11 @@ from sagellm_core.config import (
 )
 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 (
-    AscendEngine,
-    AscendEngineConfig,
-    CPUEngine,
     EmbeddingEngine,
     EmbeddingEngineConfig,
-    HFCudaEngine,
-    HFCudaEngineConfig,
-    create_ascend_engine,
 )
 from sagellm_core.factory import create_backend, create_engine
 from sagellm_core.health import HealthStatus
@@ -49,55 +78,80 @@ from sagellm_core.pd_executor import PDExecutionContext, PDSeparatedExecutor
 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
-PyTorchEngineConfig = None
-PyTorchEngineInstanceConfig = None
 create_pytorch_engine = None
 
-try:
-    from sagellm_core.engines.pytorch import (
-        PyTorchEngine,
-        PyTorchEngineConfig,
-        PyTorchEngineInstanceConfig,
-        create_pytorch_engine,
-    )
-except ImportError:
-    pass  # torch or transformers not available
-
-# Auto-register built-in engines (conditional for optional dependencies)
-try:
-    EngineFactory.register(CPUEngine)
-except Exception:
-    pass  # torch not available
-
-try:
-    EngineFactory.register(HFCudaEngine)
-except Exception:
-    pass  # torch or CUDA not available
-
-try:
-    EngineFactory.register(AscendEngine)
-except Exception:
-    pass  # torch_npu or Ascend deps not available
-
-try:
-    EngineFactory.register(EmbeddingEngine)
-except Exception:
-    pass  # sentence-transformers not available
-
-try:
-    if PyTorchEngine is not None:
-        EngineFactory.register(PyTorchEngine)
-except Exception:
-    pass  # torch or transformers not available
-
-__version__ = "0.3.0.9"
+# 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",
@@ -110,18 +164,10 @@ __all__ = [
     "EngineInstanceConfig",  # For runtime engine instantiation
     "HealthStatus",
     # Engine implementations
-    "AscendEngine",
-    "AscendEngineConfig",
-    "create_ascend_engine",
-    "CPUEngine",
-    "HFCudaEngine",
-    "HFCudaEngineConfig",
+    # DEPRECATED: Old engines removed, use LLMEngine instead
+    # Only EmbeddingEngine remains for embedding-only use cases
     "EmbeddingEngine",
     "EmbeddingEngineConfig",
-    "PyTorchEngine",
-    "PyTorchEngineConfig",
-    "PyTorchEngineInstanceConfig",
-    "create_pytorch_engine",
     # Factory functions
     "create_backend",
     "create_engine",

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",
+]

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"]

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",
+]

sagellm_core/engines/__init__.py

@@ -1,45 +1,29 @@
 """Engine implementations for sageLLM Core.
 
-This module provides built-in engine implementations:
-- CPUEngine: CPU-only inference with HuggingFace Transformers
-- HFCudaEngine: CUDA inference with HuggingFace Transformers
-- AscendEngine: Huawei Ascend NPU (PyTorch-NPU)
-- PyTorchEngine: Unified PyTorch engine (auto-detects CUDA/NPU/CPU)
-- EmbeddingEngine: Embedding model inference
+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
 
-from sagellm_core.engines.ascend import AscendEngine, AscendEngineConfig, create_ascend_engine
-from sagellm_core.engines.cpu import CPUEngine, create_cpu_engine
+# Only EmbeddingEngine remains - others have been migrated to LLMEngine
 from sagellm_core.engines.embedding import EmbeddingEngine, EmbeddingEngineConfig
-from sagellm_core.engines.hf_cuda import (
-    HFCudaEngine,
-    HFCudaEngineInstanceConfig,
-    create_hf_cuda_engine,
-)
-from sagellm_core.engines.pytorch_engine import PyTorchEngine, create_pytorch_engine
-
-# Export aliases for backward compatibility
-HFCudaEngineConfig = HFCudaEngineInstanceConfig
 
