Future plan about extending to post-train (RL/SFT/OPD)

[zhihanliu-collab]

Awesome work! Wondering if you will extend the current work to the post training like RL/SFT/OPD and use them for the auto-research?

[haok1402]

@zhihanliu-collab Thanks for your interest in our work! We start with pre-training but never intend to stop there. At the moment, we're working on the post-training part of the framework. It will too follow the four agent-native design principles, which could mean that for auto-research, agents can operate more easily.

[haok1402]

@zhihanliu-collab By the way, if you're open to sharing, we'd love to hear a bit more about your auto-research setup. Roughly what model scale are you running or planning for? And what context length do you usually need? Whatever you're comfortable sharing would help us think about where the infra could be optimized for use cases like yours.

[zhihanliu-collab]

Let us talk via wechat? cloutain666 Look forward to hearing from you. Best, Zhihan

…

On Wed, Jun 3, 2026 at 5:32 PM Hao Kang ***@***.***> wrote: *haok1402* left a comment (mlc-ai/pith-train#49) <#49 (comment)> @zhihanliu-collab <https://github.com/zhihanliu-collab> By the way, if you're open to sharing, we'd love to hear a bit more about your auto-research setup. Roughly what model scale are you running or planning for? And what context length do you usually need? Whatever you're comfortable sharing would help us think about where the infra could be optimized for use cases like yours. — Reply to this email directly, view it on GitHub <#49?email_source=notifications&email_token=CA7Y4J4W3SQ2LGRSC6F2NOL46C7QTA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTINRRG44DOMJXHE2KM4TFMFZW63VHNVSW45DJN5XKKZLWMVXHJLDGN5XXIZLSL5RWY2LDNM#issuecomment-4617871794>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/CA7Y4J6GNG3MV74MPNDFVOD46C7QTAVCNFSM6AAAAACZZOA4XKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DMMJXHA3TCNZZGQ> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[haok1402]

Thanks Zhihan! We'd prefer to keep the discussion here on the GitHub issue if that works for you, so others following along can benefit too and contribute insights. We can also discuss over email.