Skip to content

WIP: Integrate SpQR + FSDP functionalities - #840

Closed
Titus-von-Koeller wants to merge 47 commits into
bitsandbytes-foundation:mainfrom
Titus-von-Koeller:spqr_and_fsdp
Closed

WIP: Integrate SpQR + FSDP functionalities#840
Titus-von-Koeller wants to merge 47 commits into
bitsandbytes-foundation:mainfrom
Titus-von-Koeller:spqr_and_fsdp

Conversation

@Titus-von-Koeller

Copy link
Copy Markdown
Collaborator

No description provided.

@Titus-von-KoellerTitus-von-Koeller changed the title WIP: SpQR and FSDP integrationsWIP: Integrate SpQR + FSDP functionalitiesOct 23, 2023
@github-actions

Copy link
Copy Markdown

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

@TimDettmersTimDettmers reopened this Jan 1, 2024
@TimDettmersTimDettmers added High Priority (first issues that will be worked on) High Risk Risk of bugs in transformers and other libraries labels Jan 1, 2024
Titus-von-Koeller pushed a commit that referenced this pull request Jan 17, 2024
This PR adds initial FSDP support for training QLoRA models. It enables basic FSDP and CPU Offload support, with low memory training via FSDP.sync_module_states option unsupported.
This PR builds off of #840 commit 8278fca and BNB FSDP by @TimDettmers and @Titus-von-Koeller.
An example of using this PR to finetune QLoRA models with FSDP can be found in the demo repo: AnswerDotAi/fsdp_qlora.
* Minimal changes for fp32 4bit storage from BNB commit 8278fca
* Params4bit with selectable storage dtype
* possible fix for double quantizing linear weight & quant storage dtype
* minor fixes in Params4bit for peft tests
* remove redundant
* add float16
* update test
* Remove float16 quant cast as there are fp32, bf16, & fp16 quant kernels
---------
Co-authored-by: Kerem Turgutlu <keremturgutlu@gmail.com>
Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Labels

High Priority(first issues that will be worked on)High RiskRisk of bugs in transformers and other libraries

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants

@Titus-von-Koeller@TimDettmers@younesbelkada