[serve][llm][refactor] Align Ray Serve LLM Code Structure with Architectural Design (#57889)

Signed-off-by: Kourosh Hakhamaneshi <[email protected]>
Signed-off-by: elliot-barn <[email protected]>

Author: kouroshHakha

python/ray/llm/_internal/common/callbacks/base.py

@@ -6,7 +6,7 @@
 
 if TYPE_CHECKING:
     from ray.llm._internal.common.utils.download_utils import NodeModelDownloadable
-    from ray.llm._internal.serve.configs.server_models import LLMConfig
+    from ray.llm._internal.serve.core.configs.llm_config import LLMConfig
 
 logger = logging.getLogger(__name__)
 

python/ray/llm/_internal/common/callbacks/cloud_downloader.py

@@ -39,7 +39,7 @@ class CloudDownloader(CallbackBase):
     Example:
         ```
         from ray.llm._internal.common.callbacks.cloud_downloader import CloudDownloader
-        from ray.llm._internal.serve.configs.server_models import LLMConfig
+        from ray.llm._internal.serve.core.configs.llm_config import LLMConfig
         config = LLMConfig(
             ...
             callback_config={

python/ray/llm/_internal/serve/config_generator/utils/constants.py

@@ -1,6 +1,6 @@
 import os
 
-from ray.llm._internal.serve.configs.constants import RAYLLM_HOME_DIR
+from ray.llm._internal.serve.constants import RAYLLM_HOME_DIR
 
 TEMPLATE_DIR = os.path.normpath(
     os.path.join(

python/ray/llm/_internal/serve/config_generator/utils/gpu.py

@@ -7,7 +7,7 @@
     DEFAULT_DEPLOYMENT_CONFIGS_FILE,
     TEMPLATE_DIR,
 )
-from ray.llm._internal.serve.configs.server_models import GPUType
+from ray.llm._internal.serve.core.configs.llm_config import GPUType
 
 # All practical GPUs
 ALL_GPU_TYPES = [

python/ray/llm/_internal/serve/config_generator/utils/text_completion.py

@@ -15,7 +15,7 @@
 from ray.llm._internal.serve.config_generator.utils.models import (
     TextCompletionModelConfig,
 )
-from ray.llm._internal.serve.configs.server_models import LLMConfig
+from ray.llm._internal.serve.core.configs.llm_config import LLMConfig
 
 
 def get_model_default_config(model_id: str) -> Dict[str, Any]:

python/ray/llm/_internal/serve/configs/constants.py renamed to python/ray/llm/_internal/serve/constants.py

@@ -28,12 +28,12 @@
 )
 from ray.llm._internal.common.utils.download_utils import NodeModelDownloadable
 from ray.llm._internal.common.utils.import_utils import load_class, try_import
-from ray.llm._internal.serve.configs.constants import (
+from ray.llm._internal.serve.constants import (
     DEFAULT_MULTIPLEX_DOWNLOAD_TIMEOUT_S,
     DEFAULT_MULTIPLEX_DOWNLOAD_TRIES,
     MODEL_RESPONSE_BATCH_TIMEOUT_MS,
 )
-from ray.llm._internal.serve.deployments.llm.vllm.kv_transfer_backends import (
+from ray.llm._internal.serve.engines.vllm.kv_transfer import (
     SUPPORTED_BACKENDS as SUPPORTED_KV_CONNECTOR_BACKENDS,
 )
 from ray.llm._internal.serve.observability.logging import get_logger
@@ -455,7 +455,7 @@ def get_engine_config(self) -> EngineConfigType:
             return self._engine_config
 
         if self.llm_engine == LLMEngine.vLLM:
-            from ray.llm._internal.serve.deployments.llm.vllm.vllm_models import (
+            from ray.llm._internal.serve.engines.vllm.vllm_models import (
                 VLLMEngineConfig,
             )
 

python/ray/llm/_internal/serve/configs/__init__.py renamed to python/ray/llm/_internal/serve/core/__init__.py

@@ -29,7 +29,7 @@
 from vllm.utils import random_uuid
 
 if TYPE_CHECKING:
-    from ray.llm._internal.serve.configs.server_models import LLMConfig
+    from ray.llm._internal.serve.core.configs.llm_config import LLMConfig
 
 
 class ChatCompletionRequest(vLLMChatCompletionRequest):
@@ -177,6 +177,9 @@ def to_model_metadata(
         model_config: The model's YAML config.
         overrides: should only be set for LoRA fine-tuned models. The
             overrides of the fine-tuned model metadata.
+
+    Returns:
+        A ModelCard object.
     """
     metadata = {
         "model_id": model_config.model_id,