🕷️ Crawler Inspector

URL Lookup

Direct Parameter Lookup

Raw Queries and Responses

1. Shard Calculation

Query:

Response:

Calculated Shard: 174 (from laksa002)

2. Crawled Status Check

Query:

curl -X POST \
  'http://laksa174.int.ahrefs:8124/' \
  -H 'Content-Type: text/plain' \
  -H 'X-ClickHouse-Database: crawler3' \
  -H 'Authorization: Basic YXBpOg==' \
  -d 'SELECT getAhrefsURLFromUnparsed(src_unparsed) AS found_url, ifNull(toUnixTimestamp(download_stamp), 0) AS crawl_time, ifNull(toUnixTimestamp(props_url_first_seen), 0) AS first_indexed_time, download_http_code AS http_code, src_unparsed AS src_unparsed, src_root_hash AS src_root_hash, history_drop_reason AS history_drop_reason, meta_title AS meta_title, meta_descriptions AS meta_descriptions, attrs_boilerpipe_text AS attrs_boilerpipe_text, attrs_markdown AS attrs_markdown, attrs_readable_markdown AS attrs_readable_markdown, meta_canonical AS meta_canonical FROM crawler3.page_info_local FINAL PREWHERE (src_root_hash, src_unparsed) IN ((getAhrefsRootHashFromUnparsed(getAhrefsUnparsedNoserviceFromURL(\'https://github.com/openai/CLIP\')), getAhrefsUnparsedNoserviceFromURL(\'https://github.com/openai/CLIP\'))) FORMAT JSONEachRow'

Response:

