support for granite-io backend in interpreter

Signed-off-by: Mandana Vaziri <mvaziri@us.ibm.com>

examples/hello/hello-graniteio.pdl

@@ -0,0 +1,5 @@
+text:
+- "Hello!\n"
+- model: ibm-granite/granite-3.2-8b-instruct-preview
+  backend:
+    transformers: cpu

src/pdl/pdl_ast.py

@@ -378,6 +378,7 @@ class GraniteioModelBlock(ModelBlock):
     model: ExpressionType[object]
     platform: Literal[ModelPlatform.GRANITEIO] = ModelPlatform.GRANITEIO
     intrinsics: ExpressionType[list[GraniteioIntrinsicType]] = []
+    backend: ExpressionType[dict[str, Any]]
 
 
 class CodeBlock(Block):

src/pdl/pdl_llms.py

@@ -1,9 +1,15 @@
 import asyncio
 import os
+import json
+import aconfig
 import threading
 from concurrent.futures import Future
 from typing import Any, Callable, Generator, TypeVar
 
+from granite_io.io.granite_3_2 import Granite3Point2InputOutputProcessor
+from granite_io.backend.transformers import TransformersBackend
+from granite_io.io.base import ChatCompletionInputs
+
 import httpx
 import litellm
 from dotenv import load_dotenv
@@ -141,8 +147,15 @@ async def async_generate_text(
         messages: ModelInput,
     ) -> tuple[dict[str, Any], Any]:
         try:
-            outputs = block.model.process(messages)  # type: ignore # TODO
-            return outputs.response, outputs
+            if "transformers" in block.backend:    
+                input_json_str = json.dumps({"messages": messages})
+                inputs = ChatCompletionInputs.model_validate_json(input_json_str)
+                io_processor = Granite3Point2InputOutputProcessor(
+                    TransformersBackend(aconfig.Config({"model_name":block.model, "device":block.backend["transformers"]})),
+                )
+
+                result = io_processor.create_chat_completion(inputs)
+                return result.next_message.model_dump(), result.next_message.model_dump()
         except Exception as exc:
             message = f"Error during '{block.model}' model call: {repr(exc)}"
             loc = block.location