horde-engine 3.3.2__tar.gz → 3.3.3__tar.gz

{horde_engine-3.3.2 → horde_engine-3.3.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: horde-engine
-Version: 3.3.2
+Version: 3.3.3
 Summary: A wrapper around ComfyUI to allow use by AI Horde.
 Project-URL: Homepage, https://github.com/Haidra-Org/hordelib
 Project-URL: Bug Tracker, https://github.com/Haidra-Org/hordelib/issues

{horde_engine-3.3.2 → horde_engine-3.3.3}/hordelib/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '3.3.2'
-__version_tuple__ = version_tuple = (3, 3, 2)
+__version__ = version = '3.3.3'
+__version_tuple__ = version_tuple = (3, 3, 3)
 
 __commit_id__ = commit_id = None

{horde_engine-3.3.2 → horde_engine-3.3.3}/hordelib/comfy_horde.py

@@ -405,6 +405,10 @@ def pin_models_in_vram() -> bool:
 def unload_all_models_vram():
     global _comfy_current_loaded_models
 
+    from hordelib.metrics import get_metrics_collector
+
+    _unload_start = time.perf_counter()
+
     log_free_ram()
 
     from hordelib.shared_model_manager import SharedModelManager
@@ -448,6 +452,12 @@ def unload_all_models_vram():
         clear_gc_and_torch_cache()
     log_free_ram()
 
+    # Surface the VRAM-eviction cost to the in-process collector so an embedder can attribute the
+    # between-jobs reload churn (this is the "unload" half; the reload is already timed as
+    # disk_to_ram/ram_to_vram). Recorded under the job in progress when the unload happens, or the
+    # next job on this process when it is an idle eviction.
+    get_metrics_collector().record_phase("model_unload", time.perf_counter() - _unload_start)
+
 
 def unload_all_models_ram():
     global _comfy_current_loaded_models