 __all__ = [
-    # Ascend engine
-    "AscendEngine",
-    "AscendEngineConfig",
-    "create_ascend_engine",
-    # CPU engine
-    "CPUEngine",
-    "create_cpu_engine",
-    # CUDA engine
-    "HFCudaEngine",
-    "HFCudaEngineConfig",
-    "HFCudaEngineInstanceConfig",
-    "create_hf_cuda_engine",
-    # PyTorch engine (unified)
-    "PyTorchEngine",
-    "create_pytorch_engine",
-    # Embedding engine
+    # Embedding engine (still needed for embedding-only models)
     "EmbeddingEngine",
     "EmbeddingEngineConfig",
 ]

sagellm_core/executor/__init__.py

@@ -0,0 +1,16 @@
+"""Executor - Manages Workers for model execution.
+
+Executors dispatch scheduled batches to Workers for execution.
+Different executor types handle different parallelism patterns:
+- UniprocExecutor: Single-process execution
+- MultiprocessExecutor: Multi-process (future)
+- RayExecutor: Ray-based distributed (future)
+"""
+
+from sagellm_core.executor.executor_base import ExecutorBase
+from sagellm_core.executor.uniproc_executor import UniprocExecutor
+
+__all__ = [
+    "ExecutorBase",
+    "UniprocExecutor",
+]

sagellm_core/inputs/__init__.py

@@ -0,0 +1,12 @@
+"""Input processing module for sageLLM."""
+
+from __future__ import annotations
+
+from sagellm_core.inputs.processor import InputProcessor, ProcessedInput
+from sagellm_core.inputs.tokenizer_utils import TokenizerWrapper
+
+__all__ = [
+    "InputProcessor",
+    "ProcessedInput",
+    "TokenizerWrapper",
+]

sagellm_core/model/__init__.py

@@ -0,0 +1,13 @@
+"""Model loading utilities for sageLLM."""
+
+from __future__ import annotations
+
+from sagellm_core.model.model_loader import ModelLoader, load_model
+from sagellm_core.model.weight_utils import WeightLoader, QuantizedWeightLoader
+
+__all__ = [
+    "ModelLoader",
+    "load_model",
+    "WeightLoader",
+    "QuantizedWeightLoader",
+]

sagellm_core/observability/__init__.py

@@ -0,0 +1,16 @@
+"""Observability module for sageLLM.
+
+Provides structured logging, metrics, and tracing.
+"""
+
+from __future__ import annotations
+
+from sagellm_core.observability.metrics import MetricsCollector, EngineMetrics
+from sagellm_core.observability.logger import setup_logger, get_logger
+
+__all__ = [
+    "MetricsCollector",
+    "EngineMetrics",
+    "setup_logger",
+    "get_logger",
+]

sagellm_core/sampling/__init__.py

@@ -0,0 +1,14 @@
+"""Sampling module for sageLLM."""
+
+from __future__ import annotations
+
+from sagellm_core.sampling.params import SamplingParams
+from sagellm_core.sampling.sampler import Sampler, GreedySampler, TopKSampler, TopPSampler
+
+__all__ = [
+    "SamplingParams",
+    "Sampler",
+    "GreedySampler",
+    "TopKSampler",
+    "TopPSampler",
+]

sagellm_core/worker/__init__.py

@@ -0,0 +1,11 @@
+"""Worker - Executes model forward passes.
+
+Workers are responsible for:
+1. Loading model weights
+2. Running ModelRunner for forward passes
+3. Managing GPU memory (via BackendProvider)
+"""
+
+from sagellm_core.worker.worker import Worker
+
+__all__ = ["Worker"]

sagellm_core/worker/model_runner/__init__.py

@@ -0,0 +1,8 @@
+"""ModelRunner - Runs model forward passes.
+
+The ModelRunner contains the actual model execution logic.
+"""
+
+from sagellm_core.worker.model_runner.model_runner import ModelRunner, ModelRunnerOutput
+
+__all__ = ["ModelRunner", "ModelRunnerOutput"]

