feat: add BaseLogitsProcessor core interface (#2613)

Signed-off-by: Bhuvan Agrawal <[email protected]>

Author: bhuvan002

lib/bindings/python/src/dynamo/logits_processing/__init__.py

@@ -0,0 +1,13 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""
+Dynamo Logits Processing - Backend-agnostic logits processors.
+
+This module provides the BaseLogitsProcessor protocol that can be used
+across different backend adapters (TRT-LLM, vLLM, SGLang).
+"""
+
+from .base import BaseLogitsProcessor
+
+__all__ = ["BaseLogitsProcessor"]

lib/bindings/python/src/dynamo/logits_processing/base.py

@@ -0,0 +1,39 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+"""
+Base logits processor protocol for Dynamo.
+
+This module defines the core BaseLogitsProcessor interface that all
+logits processors must implement.
+"""
+
+from typing import Protocol, Sequence
+
+import torch
+
+
+class BaseLogitsProcessor(Protocol):
+    """
+    Protocol for logits processors in Dynamo.
+
+    All logits processors must implement this interface to be compatible
+    with backend adapters (TRT-LLM, vLLM, SGLang).
+    """
+
+    def __call__(
+        self,
+        input_ids: Sequence[int],
+        logits: torch.Tensor,
+    ) -> torch.Tensor:
+        """
+        Process the logits for the next token prediction.
+
+        Args:
+            input_ids: The input token IDs generated so far.
+            logits: The raw logits for the next token. Shape: (vocab_size,)
+
+        Returns:
+            A tensor with the same shape, dtype, and device as `logits`.
+        """
+        ...