[Model][NVIDIA] Route DSA models to the CUDA non-compiled path (#52861)
This commit is contained in:
@@ -1679,6 +1679,11 @@ _SPECULATIVE_DECODING_EXAMPLE_MODELS = {
|
||||
speculative_model="luccafong/deepseek_mtp_draft_random",
|
||||
trust_remote_code=True,
|
||||
),
|
||||
"DeepseekV32MTPModel": _HfExamplesInfo(
|
||||
"nvidia/DeepSeek-V3.2-NVFP4",
|
||||
speculative_model="nvidia/DeepSeek-V3.2-NVFP4",
|
||||
is_available_online=False,
|
||||
),
|
||||
"DeepSeekV4MTPModel": _HfExamplesInfo(
|
||||
"deepseek-ai/DeepSeek-V4-Flash",
|
||||
speculative_model="deepseek-ai/DeepSeek-V4-Flash",
|
||||
|
||||
Reference in New Issue
Block a user