lightseekorg--tokenspeed/python/tokenspeed/runtime/sampling/custom_logit_processor.py

# Copyright (c) 2026 LightSeek Foundation
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in
# all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.

import json
from abc import ABC, abstractmethod
from functools import cache
from typing import Any, Self

import dill
import torch


@cache
def _cache_from_str(json_str: str) -> "CustomLogitProcessor":
    """Deserialize a json string to a Callable object.
    This function is cached to avoid redundant deserialization.
    """
    data = json.loads(json_str)
    return dill.loads(bytes.fromhex(data["callable"]))


class CustomLogitProcessor(ABC):
    """Abstract base class for callable functions."""

    @abstractmethod
    def __call__(
        self,
        logits: torch.Tensor,
        custom_param_list: list[dict[str, Any]] | None = None,
    ) -> torch.Tensor:
        """Define the callable behavior."""
        raise NotImplementedError

    def to_str(self) -> str:
        """Serialize the callable function to a JSON-compatible string."""
        return json.dumps({"callable": dill.dumps(self).hex()})

    @classmethod
    def from_str(cls, json_str: str) -> Self:
        """Deserialize a callable function from a JSON string."""
        return _cache_from_str(json_str)