isagellm_core-0.3.0.10.dist-info/RECORD

@@ -1,52 +0,0 @@
-sagellm_core/__init__.py,sha256=SD78wlVmkv9ZhzcdebPEy2Qr6uWA4FGLf1UJPQOnICE,3691
-sagellm_core/__init__.pyc,sha256=E0UCvZzZHJrdpIFw0xLRu873Vqe3XoaN57Nwu-sdroY,3525
-sagellm_core/__main__.pyc,sha256=48Ej1ycqV-z87qawGOTNBKNgY3EcbffUOOXrDecOR3g,384
-sagellm_core/config.pyc,sha256=cCMLvMVb_s3HzPYSfqlbRWodQ4Uby8U962l1vmunT5Y,8453
-sagellm_core/demo.pyc,sha256=0hZkAJch6ETjSDsPq8XkDLNcd8qAys_hqOH1qi0xBzg,6028
-sagellm_core/engine.pyc,sha256=o5yzFfLhpRx5lEYC2E75q9b04UqsDY63fC4nZdkHyeU,12258
-sagellm_core/engine_factory.pyc,sha256=cNbEImM0MNcCgVeC0zViAT1AZkPAt2WARrVd1U__tS4,14886
-sagellm_core/engine_server.pyc,sha256=oKPJNu_s5v_wBvhkoMfPfCU-bD_pVA9syLcHLfN-rs4,19389
-sagellm_core/factory.pyc,sha256=B0s_JSOjVPwE95Oj2moP_vVA3vCrWsi2753UrwF8-eM,2737
-sagellm_core/health.pyc,sha256=c4A60b1ZANbAqck59AV-kczK2b355EPi9JxoYHRlhl8,684
-sagellm_core/pd_executor.pyc,sha256=1ytnDJi56sTSnZjAa7cRlN_1yQU4t7OJwXYgeUgjqA4,17784
-sagellm_core/plugins.pyc,sha256=g6eke_yH65_7Bu6yfT7v19cZ7gjIAWbkyG7gv1JSXy0,2617
-sagellm_core/py.typed,sha256=ixa8YukDZ3kLo0WsFJRGohLMyHzbMur1ALmmASML2cs,64
-sagellm_core/runner.pyc,sha256=GK_t15IH2etN6flRqUui4iUPkD39zl3_rw1wnOSG5Jg,14238
-sagellm_core/runtime.pyc,sha256=RFPMN7Vu9M5O1ZYgFQ9REz4wkMFznRPVPt2ra3u2vns,10018
-sagellm_core/workload.pyc,sha256=G9yMm2L4ZyZpGsCxuEOQ83t78Y_lCQmbg5p6OHl5rn0,4127
-sagellm_core/__pycache__/__init__.cpython-311.pyc,sha256=Ic7SZye-CxAZ_1ftXwr1XnyqmJWXcy9CCoJkRNg9ZGQ,3555
-sagellm_core/__pycache__/base_engine.cpython-311.pyc,sha256=XiynMTxG3srYsUwzIoEbwEJOmpjBBB6UNEoAXDxDgWM,9563
-sagellm_core/__pycache__/config.cpython-311.pyc,sha256=6gUdzUs3m31zGQVa2OzlPcPbX1it2Ziag0cQbCaXvIU,8483
-sagellm_core/__pycache__/demo.cpython-311.pyc,sha256=PN6yZ3PgPWjiZ4XqpqRT-s5YYWitMwR_muYFvPmXdWA,6058
-sagellm_core/__pycache__/engine.cpython-311.pyc,sha256=B6ynwwETTD_kjZ2CvUao_QRjM0EluEWg60Be-nRiVeE,12288
-sagellm_core/__pycache__/engine_factory.cpython-311.pyc,sha256=Qte3LXKSpcOWGhu-Hns7xPvTdkyYry3v_EeYQbfVp4w,14916
-sagellm_core/__pycache__/engine_server.cpython-311.pyc,sha256=m5YWclUbXt_IHuJLesJwEtFQF6zgTVrykD5T_9YznMk,19419
-sagellm_core/__pycache__/factory.cpython-311.pyc,sha256=R4MbGyreD3U2iWmwQ0r14R347GAY9RM_I1xOiWovozo,2767
-sagellm_core/__pycache__/health.cpython-311.pyc,sha256=Z2mo-4iGINZp0w9AmCJomhHGRSzvbwxCfmdMAjg5vc4,714
-sagellm_core/__pycache__/mock_engine.cpython-311.pyc,sha256=zCHNncwIzEgqZ0lc7kiWkS-zOHf7CQ73UhCXztR6mFM,581
-sagellm_core/__pycache__/pd_executor.cpython-311.pyc,sha256=XuMKNOZEYiupvhyQjgN1iM3p2Q0daXQvTNK745BPL5U,17814
-sagellm_core/__pycache__/plugins.cpython-311.pyc,sha256=gaoaHApylGSrmMxXVwFs2v4idiZJnisttawlXTbXBWY,2647
