Search results

Llama3.2 vision does not run with distributed state dict

Status: Open.

#2277 In pytorch/torchtune;

· acisseJZhong opened on Jan 17, 2025

The current instantiation does not trigger the initialization of submodules

discussionStart a discussionStart a discussion

triagedThis issue has been assigned an owner and appropriate labelThis issue has been assigned an owner and appropriate label

Status: Open.

#2273 In pytorch/torchtune;

· dz1iang opened on Jan 16, 2025

DPO after / on top of LoRA tuning

discussionStart a discussionStart a discussion

triagedThis issue has been assigned an owner and appropriate labelThis issue has been assigned an owner and appropriate label

Status: Open.

#2272 In pytorch/torchtune;

· albertbn opened on Jan 16, 2025

Don't use _get_clones

best practiceThings we should be doing but aren'tThings we should be doing but aren't

community help wantedWe would love the community's help completing this issueWe would love the community's help completing this issue

Status: Open.

#2270 In pytorch/torchtune;

· ebsmothers opened on Jan 16, 2025

About the CLS token for the llama3_2_vision_encoder

discussionStart a discussionStart a discussion

triagedThis issue has been assigned an owner and appropriate labelThis issue has been assigned an owner and appropriate label

Status: Open.

#2268 In pytorch/torchtune;

· dfloreaa opened on Jan 15, 2025

Expose FSDP2 MixedPrecisionPolicy params

enhancementNew feature or requestNew feature or request

triagedThis issue has been assigned an owner and appropriate labelThis issue has been assigned an owner and appropriate label

Status: Open.

#2267 In pytorch/torchtune;

· EugenHotaj opened on Jan 14, 2025

adding support for LR schedule for full distributed finetune

best practiceThings we should be doing but aren'tThings we should be doing but aren't

better engineeringTasks which help improve eng productivity e.g. building tools, cleaning up code, writing docsTasks which help improve eng productivity e.g. building tools, cleaning up code, writing docs

triagedThis issue has been assigned an owner and appropriate labelThis issue has been assigned an owner and appropriate label

Status: Open.

#2263 In pytorch/torchtune;

· tginart opened on Jan 13, 2025

[RFC] Additional chat loss masking strategies

community help wantedWe would love the community's help completing this issueWe would love the community's help completing this issue

discussionStart a discussionStart a discussion

enhancementNew feature or requestNew feature or request

good first issueGood for newcomersGood for newcomers

rfcRequest for commentsRequest for comments

Status: Open.

#2261 In pytorch/torchtune;

· RdoubleA opened on Jan 13, 2025

Request: adding py.typed for type checkers

better engineeringTasks which help improve eng productivity e.g. building tools, cleaning up code, writing docsTasks which help improve eng productivity e.g. building tools, cleaning up code, writing docs

triagedThis issue has been assigned an owner and appropriate labelThis issue has been assigned an owner and appropriate label

Status: Open.

#2258 In pytorch/torchtune;

· jamesbraza opened on Jan 13, 2025

Qlora uses more memory than regular lora

triagedThis issue has been assigned an owner and appropriate labelThis issue has been assigned an owner and appropriate label

Status: Open.

#2255 In pytorch/torchtune;

· AndrewMead10 opened on Jan 11, 2025

Very slow convergence with bf16

Status: Open.

#2254 In pytorch/torchtune;

· EugenHotaj opened on Jan 11, 2025

Lora and Dora finetuning produces identical results

bugSomething isn't workingSomething isn't working

high-priority

Status: Open.

#2250 In pytorch/torchtune;

· AndrewMead10 opened on Jan 10, 2025