| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | """Qwen4Exp model package.""" |
| 4 | |
| 5 | from typing import TYPE_CHECKING, Any |
| 6 | |
| 7 | from .common.hyperconnection import ( |
| 8 | GatedResidual, |
| 9 | GroupedGemmaRMSNorm, |
| 10 | HyperConnectionBase, |
| 11 | HyperConnectionConfig, |
| 12 | ) |
| 13 | |
| 14 | if TYPE_CHECKING: |
| 15 | from .nvidia.model import ( |
| 16 | Qwen4ExpForCausalLM, |
| 17 | Qwen4ExpForConditionalGeneration, |
| 18 | ) |
| 19 | from .nvidia.mtp import Qwen4ExpMTP |
| 20 | |
| 21 | |
| 22 | def __getattr__(name: str) -> Any: |
| 23 | if name in { |
| 24 | "Qwen4ExpForCausalLM", |
| 25 | "Qwen4ExpForConditionalGeneration", |
| 26 | "Qwen4ExpMTP", |
| 27 | }: |
| 28 | from vllm.platforms import current_platform |
| 29 | |
| 30 | if current_platform.is_xpu() or current_platform.is_tpu(): |
| 31 | raise NotImplementedError("Qwen4Exp currently supports CUDA and ROCm only") |
| 32 | if current_platform.is_rocm(): |
| 33 | from .amd.model import ( |
| 34 | Qwen4ExpForCausalLM, |
| 35 | Qwen4ExpForConditionalGeneration, |
| 36 | ) |
| 37 | from .amd.mtp import Qwen4ExpMTP |
| 38 | else: |
| 39 | from .nvidia.model import ( |
| 40 | Qwen4ExpForCausalLM, |
| 41 | Qwen4ExpForConditionalGeneration, |
| 42 | ) |
| 43 | from .nvidia.mtp import Qwen4ExpMTP |
| 44 | |
| 45 | return { |
| 46 | "Qwen4ExpForCausalLM": Qwen4ExpForCausalLM, |
| 47 | "Qwen4ExpForConditionalGeneration": (Qwen4ExpForConditionalGeneration), |
| 48 | "Qwen4ExpMTP": Qwen4ExpMTP, |
| 49 | }[name] |
| 50 | raise AttributeError(name) |
| 51 | |
| 52 | |
| 53 | __all__ = [ |
| 54 | "GatedResidual", |
| 55 | "GroupedGemmaRMSNorm", |
| 56 | "HyperConnectionBase", |
| 57 | "HyperConnectionConfig", |
| 58 | "Qwen4ExpForCausalLM", |
| 59 | "Qwen4ExpForConditionalGeneration", |
| 60 | "Qwen4ExpMTP", |
| 61 | ] |