[mcore] verl+megatron development tracking

[ISEEKYAN]

veRL Megatron-core Development Tracking

This page focuses on development of verl+mcore.
The milestone target is to enable training deepseek-v3 on veRL as #708 and the further target is to continuously enhance the verl training experience of the mcore backend.

Progress and TODO

Recent

• update mcore version to 0.11 megatron：Update megatron-lm to core_r0.11.0 #392
• use mcore GPTModel api instead of huggingface workaround with sequence packing Use Mcore GPTModel #706
• support context parallel [Mcore] context parallel #970
• support loading mcore dist_checkpointing [mcore] option to use dist checkpoint #1030
• support Megatron 0.11.0 and vLLM 0.8.2 Support Megatron 0.11.0 and vLLM 0.8.2, update images to use latest vllm and Megatron #851
• support qwen2moe training [mcore] qwen2moe support #1139
• support Moonlight-16B-A3B training (WIP) [mcore] moonlight (small model with deepseekv3 arch) #1284
• support Qwen2.5-VL training [megatron] feat: qwen2.5vl #1286
• support EP(expert parallel) [megatron] support megatron expert parallel #1467

Further

• FP8 training
• training efficiency related optimization
• support sglang inference engine
• support trtllm inference engine

[eric-haibin-lin]

Could you merge the todo list from this as well? #825

[ISEEKYAN]

Could you merge the todo list from this as well? #825

ok

[rbao2018]

mark👀

[yushuiwx]

mark👀

[Aurelius84]

mark👀

[Wolfwjs]

How does the mcore backend support make_vocab_size_divisible_by, and where to pad the vocabulary to meet the splitting requirements?

[supermancmk]

Could you support qwen to use Megatron for SFT training?

[eric-haibin-lin]

Referencing a related issue: #708