transformers
[deepseek_v4] Add DeepseekV4NextNPredictor class for MTP draft-head support
#46127
Open
Open
[deepseek_v4] Add DeepseekV4NextNPredictor class for MTP draft-head support #46127
pasta-paul
wants to merge
2 commits into
huggingface:main
from
pasta-paul:dsv4-mtp-class
Add DeepseekV4NextNPredictor class for MTP draft-head support
e229b31d
Use layer_type=sliding_attention for MTP block (not last main-layer's…
9f565da0
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub