A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 3, 08:00 UTC
[Model Refactoring] Rename deepseek_v4.py to model.py [4/N] (#43077)
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Woosuk Kwon committed
May 19, 2026 at 01:12 UTC
07beaed8422d2df34a20e8ebd22b7924d563a566
8 files changed
+8
-8
tests/models/test_deepseek_v4_mega_moe.py
+1
-1
@@ -6,7 +6,7 @@ from types import SimpleNamespace
6
import pytest
7
import torch
8
9
-from vllm.models.deepseek_v4.nvidia.deepseek_v4 import (
9
+from vllm.models.deepseek_v4.nvidia.model import (
10
DeepseekV4MegaMoEExperts,
11
_stage_deepseek_v4_mega_moe_inputs,
12
make_deepseek_v4_expert_params_mapping,
vllm/models/deepseek_v4/__init__.py
+4
-4
@@ -17,11 +17,11 @@ from .quant_config import DeepseekV4FP8Config
17
# default that mypy sees; the ROCm branch overrides it at runtime and is
18
# kept type-compatible via ``# type: ignore[assignment]``.
19
if TYPE_CHECKING or not current_platform.is_rocm():
20
- from .nvidia.deepseek_v4 import DeepseekV4ForCausalLM
21
- from .nvidia.deepseek_v4_mtp import DeepSeekV4MTP
20
+ from .nvidia.model import DeepseekV4ForCausalLM
21
+ from .nvidia.mtp import DeepSeekV4MTP
22
else:
23
- from .amd.deepseek_v4 import DeepseekV4ForCausalLM # type: ignore[assignment]
24
- from .amd.deepseek_v4_mtp import DeepSeekV4MTP # type: ignore[assignment]
23
+ from .amd.model import DeepseekV4ForCausalLM # type: ignore[assignment]
24
+ from .amd.mtp import DeepSeekV4MTP # type: ignore[assignment]
25
26
__all__ = [
27
"DeepSeekV4MTP",
vllm/models/deepseek_v4/amd/deepseek_v4.py
deleted
-1
@@ -1 +0,0 @@
1
-../nvidia/deepseek_v4.py
\ No newline at end of file
vllm/models/deepseek_v4/amd/deepseek_v4_mtp.py
deleted
-1
@@ -1 +0,0 @@
1
-../nvidia/deepseek_v4_mtp.py
\ No newline at end of file
vllm/models/deepseek_v4/amd/model.py
new
+1
@@ -0,0 +1 @@
1
+../nvidia/model.py
\ No newline at end of file
vllm/models/deepseek_v4/amd/mtp.py
new
+1
@@ -0,0 +1 @@
1
+../nvidia/mtp.py
\ No newline at end of file
vllm/models/deepseek_v4/nvidia/model.py
renamed
vllm/models/deepseek_v4/nvidia/mtp.py
renamed
+1
-1
@@ -40,7 +40,7 @@ from vllm.model_executor.models.utils import maybe_prefix
40
from vllm.platforms import current_platform
41
from vllm.sequence import IntermediateTensors
42
43
-from .deepseek_v4 import (
43
+from .model import (
44
DeepseekV4DecoderLayer,
45
make_deepseek_v4_expert_params_mapping,
46
)