checkpoint-engine 0.3.3__py3-none-any.whl → 0.3.4__py3-none-any.whl

checkpoint_engine/_version.py

@@ -28,7 +28,7 @@ version_tuple: VERSION_TUPLE
 commit_id: COMMIT_ID
 __commit_id__: COMMIT_ID
 
-__version__ = version = '0.3.3'
-__version_tuple__ = version_tuple = (0, 3, 3)
+__version__ = version = '0.3.4'
+__version_tuple__ = version_tuple = (0, 3, 4)
 
 __commit_id__ = commit_id = None

checkpoint_engine/worker.py

@@ -10,6 +10,9 @@ import zmq
 from checkpoint_engine.device_utils import DeviceManager, npu_generate_uuid
 
 
+_WEIGHTS_TYPE = list[tuple[str, torch.Tensor]]
+
+
 def _rebuild_ipc(handle: tuple[Callable, tuple], device_id: int | None = None) -> torch.Tensor:
     func, args = handle
     list_args = list(args)
@@ -29,11 +32,9 @@ class FlattenedTensorMetadata(TypedDict):
     offset: int
 
 
-def _extract_weights(
-    payload: list[FlattenedTensorMetadata], buffer: torch.Tensor
-) -> list[tuple[str, torch.Tensor]]:
+def _extract_weights(payload: list[FlattenedTensorMetadata], buffer: torch.Tensor) -> _WEIGHTS_TYPE:
     assert buffer is not None
-    weights: list[tuple[str, torch.Tensor]] = []
+    weights: _WEIGHTS_TYPE = []
     for item in payload:
         shape = item["shape"]
         if isinstance(shape, list | tuple):
@@ -166,12 +167,31 @@ class VllmColocateWorkerExtension:
             self.device = torch.device(f"npu:{self.local_rank}")
         assert self.device is not None
 
+        def _load_weights(weights: _WEIGHTS_TYPE):
+            # Load main model weights
+            self.model_runner.model.load_weights(weights)
+            # Load drafter model weights if MTP/speculative decoding is enabled
+            if (
+                getattr(self.model_runner, "drafter", None) is not None
+                and getattr(self.model_runner.drafter, "model", None) is not None
+            ):
+                self.model_runner.drafter.model.load_weights(weights=weights)
+
+        def _post_hook():
+            process_weights_after_loading(self.model_runner.model, self.model_config, self.device)
+            # Also trigger drafter model's post processing if MTP is enabled
+            if (
+                getattr(self.model_runner, "drafter", None) is not None
+                and getattr(self.model_runner.drafter, "model", None) is not None
+            ):
+                process_weights_after_loading(
+                    self.model_runner.drafter.model, self.model_config, self.device
+                )
+
         update_weights_from_ipc(
             self._zmq_ctx,
             zmq_handles[self._device_uuid],
             device_id=self.device.index,
-            run=self.model_runner.model.load_weights,
-            post_hook=lambda: process_weights_after_loading(
-                self.model_runner.model, self.model_config, self.device
-            ),
+            run=_load_weights,
+            post_hook=_post_hook,
         )

{checkpoint_engine-0.3.3.dist-info → checkpoint_engine-0.3.4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: checkpoint-engine
-Version: 0.3.3
+Version: 0.3.4
 Summary: checkpoint-engine is a lightweight, decoupling and efficient weight update middleware
 Project-URL: Homepage, https://github.com/MoonshotAI/checkpoint-engine
 Project-URL: Repository, https://github.com/MoonshotAI/checkpoint-engine

{checkpoint_engine-0.3.3.dist-info → checkpoint_engine-0.3.4.dist-info}/RECORD

@@ -1,15 +1,15 @@
 checkpoint_engine/__init__.py,sha256=OeWxe9mxl2sZ6cW-blSTg6JbFlOMpGbBghLZtxGOqXk,942
 checkpoint_engine/__main__.py,sha256=yzQlApuYo6eIOqtqM018RosyxNzXzB5a-stxUvsh-dg,709
-checkpoint_engine/_version.py,sha256=lemL_4Kl75FgrO6lVuFrrtw6-Dcf9wtXBalKkXuzkO4,704
+checkpoint_engine/_version.py,sha256=3nDaC5e0d_scBB1bUEKPlItbvbY0PmXNNyyOTNFNWNI,704
 checkpoint_engine/api.py,sha256=JDiQ4i3Gb6GoaBhlp8lNuUPaVURoFFdeGJY9ZDDGvPc,3518
 checkpoint_engine/data_types.py,sha256=O9uAXjwB20iwrOHfEEQd8Y9CmaFspNJ9ks9noHqwQKk,2716
 checkpoint_engine/device_utils.py,sha256=iKrof60j3CY3fStRTq3DRTt_kE1vYoEWHhAeyh0lByA,3020
 checkpoint_engine/p2p_store.py,sha256=abiCDVmRISPt9QFfavHB9Jo7ZpBbSjUS1NevGuB-AVA,8721
 checkpoint_engine/pin_memory.py,sha256=b7nABKJV2bSIsOfX2YTHzUk1OkOze6AQjCaOIFaQnbA,16708
 checkpoint_engine/ps.py,sha256=wBsHu2qWy5oRBrvLc7aEOroG_j58UJoWT6lFH4ylMRk,41092
-checkpoint_engine/worker.py,sha256=ghj9d2u8hY_U2uiOZWIN2CqRNZH6PrzujT22fHUFBWI,6879
-checkpoint_engine-0.3.3.dist-info/licenses/LICENCE,sha256=D3gPmHKpGtF1yxYNhqjtBtZY_brZjDotJTzpnmClzlY,1067
-checkpoint_engine-0.3.3.dist-info/METADATA,sha256=WyyGLw1qrteQgRGOWhAm15NN2nzklTqw4iiQ9U2nYpQ,11559
-checkpoint_engine-0.3.3.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-checkpoint_engine-0.3.3.dist-info/top_level.txt,sha256=66sik_1eLakLYmcllOEJzFaNbSfjsueuP0tHYEzhMSs,18
-checkpoint_engine-0.3.3.dist-info/RECORD,,
+checkpoint_engine/worker.py,sha256=CDWbxwvMpid19yriuwAsyZLUZtqfkh9Lybn8KpiuKCw,7781
+checkpoint_engine-0.3.4.dist-info/licenses/LICENCE,sha256=D3gPmHKpGtF1yxYNhqjtBtZY_brZjDotJTzpnmClzlY,1067
+checkpoint_engine-0.3.4.dist-info/METADATA,sha256=P23Txz8z5WvM3km3EHFtKBEc5299c5UZcd0UTABN-u8,11559
+checkpoint_engine-0.3.4.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+checkpoint_engine-0.3.4.dist-info/top_level.txt,sha256=66sik_1eLakLYmcllOEJzFaNbSfjsueuP0tHYEzhMSs,18
+checkpoint_engine-0.3.4.dist-info/RECORD,,

{checkpoint_engine-0.3.3.dist-info → checkpoint_engine-0.3.4.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (80.9.0)
+Generator: setuptools (80.10.2)
 Root-Is-Purelib: true
 Tag: py3-none-any