| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | """DeepSeek V4 model — hardware-isolated entry point. |
| 4 | |
| 5 | The actual implementation lives under ``nvidia/`` and ``amd/``; this module |
| 6 | picks the right one for the current platform and re-exports the public |
| 7 | classes used by the model registry and quantization config lookup. |
| 8 | """ |
| 9 | |
| 10 | from vllm.platforms import current_platform |
| 11 | |
| 12 | from .quant_config import DeepseekV4FP8Config |
| 13 | |
| 14 | # Pick the per-platform implementation. The NVIDIA branch is the static |
| 15 | # default that mypy sees; the ROCm/XPU branches override at runtime and are |
| 16 | # kept type-compatible via ``# type: ignore[assignment]``. |
| 17 | if current_platform.is_rocm(): |
| 18 | from .amd.dspark import ( # type: ignore[assignment] |
| 19 | DSparkDeepseekV4ForCausalLM, |
| 20 | ) |
| 21 | from .amd.model import DeepseekV4ForCausalLM |
| 22 | from .amd.mtp import DeepSeekV4MTP |
| 23 | from .common.vl_model import ( # type: ignore[assignment] |
| 24 | DeepseekV4ForConditionalGeneration, |
| 25 | ) |
| 26 | elif current_platform.is_xpu(): |
| 27 | from .vl_stub import ( # type: ignore[assignment] |
| 28 | DeepseekV4ForConditionalGeneration, |
| 29 | ) |
| 30 | from .xpu.dspark import DSparkDeepseekV4ForCausalLM # type: ignore[assignment] |
| 31 | from .xpu.model import DeepseekV4ForCausalLM # type: ignore[assignment] |
| 32 | from .xpu.mtp import DeepSeekV4MTP # type: ignore[assignment] |
| 33 | elif current_platform.is_cpu(): |
| 34 | from .cpu.dspark import DSparkDeepseekV4ForCausalLM # type: ignore[assignment] |
| 35 | from .cpu.model import DeepseekV4ForCausalLM # type: ignore[assignment] |
| 36 | from .cpu.mtp import DeepSeekV4MTP # type: ignore[assignment] |
| 37 | from .vl_stub import ( # type: ignore[assignment] |
| 38 | DeepseekV4ForConditionalGeneration, |
| 39 | ) |
| 40 | else: |
| 41 | from .common.vl_model import ( # type: ignore[assignment] |
| 42 | DeepseekV4ForConditionalGeneration, |
| 43 | ) |
| 44 | from .nvidia.dspark import ( # type: ignore[assignment] |
| 45 | DSparkDeepseekV4ForCausalLM, |
| 46 | ) |
| 47 | from .nvidia.model import DeepseekV4ForCausalLM # type: ignore[assignment] |
| 48 | from .nvidia.mtp import DeepSeekV4MTP # type: ignore[assignment] |
| 49 | |
| 50 | __all__ = [ |
| 51 | "DSparkDeepseekV4ForCausalLM", |
| 52 | "DeepSeekV4MTP", |
| 53 | "DeepseekV4FP8Config", |
| 54 | "DeepseekV4ForCausalLM", |
| 55 | "DeepseekV4ForConditionalGeneration", |
| 56 | ] |