Model: Bypass lock checks when shutting down

Previously, when a SIGINT was emitted and a model load is running,
the API didn't shut down until the load finished due to waitng for
the lock. However, when shutting down, the lock doesn't matter since
the process is being killed anyway.

Signed-off-by: kingbri <bdashore3@proton.me>

backends/exllamav2/model.py

@@ -734,11 +734,15 @@ class ExllamaV2Container:
         Free all VRAM resources used by this model
         """
 
-        try:
-            await self.load_lock.acquire()
+        # Shutdown immediately unloads and bypasses all locks
+        do_shutdown = kwargs.get("shutdown")
 
-            # Wait for other jobs to finish
-            await self.wait_for_jobs(kwargs.get("skip_wait"))
+        try:
+            if not do_shutdown:
+                await self.load_lock.acquire()
+
+                # Wait for other jobs to finish
+                await self.wait_for_jobs(kwargs.get("skip_wait"))
 
             # Delete references held in the grammar module
             clear_grammar_func_cache()
@@ -778,10 +782,11 @@ class ExllamaV2Container:
 
             logger.info("Loras unloaded." if loras_only else "Model unloaded.")
         finally:
-            self.load_lock.release()
+            if not do_shutdown:
+                self.load_lock.release()
 
-            async with self.load_condition:
-                self.load_condition.notify_all()
+                async with self.load_condition:
+                    self.load_condition.notify_all()
 
     def encode_tokens(self, text: str, **kwargs):
         """Wrapper to encode tokens from a text string"""

common/model.py

@@ -43,11 +43,11 @@ def load_progress(module, modules):
     yield module, modules
 
 
-async def unload_model(skip_wait: bool = False):
+async def unload_model(skip_wait: bool = False, shutdown: bool = False):
     """Unloads a model"""
     global container
 
-    await container.unload(skip_wait=skip_wait)
+    await container.unload(skip_wait=skip_wait, shutdown=shutdown)
     container = None
 
 

common/signals.py

@@ -29,7 +29,7 @@ async def signal_handler_async(*_):
     """Internal signal handler. Runs all async code to shut down the program."""
 
     if model.container:
-        await model.unload_model(skip_wait=True)
+        await model.unload_model(skip_wait=True, shutdown=True)
 
     if model.embeddings_container:
         await model.unload_embedding_model()