Training a one-stage model

TRI-ML / prismatic-vlms

A flexible and efficient codebase for training visually-conditioned language models (VLMs)

MIT License

425 stars 194 forks source link

Closed shikhar-srivastava closed 6 months ago

shikhar-srivastava commented 6 months ago

Hey @siddk-tri @siddk !

Can you clarify what the command to reproduce the one-stage model training is?

In particular, it's unclear what stage specification, dataset specification is to be provided.

Thanks!

shikhar-srivastava commented 6 months ago

This was solved. Thanks.

zjysteven commented 4 months ago

@shikhar-srivastava I'm actually having the same question. I guess it would be just running with stage="finetune"?

siddk commented 4 months ago

Yeah -- you just set stage="finetune" for single stage training!