| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | """CPU DeepSeek-V4 DSpark (speculative decoding) — deferred, not implemented. |
| 4 | |
| 5 | ``DSparkDeepseekV4ForCausalLM`` is only instantiated when the speculative- |
| 6 | decoding registry resolves it by name for a DSpark speculative config; a |
| 7 | plain (non-spec) DeepSeek-V4 CPU model never constructs this class. This |
| 8 | stub exists solely so ``vllm.models.deepseek_v4.cpu`` is import-clean. |
| 9 | """ |
| 10 | |
| 11 | from vllm.config import VllmConfig |
| 12 | |
| 13 | |
| 14 | class DSparkDeepseekV4ForCausalLM: |
| 15 | def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None: |
| 16 | raise NotImplementedError( |
| 17 | "DeepSeek-V4 speculative decoding (DSpark) is not implemented on CPU." |
| 18 | ) |