dvlab-research/LLaMA-VID

Issues

• About ZERO3

  #75 opened 2 months ago by xxtars
  7

• training loss in stage-1

  #88 opened a month ago by Nastu-Ho
  0

• code details

  #87 opened a month ago by Nastu-Ho
  0

• Extract context relevancy

  #86 opened 2 months ago by IgnacioSan22
  0

• KeyError: 'LlavaConfig'

  #85 opened 2 months ago by skyol99
  0

• How to resume the checkpoint to continue pretraining？

  #84 opened 2 months ago by Einstone-rose
  0

• About the WebVid dataset

  #83 opened 2 months ago by szbcasia
  0

• why not use LoRA for tunning Vicuna?

  #72 opened 2 months ago by dragen1860
  1

• Confusion in pre-process images for long video

  #77 opened 2 months ago by zhuqiangLu
  0

• Are all video-based checkpoints trained with 2 tokens?

  #82 opened 2 months ago by haodi19
  0

• HF model format : vlm weights not in llama-vid-7b-full-336

  #81 opened 2 months ago by nileshkokane01
  0

• Zero-3 offload support

  #60 opened 4 months ago by XenonLamb
  5

• About the json in stage2 and stage3

  #79 opened 2 months ago by liziming5353
  1

• Questions about Text Decoder and Text Query

  #80 opened 2 months ago by SeuXiao
  0

• about the context length for long video

  #78 opened 2 months ago by zhuqiangLu
  0

• RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cuda:1! (when checking argument for argument weight in method wrapper_CUDA__native_layer_norm)

  #76 opened 2 months ago by daocodedao
  1

• An error occurs during the stage 2 fine-tuning

  #74 opened 3 months ago by ShuoZhang2003
  1

• AttributeError: 'NoneType' object has no attribute 'is_loaded'

  #73 opened 3 months ago by sykuann
  1

• Multi-image inference

  #71 opened 3 months ago by g-h-chen
  1

• Requirements needed for inferring llama-vid llama-vid-13b-full-224-video-fps-1

  #69 opened 3 months ago by sykuann
  1

• abnormal outputs for llama-vid-7b-full-224-video-fps-1 ckpt

  #68 opened 3 months ago by YulongBonjour
  1

• error: llava key

  #64 opened 4 months ago by menahem-borges-rodrigues
  1

• Sharing training loss

  #59 opened 4 months ago by Deaddawn
  2

• Computation costs for each stage?

  #70 opened 3 months ago by Becomebright
  1

• How to change default path for model_zoo

  #67 opened 3 months ago by sykuann
  2

• Questions about the subtitles.

  #66 opened 3 months ago by Yxxxb
  1

• Long video dataset (only available 167 movies)

  #62 opened 3 months ago by KerolosAtef
  2

• Long Video dataset

  #61 opened 4 months ago by eslambakr
  1

• flash-attn

  #65 opened 4 months ago by ismailukman
  2

• About evaluation on vqav2 dataset

  #63 opened 4 months ago by liziming5353
  1

• Incomplete evaluation on MSVD-QA dataset.

  #52 opened 5 months ago by XenonLamb
  5

• About text encoder

  #51 opened 4 months ago by liziming5353
  3

• MSVD ACC decrease after stage3

  #58 opened 4 months ago by Deaddawn
  3

• 自定义长视频完全跑不了

  #54 opened 4 months ago by TotoroDHL
  1

• Logic error in code: img_in_text and img_token not in sentence["value"]

  #50 opened 5 months ago by dragen1860
  3

• The GPU's graphics card usage is also constantly increasing,

  #57 opened 5 months ago by kunkunsheng
  3

• is eva_vit_g.pth trained by yourself?

  #56 opened 5 months ago by Deaddawn
  1

• why stage 1 and 2 use differenct ` --version plain_guided ` ` --version imgsp_v1 ` parameters?

  #55 opened 5 months ago by dragen1860
  1

• Enquiry on Download Permission

  #53 opened 5 months ago by HenryHZY
  2

• A question in stage3

  #45 opened 5 months ago by liziming5353
  2

• two types of tokenizer?

  #43 opened 5 months ago by dragen1860
  3

• multiple json for training?

  #39 opened 5 months ago by dragen1860
  1

• When can the Customed Long Video Gradio Web UI be released？

  #40 opened 5 months ago by QiSu77
  4

• Long Video CLI wrong

  #48 opened 5 months ago by QiSu77
  2

• is the LLM weight trainable during stage1-2-3?

  #49 opened 5 months ago by dragen1860
  1

• why delay_load in build_vision_tower(config, delay_load=True)?

  #47 opened 5 months ago by dragen1860
  1

• why different architectures in stage2 and stage3?

  #46 opened 5 months ago by dragen1860
  1

• stage 2: freezing the visual encoder?

  #44 opened 5 months ago by dragen1860
  1

• you build `build_vision_tower` twice?

  #42 opened 5 months ago by dragen1860
  1

• what does `lazy_preprocess` mean?

  #41 opened 5 months ago by dragen1860
  1