All issues

Search results

[Bug] MultiheadAttentionLoRA materializes full attention matrix → OOM on 24GB even with LoRA

Status: Open.

#39 In Sompote/SAM3_LoRA;

· ycdev-liu opened on Aug 14, 2026

Related to dataset making

Status: Open.

#38 In Sompote/SAM3_LoRA;

· BobbyZhang073 opened on Aug 14, 2026

About the validation code

Status: Open.

#37 In Sompote/SAM3_LoRA;

· Wangduo428 opened on Jun 28, 2026

Question about the validation strategy(NMS) mentioned in the README

Status: Open.

#36 In Sompote/SAM3_LoRA;

· Creator404 opened on Jun 23, 2026

SAM3 LoRA Fine-tuning: Severe Performance Degradation, Epoch Training Time Explodes After Epoch 1

Status: Open.

#35 In Sompote/SAM3_LoRA;

· qwork2010 opened on May 22, 2026

ValueError: matrix contains invalid numeric entries

Status: Open.

#34 In Sompote/SAM3_LoRA;

· D1fficult-hua opened on May 16, 2026

有相关论文的发布或投稿吗？如有是否可以提供预印版本

Status: Open.

#33 In Sompote/SAM3_LoRA;

· xiaohongri opened on May 7, 2026

Train with categories code imports SAM3Loss

Status: Open.

#32 In Sompote/SAM3_LoRA;

· Randophilus opened on Apr 24, 2026

Why the resolution must be 1008 in train sam3_lora_native.py

Status: Open.

#31 In Sompote/SAM3_LoRA;

· Barsuen opened on Apr 19, 2026

How to merge the Lora weights to the SAM3

Status: Open.

#30 In Sompote/SAM3_LoRA;

· GallonDeng opened on Apr 9, 2026

VRAM requirements for full lora finetuning ?

Status: Open.

#29 In Sompote/SAM3_LoRA;

· mathanprasannakumar opened on Mar 31, 2026

Train with new vocabulary word

Status: Open.

#5 In Sompote/SAM3_LoRA;

· spirosbond opened on Dec 13, 2025