mlfoundations/open_clip issues and pull requests

#1032 - How to Convert a dataset to WebDataset Format for Use with OpenCLIP

Issue - State: closed - Opened by zhangyongren 6 days ago

#1031 - AttributeError: 'WebLoader' object has no attribute 'batch_size'

Issue - State: closed - Opened by zhangyongren 6 days ago

#1029 - 'Namespace' object has no attribute 'train_data_upsampling_factors'

Issue - State: closed - Opened by zhangyongren 7 days ago - 2 comments

#1028 - Different performance when pre-training on A100 and H100

Issue - State: closed - Opened by MingliangLiang3 9 days ago - 3 comments

#1027 - Ask for help with an alternative CLIP loss implementation

Issue - State: closed - Opened by Zch0414 12 days ago

#1026 - @rwightman Could you clarify the FLOPs calculation for the EVA models?

Issue - State: closed - Opened by ghost 15 days ago - 3 comments

#1025 - Not able to reproduce the results for Conceptual Captions 3M

Issue - State: closed - Opened by raja-7-c 23 days ago - 2 comments

#1024 - This bug appeared when I was running the test set near completion. My torch version: 2.5.1+cu124 python 3.10,2080Ti*8

Issue - State: closed - Opened by benwuming 25 days ago - 3 comments

#1023 - I ran this command, but he reported this error, I don't know why he also reported this error when the precision is amp

Issue - State: closed - Opened by benwuming 28 days ago - 4 comments

#1021 - RuntimeError: expected scalar type Float but found BFloat16 (ComfyUI)

Issue - State: closed - Opened by Ehsan7104 about 1 month ago

#1020 - How to customize model structures, instead of using existing models?

Issue - State: closed - Opened by Leslie-Jiang-Hamster about 1 month ago - 1 comment

#1018 - Support for Concurrent Data Downloading and Processing

Issue - State: closed - Opened by tandalalam about 1 month ago - 1 comment

#1016 - Training stuck on first epoch

Issue - State: closed - Opened by alexisdrakopoulos about 2 months ago - 12 comments

#1015 - Add clips

Pull Request - State: open - Opened by Yanqing0327 about 2 months ago

#1012 - error when use try to use int8 operations with OpenCLIP.

Issue - State: open - Opened by xiaohoua about 2 months ago - 1 comment

#1011 - [AssertionError] when trying to inference model

Issue - State: closed - Opened by bright-arparwut 2 months ago - 3 comments

#1010 - AttributeError: module 'triton.language' has no attribute 'libdevice', when using method "convert_int8_model_to_inference_mode()"

Issue - State: closed - Opened by bright-arparwut 2 months ago - 1 comment

#1009 - Intermediate checkpoints with ResNet backbone

Issue - State: closed - Opened by ndrsn0208 2 months ago - 1 comment

#1008 - Add CLIPS to open_clip

Pull Request - State: closed - Opened by Yanqing0327 2 months ago - 3 comments

#1007 - Dataset's mean and std

Issue - State: closed - Opened by xiaohoua 2 months ago - 1 comment

#1005 - ZERO-SHOT Classification On SUN397

Issue - State: closed - Opened by xiaohoua 3 months ago

#1003 - Where can I get the LAION-80M dataset?

Issue - State: closed - Opened by XxFChen 3 months ago

#1002 - is it a bug while computing contrastive loss?

Issue - State: closed - Opened by gaofei 3 months ago - 1 comment

#1001 - Fix ViT-SO400M-14-SigLIP context length

Pull Request - State: closed - Opened by shkarupa-alex 3 months ago - 4 comments

#999 - ViT-L-14-336 fine-tuning failed

Issue - State: closed - Opened by zhaozhipeng1997 3 months ago - 1 comment

#998 - Cannot train again on pretrained checkpoint due to change in default `weights_only=True`

Issue - State: closed - Opened by ishaaq 3 months ago - 9 comments

#997 - how to update the position_embedding from embedding to rope position？

Issue - State: closed - Opened by Johnson-yue 3 months ago - 1 comment

#993 - Clarification on using --train-num-samples (lower value) without --dataset-resampled

Issue - State: closed - Opened by fadamsyah 3 months ago - 6 comments

#992 - could you please give me some advice on how to read directory or a lot of tar, thanks!!!

Issue - State: closed - Opened by leo23ui 3 months ago - 1 comment

#991 - I wonder how to extract patch/local features in 768 dim of CoCa for downstream tasks? Should I use the attn_pool (for caption) to get (256,768)?

Issue - State: closed - Opened by Arsiuuu 3 months ago

#988 - Hello, I divided yfcc15m into 16 parts, but dataset download too many files, compression failure , could you please give me some advice on how to solve this!! thanks

Issue - State: closed - Opened by leo23ui 3 months ago

#985 - Fine-tune ViT Models on Higher Resolution Images

Issue - State: closed - Opened by D0miH 3 months ago - 1 comment

#984 - SigLIP attention mask

Issue - State: closed - Opened by chs20 3 months ago - 1 comment

#983 - The model and pretraining parameters do not match.

Issue - State: closed - Opened by q664171689 3 months ago - 2 comments

#981 - Inference speed

Issue - State: closed - Opened by tppqt 4 months ago - 1 comment

#980 - Issue with all_gather

Issue - State: closed - Opened by scopello 4 months ago - 3 comments

#979 - Support using timm optimizers for alternative to adamw default

Pull Request - State: closed - Opened by rwightman 4 months ago

#978 - add missing ViTamin configs

Pull Request - State: closed - Opened by xywei00 4 months ago

#973 - Fine tune arguments to learn new knowledge without forgetting previous

Issue - State: closed - Opened by SutirthaChakraborty 4 months ago - 1 comment

#971 - Experimenting with alternative siglip loss impl for better dist scaling

Pull Request - State: closed - Opened by rwightman 4 months ago - 2 comments

#970 - All default pretrained weights pushed to HF hub

Pull Request - State: closed - Opened by rwightman 4 months ago - 2 comments

#968 - Unable to load model in offline mode using local files

Issue - State: closed - Opened by yarden4998 4 months ago - 5 comments

#967 - Add safeglobals to allow metaclip models to load with weights_only=True

Pull Request - State: closed - Opened by rwightman 4 months ago

#966 - Error loading ViT-L-14-quickgelu (metaclip_fullcc) model with version v2.27.0+

Issue - State: closed - Opened by aivarasbaranauskas 4 months ago - 12 comments

#965 - Device cleanup

Pull Request - State: closed - Opened by rwightman 4 months ago

#964 - Add GLIP (Centered Masking for Language-Image Pre-Training)

Pull Request - State: open - Opened by MingliangLiang3 4 months ago

#962 - How to subdivide the same category, for example, how to distinguish Persian cats, coffee cats, and jingle cats, which are all cat categories?

Issue - State: closed - Opened by watertianyi 4 months ago - 2 comments

#961 - Does open_clip support add_tokens?

Issue - State: closed - Opened by HyelinNAM 4 months ago - 2 comments

#960 - Switch timm import fallback order to favour new paths

Pull Request - State: closed - Opened by rwightman 4 months ago

#959 - Combined merge PR, safetensors load & SigLIP i18n SO400m

Pull Request - State: closed - Opened by rwightman 4 months ago

#958 - Couldnot find MobileCLIP-S0

Issue - State: closed - Opened by SutirthaChakraborty 4 months ago - 1 comment

#957 - Apply scale loss when performing accum_freq

Issue - State: open - Opened by AshStuff 4 months ago - 1 comment

#956 - Add model defs & weights for new so400m i18n variant

Pull Request - State: closed - Opened by rwightman 4 months ago

#954 - RuntimeError: Mask shape should match input shape

Issue - State: closed - Opened by MINXIANWEN 5 months ago - 1 comment

#948 - when I finetune CLIP_ViT_L_14 model , Logit Scale is decrease from 100.0 to 95. and keep going , is right?

Issue - State: closed - Opened by Johnson-yue 5 months ago

#947 - Does the “ViT-L/14” support conversion to TensorRT parameter files?

Issue - State: closed - Opened by gyd-a 5 months ago

#946 - Separately Optimizing CLIP Image and Text Encoders with Different Loss Functions

Issue - State: closed - Opened by omrisuissabrown 5 months ago

#945 - Fine Tune for emotion

Issue - State: closed - Opened by SutirthaChakraborty 5 months ago

#944 - Questions regarding the implementation of SigLIP loss

Issue - State: closed - Opened by binwang777 5 months ago - 1 comment

#942 - SigLip memory consumption increases as we scale number of GPUs

Issue - State: open - Opened by khalidsaifullaah 5 months ago - 5 comments

#941 - Question about fine-tuning

Issue - State: closed - Opened by jimmyparadm 5 months ago - 1 comment

#940 - Question about the SigLipTokenizer

Issue - State: closed - Opened by LuFan31 5 months ago - 2 comments

#939 - Question about the SigLipTokenizer

Issue - State: closed - Opened by LuFan31 5 months ago

#938 - Is it possible to load only the visual encoder or the text encoder?

