Skip to content

fix: shared memory - #51

Merged
EYH0602 merged 2 commits into
mainfrom
fix/shared-memory
Feb 8, 2026
Merged

fix: shared memory#51
EYH0602 merged 2 commits into
mainfrom
fix/shared-memory

Conversation

@EYH0602

@EYH0602 EYH0602 commented Feb 8, 2026

Copy link
Copy Markdown
Member

No description provided.

SplitHeadWrapper wraps RobertaForMaskedLM which has tied weights
(word_embeddings.weight ↔ lm_head.decoder.weight). Safetensors rejects
shared tensors during checkpoint saving. Switch to torch.save via
save_safetensors=False.

Replace torch.distributed with accelerate.PartialState for world-size
detection, and torchrun with accelerate launch for cleaner distributed

training management.ix: resolve safetensors shared tensor error and migrate to accelerate
SplitHeadWrapper wraps RobertaForMaskedLM which has tied weights
(word_embeddings.weight ↔ lm_head.decoder.weight). Safetensors rejects
shared tensors during checkpoint saving. Switch to torch.save via
save_safetensors=False.

Replace torch.distributed with accelerate.PartialState for world-size
detection, and torchrun with accelerate launch for cleaner distributed
training management.
@EYH0602
EYH0602 merged commit fe10288 into main Feb 8, 2026
3 checks passed
@EYH0602
EYH0602 deleted the fix/shared-memory branch February 8, 2026 23:30
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant