add more in-line docs

packages/vllm/src/main.py

@@ -35,7 +35,7 @@ def clamp(n: float | int, smallest: float | int, largest: float | int):
 
 
 class RandomAsyncIterator:
-    """Manages multiple async iterables and allows iterating over them in a random order."""
+    """Manages multiple async iterables and iterates over them in a random order."""
 
     def __init__(self, async_iterables):
         # Convert each iterable into an async iterator
@@ -57,6 +57,8 @@ async def __anext__(self):
         except StopAsyncIteration:
             # If the selected iterator is exhausted, remove it from the list
             del self.async_iterators[random_index]
+
+            # Continue to the next iterator
             if self.async_iterators:
                 return await self.__anext__()
 
@@ -80,6 +82,8 @@ def remove_iterator(self, async_iterable):
 
 
 def get_backend_configs():
+    """Get vLLM generation parameters from Confz environment variables"""
+
     # Manually load "STOP_TOKENS", since Confz can't handle complex types (i.e., lists)
     stop_tokens: str | None = os.getenv("LAI_STOP_TOKENS")
     processed_stop_tokens = json.loads(stop_tokens) if stop_tokens else []
@@ -113,6 +117,7 @@ def get_backend_configs():
 
 
 def get_config_from_request(request: ChatCompletionRequest | CompletionRequest):
+    """Get vLLM generation parameters from the request object"""
     return GenerationConfig(
         max_new_tokens=request.max_new_tokens,
         temperature=request.temperature,
@@ -151,11 +156,11 @@ def __init__(self):
         self.backend_config = get_backend_configs()
         self.model = self.backend_config.model_path
         self.engine_args = AsyncEngineArgs(
-            # Taken from LFAI SDK config pattern
+            # Taken from LFAI SDK environment variables
             model=self.backend_config.model_path,
             max_model_len=self.backend_config.max_context_length,
             max_seq_len_to_capture=self.backend_config.max_context_length,
-            # Taken from Confz
+            # Taken from Confz environment variables
             engine_use_ray=AppConfig().backend_options.engine_use_ray,
             worker_use_ray=AppConfig().backend_options.worker_use_ray,
             gpu_memory_utilization=AppConfig().backend_options.gpu_memory_utilization,
@@ -209,6 +214,7 @@ async def iterate_outputs(self):
                         request_output.outputs[0].token_ids
                     )
 
+                    # Add the result to the queue for this request
                     await self.delta_queue_by_id[request_id].put(text_delta)
             await asyncio.sleep(0)
 
@@ -254,6 +260,7 @@ async def generate(
         request_id = random_uuid()
         self.done_by_id[request_id] = False
 
+        # Create new request in the queue
         await self.generate_session(request_id, prompt, config)
 
         logging.info(f"Begin reading the output for request {request_id}")