Skip to content

feat(RL): add RL support for verl#1298

Open
shihaobai wants to merge 259 commits into
mainfrom
rl_verl_rebase_main
Open

feat(RL): add RL support for verl#1298
shihaobai wants to merge 259 commits into
mainfrom
rl_verl_rebase_main

Conversation

@shihaobai

Copy link
Copy Markdown
Collaborator

No description provided.

hiworldwzj and others added 30 commits June 17, 2026 02:42
… contract

- MTP draft 模型不再挂 routing capture 回调:draft 与主模型共享 KV 槽位,
  verify/prefill 后会用相同 mem_indexes 重写 layer-0 路由数据
- routed_experts 行数统一为 input_len + 可见输出长度 - 1(infer 侧与
  httpserver 读取侧一致);MTP 下 EOS 后被接受的 draft token 会继续推进
  cur_output_len/cur_kv_len,需用 shm_cur_output_len 截断
- extract_routing_data 改为全设备同步,scatter 核跑在 overlap/graph 流上
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants