[Refactor] Normalize DeepSeek V4.1 model package naming (#56741)

Signed-off-by: Bugen Zhao <i@bugenzhao.com>
This commit is contained in:
Bugen Zhao
2026-09-13 23:17:50 -07:00
committed by GitHub
parent 10e8d5614c
commit cf1584f373
44 changed files with 71 additions and 72 deletions
+2 -2
View File
@@ -18,8 +18,8 @@ from vllm.models.deepseek_v4.nvidia.model import (
)
from vllm.models.deepseek_v4.nvidia.mtp import DeepSeekV4MTP
from vllm.models.deepseek_v4.nvidia.ops.prepare_megamoe import prepare_megamoe_inputs
from vllm.models.deepseek_v4_1.common.mm_preprocess import IMAGE_SENTINEL_BASE_ID
from vllm.models.deepseek_v4_1.nvidia.model import DeepseekV4MoE as DeepseekV41MoE
from vllm.models.deepseek_v41.common.mm_preprocess import IMAGE_SENTINEL_BASE_ID
from vllm.models.deepseek_v41.nvidia.model import DeepseekV4MoE as DeepseekV41MoE
from vllm.platforms import current_platform
from vllm.transformers_utils.configs.deepseek_v41 import DeepseekV41Config
+1 -1
View File
@@ -216,7 +216,7 @@ def test_v41_dspark_loads_linear_scales(
"""Checkpoint ``.scale`` maps to the quant method's scale parameter and
loads untouched. MXFP8 block-scale expansion lives in the KMxfp8Static
loader (see tests/quantization/test_modelopt.py), not in load_weights."""
from vllm.models.deepseek_v4_1.nvidia import dspark
from vllm.models.deepseek_v41.nvidia import dspark
mxfp8 = scale_dtype != torch.float32
scale_name = "weight_scale" if mxfp8 else "weight_scale_inv"