[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
@@ -120,7 +120,7 @@ def get_quantization_config(quantization: str) -> type[QuantizationConfig]:
# The v4.1 class is the v4 one extended to accept model_type
# "deepseek_v41" and its 32x32 MXFP8 linear layout. V4.1 has
# platform-specific implementations for both CUDA and ROCm.
from vllm.models.deepseek_v4_1 import (
from vllm.models.deepseek_v41 import (
DeepseekV4FP8Config as DeepseekV41FP8Config,
)
+2 -2
View File
@@ -377,7 +377,7 @@ _MULTIMODAL_MODELS = {
"DeepseekV4ForConditionalGeneration",
),
"DeepseekV41ForCausalLM": (
"vllm.models.deepseek_v4_1",
"vllm.models.deepseek_v41",
"DeepseekV41ForCausalLM",
),
"Dots3NoteForCausalLM": (
@@ -645,7 +645,7 @@ _SPECULATIVE_DECODING_MODELS = {
"DFlashMuseGlimmerAssistantModel": ("qwen3_dflash", "DFlashQwen3ForCausalLM"),
"DSparkDraftModel": ("vllm.models.deepseek_v4", "DSparkDeepseekV4ForCausalLM"),
"DSparkV41DraftModel": (
"vllm.models.deepseek_v4_1",
"vllm.models.deepseek_v41",
"DSparkDeepseekV4ForCausalLM",
),
"Qwen3DSparkModel": ("qwen3_dspark", "Qwen3DSparkForCausalLM"),