Adding StateDictAdapter #1601
Adding StateDictAdapter #1601
Conversation
meta-cla
bot
added
the
CLA Signed
Fixed |
|
@HosseinKaviani-H has imported this pull request. If you are a Meta employee, you can view this in D80660953. |
|
@HosseinKaviani-H has imported this pull request. If you are a Meta employee, you can view this in D80660953. |
|
@HosseinKaviani-H has imported this pull request. If you are a Meta employee, you can view this in D80660953. |
|
@HosseinKaviani-H has imported this pull request. If you are a Meta employee, you can view this in D80660953. |
|
Can you check again what's going on with this PR? Do you export the latest version of the diff? |
|
@HosseinKaviani-H has imported this pull request. If you are a Meta employee, you can view this in D80660953. |
I did and imported to the base again. I don't know why it's throwing an error like this. |
pytorch-bot
bot
dismissed stale reviews from wwwjn and wwwjn
This PR was reopened (likely due to being reverted), so your approval was removed. Please request another review.
|
Please check the internal diff message. |
|
I haven't meet this error before, and please check the fix the signal |
In this PR, I'm adding the StateDictAdapter for Qwen3 to enable loading HF checkpoints. We can use this script to adapt the checkpoint from HF to the format that we can load into the torchtitan model and vice versa. This can enable us to do a parity test with the HF implementation and make sure that our results are aligned with the HF implementation. --------- Co-authored-by: Hossein Kavianihamedani <[email protected]>
In this PR, I'm adding the StateDictAdapter for Qwen3 to enable loading HF checkpoints. We can use this script to adapt the checkpoint from HF to the format that we can load into the torchtitan model and vice versa. This can enable us to do a parity test with the HF implementation and make sure that our results are aligned with the HF implementation.