Add LoRA co-training support for HF EAGLE speculative decoding#1060

Open

yeyu-nvidia wants to merge 7 commits intomainfrom

yeyu/speculative-lora-cotrain

Commits on Apr 10, 2026

Fix dp_shard_size for multinode: use WORLD_SIZE instead of local GPU count
yeyu-nvidia
committed

Commits on Apr 15, 2026

fix: rope_theta compatibility for EAGLE3 LoRA export and merge

yeyu-nvidia
and
claude
committed