-sagellm_core/__pycache__/runner.cpython-311.pyc,sha256=krW5MagsiQcX5J9fljblEshfsQB_gAIAWR7S0uD-TzU,14268
-sagellm_core/__pycache__/runtime.cpython-311.pyc,sha256=STMbXptWWqislgsqEk4kIXmUYXdFzGDihSeXGWLy_6k,10048
-sagellm_core/__pycache__/workload.cpython-311.pyc,sha256=Y01UUCDE85D8GY5kF_UsHyNCBbMkmsoB3vOUJV-vulA,4157
-sagellm_core/engines/__init__.py,sha256=dK5-xPVkGftvSUDg7B6rNuaMohibckfXrklaWFzz_Jo,1418
-sagellm_core/engines/__init__.pyc,sha256=k_8HvoXGNmRVikV24Dxy0QIoLJ_XLP2aKwu7Dw3CIkQ,1438
-sagellm_core/engines/ascend.pyc,sha256=T2qveWOhxV1AyVwJ5n_q8-TLg0uBJN_G6EZuVpA5rCU,17263
-sagellm_core/engines/cpu.pyc,sha256=CLLNyuRLDIER5Zgny91-eGYUSMxmVVAw4E0rUMJDDsU,27833
-sagellm_core/engines/embedding.pyc,sha256=Be4To-qpamFuGcyFZ1MKPk6XOf_HDBKnBGoIisxNemE,11476
-sagellm_core/engines/hf_cuda.pyc,sha256=zo-Ht70Q3aba7L5xugpc6fwTIRrpkEFe9SeTy3exI-s,61103
-sagellm_core/engines/pytorch.pyc,sha256=MLcymoDnwMduWoWVDZmYuG6UoQavNO-FjddPQ8ccccI,18408
-sagellm_core/engines/pytorch_engine.pyc,sha256=fdzf-VN2hVRWzP8PCmGV-pg9qUIJWWqEH662CRmh5_0,11645
-sagellm_core/engines/__pycache__/__init__.cpython-311.pyc,sha256=dI0jrBb6-rlnTvVQZojWA2n-lUhE9j0aGn4cz3SxmqY,1468
-sagellm_core/engines/__pycache__/ascend.cpython-311.pyc,sha256=UvdVYqCptZqrPhyrbqK2dxnVa9NY-HH_vsXffsblzC8,17293
-sagellm_core/engines/__pycache__/cpu.cpython-311.pyc,sha256=lxuj9LGXjsg7WwB0GXQUsYBphevFENmeLAjKYiswNxY,27863
-sagellm_core/engines/__pycache__/embedding.cpython-311.pyc,sha256=ZNqFpuCsS4Wp0xg7yV3GBbaJ4G0saqGNaKMdivxEBkA,11506
-sagellm_core/engines/__pycache__/hf_cuda.cpython-311.pyc,sha256=WcWhj1ktaGONooxWJZKRv-os5Hw-InlfV2ZfZ6myrLY,61133
-sagellm_core/engines/__pycache__/mock.cpython-311.pyc,sha256=1g9YyKfo9yv-6VTjzzfacv-9ZICHjsVXjyJ_IajRnQo,14641
-sagellm_core/engines/__pycache__/pytorch.cpython-311.pyc,sha256=MRSQN2ZBJmuIxFIA_yflaxarFhslfT0SG0Lw7-EFyDQ,18438
-sagellm_core/engines/__pycache__/pytorch_engine.cpython-311.pyc,sha256=dLBcBODxEqgmH25nCBPWhwy3uu3-dqzyzOwQy-izN7A,11675
-isagellm_core-0.3.0.10.dist-info/METADATA,sha256=xTzEVbHCbeZr5S3gTu6lTvDy9QR_1I34DiU2iWvS2-0,9254
-isagellm_core-0.3.0.10.dist-info/WHEEL,sha256=JNWh1Fm1UdwIQV075glCn4MVuCRs0sotJIq-J6rbxCU,109
-isagellm_core-0.3.0.10.dist-info/entry_points.txt,sha256=QWQd0kFD5erCgpS6DqfLpgkbD03vLf1ouojaUXHGX8w,312
-isagellm_core-0.3.0.10.dist-info/top_level.txt,sha256=wcgdWrvkaoYYh_dWSFI5Toi8PZsHutVqfhTB2tb0K6g,13
-isagellm_core-0.3.0.10.dist-info/RECORD,,

isagellm_core-0.3.0.10.dist-info/entry_points.txt

@@ -1,8 +0,0 @@
-[console_scripts]
-sage-engine = sagellm_core.engine_server:main
-
-[sagellm.engines]
-ascend = sagellm_core.engines.ascend:create_ascend_engine
-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