4 Issues opened by 4 people

• [Question] Fintune llava-v1.5-7b just use all textvqa data, but get Accuracy: 0.00%

  #1789 opened Dec 4, 2024

• [Usage] finetune_task_lora.sh -> Error "exits with return code = -7"

  #1788 opened Dec 3, 2024

• [Question] Hello, I am currently fine-tuning the visual encoder using LLaMA2 + CLIP-14-336 with the finetune_full_schedule.sh script. However, I noticed that the loss does not decrease and fluctuates significantly. What could be the reason for this?

  #1787 opened Dec 3, 2024

• [Question] demo website error https://llava.hliu.cc/

  #1786 opened Nov 27, 2024

12 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• How to fine tune Lora without images……

  #1585 commented on Nov 27, 2024 • 0 new comments

• [Question] Text only training?

  #1467 commented on Nov 27, 2024 • 0 new comments

• Need either a `state_dict` or a `save_folder` containing offloaded weights.

  #1600 commented on Nov 28, 2024 • 0 new comments

• [Usage] Infer on model finetuned using finetune_qlora.sh

  #1771 commented on Nov 29, 2024 • 0 new comments

• [Question] How to use in-context learning in LLaVA?

  #1357 commented on Nov 29, 2024 • 0 new comments

• Some images of ocr vqa data in llava_v1_5_mix665k.json do not exist!

  #1618 commented on Dec 1, 2024 • 0 new comments

• [Usage] finetune_task_lora.sh checkpoints usage

  #1423 commented on Dec 2, 2024 • 0 new comments

• [Usage] can not print weight in callback function

  #1751 commented on Dec 2, 2024 • 0 new comments

• Can I load the parameters of llava1.5 with full parameter fine-tuning

  #1694 commented on Dec 2, 2024 • 0 new comments

• [Question] In-Context-Learning for Batch Inference 上下文学习怎么批量推理？

  #1315 commented on Dec 2, 2024 • 0 new comments

• [Usage] 7B Inferece CUDA Out of Memory for RTX 4090 24GB VRAM

  #1570 commented on Dec 2, 2024 • 0 new comments

• NETWORK ERROR DUE TO HIGH TRAFFIC. PLEASE REGENERATE OR REFRESH THIS PAGE. (error_code: 1)

  #1332 commented on Dec 2, 2024 • 0 new comments