Skip to content

Conversation

@winglian
Copy link
Collaborator

@winglian winglian commented Mar 8, 2024

Description

AnswerAI managed to get qlora+fsdp working, taking a stab at implementing the fundamentals.

https://github.com/AnswerDotAI/fsdp_qlora/blob/47c17224b86425cd9cf03c1d7c760a5035656859/train.py

cc: @jph00 @johnowhitaker

@hamelsmu
Copy link
Collaborator

hamelsmu commented Mar 8, 2024

I'm really excited about this

@hamelsmu hamelsmu changed the title WIP: Fsdp qlora Fsdp + qlora Mar 8, 2024
@hamelsmu hamelsmu changed the title Fsdp + qlora FDSP + QLoRA Mar 8, 2024
@hamelsmu hamelsmu added the enhancement New feature or request label Mar 8, 2024
@winglian winglian merged commit 9b6ee83 into main Mar 8, 2024
@winglian winglian deleted the fsdp-qlora branch March 8, 2024 20:01
djsaunde pushed a commit that referenced this pull request Dec 17, 2024
* wip qlora + fsdp fixes

* more fixes

* make sure to load the lora 🤦

* only setup quantized meta on non-zero rank:

* only run setup_quantized_peft_meta_for_training for qlora+fsdp

* more fixes for qlora+fsdp

* chore: lint

* add example yml

* support mistral too

* fix for model_type and add mixtral support too

* set cpu_offload: false to reduce vram, constrain new accleerator logic to qlora + fsdp

* refactor for duplicate code
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

enhancement New feature or request

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants