Files
vllm-project--vllm/vllm/transformers_utils/configs/deepseek_v4.py
T
wehub-resource-sync 7ce4c8e27e
pre-commit / pre-run-check (push) Has been cancelled
pre-commit / pre-commit (push) Has been cancelled
chore: import upstream snapshot with attribution
2026-07-13 12:55:37 +08:00

24 lines
737 B
Python
Executable File

# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from typing import Any
from transformers import PretrainedConfig
class DeepseekV4Config(PretrainedConfig):
model_type = "deepseek_v4"
def __init__(
self,
max_position_embeddings: int = 1048576,
rope_scaling: dict[str, Any] | None = None,
rope_parameters: dict[str, Any] | None = None,
rope_theta: float = 10000.0,
**kwargs,
):
self.max_position_embeddings = max_position_embeddings
self.rope_scaling = rope_scaling
self.rope_theta = rope_theta
self.rope_parameters = rope_scaling or rope_parameters
super().__init__(**kwargs)