Issue - State: closed - Opened by zhangyx1998 5 months ago - 2 comments

#937 - RuntimeError: The shape of the 2D attn_mask is torch.Size([77, 77]), but should be (4, 4)

Issue - State: closed - Opened by JaspinXu 5 months ago - 3 comments

#936 - Inconsistent performance on pretrained checkpoints of the same architecture from different sources

Issue - State: closed - Opened by bdevnani3 5 months ago - 1 comment

#935 - Conflict between webdataset implementation and OpenCLIP

Issue - State: closed - Opened by wanghao14 5 months ago - 8 comments

#934 - Load model is Error

Issue - State: closed - Opened by ROC-Star 5 months ago - 2 comments

#933 - Any Plan for direct support of parquet dataset?

Issue - State: closed - Opened by cxxgtxy 6 months ago - 2 comments

#932 - training details for convnext_large_d.laion2B-s26B-b102K-augreg

Issue - State: closed - Opened by mumu-mu 6 months ago - 1 comment

#931 - segmentation fault

Issue - State: closed - Opened by vadim0x60 6 months ago - 2 comments

#930 - ❓ [Question] Can't reproduce imagenet results of RN50 model trained on `pixparse/cc3m-wds`

Issue - State: closed - Opened by clownrat6 6 months ago - 1 comment

#928 - Fix torch.load weights_only FutureWarning

Issue - State: closed - Opened by johnbradley 6 months ago - 2 comments

#922 - How to fine tune open clip?

Issue - State: closed - Opened by capricixhk 7 months ago

#921 - How to persist int8 model?

Issue - State: open - Opened by EdenChen233 7 months ago

#920 - Image preprocessor default when loading checkpoints

Issue - State: closed - Opened by jn2clark 7 months ago - 3 comments

#919 - ViT-B-32::laion2b-s34b-b79k Fix

Pull Request - State: closed - Opened by hem9984 7 months ago

#918 - Error loading state_dict using pre-trained checkpoint with CustomTextCLIP model

Issue - State: closed - Opened by cubense 7 months ago - 2 comments

#917 - Duplicate Class Names in IMAGENET_CLASSNAMES

Issue - State: closed - Opened by raytrun 7 months ago - 1 comment

#916 - RuntimeError: expected scalar type Float but found BFloat16

Issue - State: closed - Opened by NineMeowICT 7 months ago - 2 comments

#915 - [bug] loading saved weights of an open_clip model does give back the same results

Issue - State: closed - Opened by yxchng 7 months ago - 2 comments

#914 - Hello, where can I find the 'vision.cfg' and 'text.cfg' files？

Issue - State: closed - Opened by SuzyH9 7 months ago - 1 comment

#913 - Text Similarity not equal to 1 for identical inputs

Issue - State: closed - Opened by sugary199 7 months ago - 1 comment

#910 - RuntimeError: The shape of the 2D attn_mask is torch.Size([77, 77]), but should be (128, 128).

Issue - State: closed - Opened by KYKong97 8 months ago - 4 comments

#909 - Refactor build / dist to use pyproject.toml

Pull Request - State: closed - Opened by rwightman 8 months ago

#908 - Build refactor, move to pyproject.toml

Pull Request - State: closed - Opened by rwightman 8 months ago

#907 - Fixing flaky inference test

Pull Request - State: closed - Opened by rwightman 8 months ago

#906 - Default Transformer MHA to batch_first, enables fastpath for vit

Pull Request - State: closed - Opened by rwightman 8 months ago

#905 - Fix #898, stopping criteria update. Remove any from logic after testi…

Pull Request - State: closed - Opened by rwightman 8 months ago

#903 - Huge performance drop in zero-shot image classification for all SigLIP models

Issue - State: closed - Opened by shyammarjit 8 months ago

#902 - How do you profile the CLIP models

Issue - State: closed - Opened by X-funbean 8 months ago - 1 comment

#901 - fastpath inference is not supported

Issue - State: closed - Opened by bigtree2020 8 months ago - 2 comments

#900 - Remove explicit sentencepiece / protobuf deps

Pull Request - State: closed - Opened by rwightman 8 months ago

#899 - Help With Training Bottleneck

Issue - State: closed - Opened by ShijianXu 8 months ago - 1 comment

#898 - Coca Image-to-Text, RuntimeError: Boolean value of Tensor with more than one value is ambiguous

Issue - State: closed - Opened by sky-cake 8 months ago - 4 comments

#897 - `coca_ViT-B-32`: `self._encode_text` bug

Issue - State: closed - Opened by haideraltahan 8 months ago - 2 comments

#894 - Best practice with distributed training?

Issue - State: closed - Opened by songtianhui 8 months ago - 1 comment

#893 - Preparing for final MobileCLIP hub locations

Pull Request - State: closed - Opened by rwightman 9 months ago

#889 - Conceptual help on SigLIP + pre-trained CLIP

Issue - State: closed - Opened by miguelalba96 9 months ago

#888 - 1111

Issue - State: closed - Opened by xpdd123 9 months ago

GitHub / mlfoundations/open_clip issues and pull requests