{"found_url":"https:\/\/github.com\/openai\/CLIP","crawl_time":1775755056,"first_indexed_time":1609881675,"http_code":200,"src_unparsed":"com,github!\/openai\/CLIP s443","src_root_hash":"6325672905007345774","history_drop_reason":null,"meta_title":"GitHub - openai\/CLIP: CLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image · GitHub","meta_descriptions":["CLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image - openai\/CLIP"],"attrs_boilerpipe_text":"[Blog]\n[Paper]\n[Model Card]\n[Colab]\nCLIP (Contrastive Language-Image Pre-Training) is a neural network trained on a variety of (image, text) pairs. It can be instructed in natural language to predict the most relevant text snippet, given an image, without directly optimizing for the task, similarly to the zero-shot capabilities of GPT-2 and 3. We found CLIP matches the performance of the original ResNet50 on ImageNet “zero-shot” without using any of the original 1.28M labeled examples, overcoming several major challenges in computer vision.\nApproach\nUsage\nFirst,\ninstall PyTorch 1.7.1\n(or later) and torchvision, as well as small additional dependencies, and then install this repo as a Python package. On a CUDA GPU machine, the following will do the trick:\n$ conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.0\n$ pip install ftfy regex tqdm\n$ pip install git+https:\/\/github.com\/openai\/CLIP.git\nReplace\ncudatoolkit=11.0\nabove with the appropriate CUDA version on your machine or\ncpuonly\nwhen installing on a machine without a GPU.\nimport\ntorch\nimport\nclip\nfrom\nPIL\nimport\nImage\ndevice\n=\n\"cuda\"\nif\ntorch\n.\ncuda\n.\nis_available\n()\nelse\n\"cpu\"\nmodel\n,\npreprocess\n=\nclip\n.\nload\n(\n\"ViT-B\/32\"\n,\ndevice\n=\ndevice\n)\nimage\n=\npreprocess\n(\nImage\n.\nopen\n(\n\"CLIP.png\"\n)).\nunsqueeze\n(\n0\n).\nto\n(\ndevice\n)\ntext\n=\nclip\n.\ntokenize\n([\n\"a diagram\"\n,\n\"a dog\"\n,\n\"a cat\"\n]).\nto\n(\ndevice\n)\nwith\ntorch\n.\nno_grad\n():\nimage_features\n=\nmodel\n.\nencode_image\n(\nimage\n)\ntext_features\n=\nmodel\n.\nencode_text\n(\ntext\n)\nlogits_per_image\n,\nlogits_per_text\n=\nmodel\n(\nimage\n,\ntext\n)\nprobs\n=\nlogits_per_image\n.\nsoftmax\n(\ndim\n=\n-\n1\n).\ncpu\n().\nnumpy\n()\nprint\n(\n\"Label probs:\"\n,\nprobs\n)\n# prints: [[0.9927937  0.00421068 0.00299572]]\nAPI\nThe CLIP module\nclip\nprovides the following methods:\nclip.available_models()\nReturns the names of the available CLIP models.\nclip.load(name, device=..., jit=False)\nReturns the model and the TorchVision transform needed by the model, specified by the model name returned by\nclip.available_models()\n. It will download the model as necessary. The\nname\nargument can also be a path to a local checkpoint.\nThe device to run the model can be optionally specified, and the default is to use the first CUDA device if there is any, otherwise the CPU. When\njit\nis\nFalse\n, a non-JIT version of the model will be loaded.\nclip.tokenize(text: Union[str, List[str]], context_length=77)\nReturns a LongTensor containing tokenized sequences of given text input(s). This can be used as the input to the model\nThe model returned by\nclip.load()\nsupports the following methods:\nmodel.encode_image(image: Tensor)\nGiven a batch of images, returns the image features encoded by the vision portion of the CLIP model.\nmodel.encode_text(text: Tensor)\nGiven a batch of text tokens, returns the text features encoded by the language portion of the CLIP model.\nmodel(image: Tensor, text: Tensor)\nGiven a batch of images and a batch of text tokens, returns two Tensors, containing the logit scores corresponding to each image and text input. The values are cosine similarities between the corresponding image and text features, times 100.\nMore Examples\nZero-Shot Prediction\nThe code below performs zero-shot prediction using CLIP, as shown in Appendix B in the paper. This example takes an image from the\nCIFAR-100 dataset\n, and predicts the most likely labels among the 100 textual labels from the dataset.\nimport\nos\nimport\nclip\nimport\ntorch\nfrom\ntorchvision\n.\ndatasets\nimport\nCIFAR100\n# Load the model\ndevice\n=\n\"cuda\"\nif\ntorch\n.\ncuda\n.\nis_available\n()\nelse\n\"cpu\"\nmodel\n,\npreprocess\n=\nclip\n.\nload\n(\n'ViT-B\/32'\n,\ndevice\n)\n# Download the dataset\ncifar100\n=\nCIFAR100\n(\nroot\n=\nos\n.\npath\n.\nexpanduser\n(\n\"~\/.cache\"\n),\ndownload\n=\nTrue\n,\ntrain\n=\nFalse\n)\n# Prepare the inputs\nimage\n,\nclass_id\n=\ncifar100\n[\n3637\n]\nimage_input\n=\npreprocess\n(\nimage\n).\nunsqueeze\n(\n0\n).\nto\n(\ndevice\n)\ntext_inputs\n=\ntorch\n.\ncat\n([\nclip\n.\ntokenize\n(\nf\"a photo of a\n{\nc\n}\n\"\n)\nfor\nc\nin\ncifar100\n.\nclasses\n]).\nto\n(\ndevice\n)\n# Calculate features\nwith\ntorch\n.\nno_grad\n():\nimage_features\n=\nmodel\n.\nencode_image\n(\nimage_input\n)\ntext_features\n=\nmodel\n.\nencode_text\n(\ntext_inputs\n)\n# Pick the top 5 most similar labels for the image\nimage_features\n\/=\nimage_features\n.\nnorm\n(\ndim\n=\n-\n1\n,\nkeepdim\n=\nTrue\n)\ntext_features\n\/=\ntext_features\n.\nnorm\n(\ndim\n=\n-\n1\n,\nkeepdim\n=\nTrue\n)\nsimilarity\n=\n(\n100.0\n*\nimage_features\n@\ntext_features\n.\nT\n).\nsoftmax\n(\ndim\n=\n-\n1\n)\nvalues\n,\nindices\n=\nsimilarity\n[\n0\n].\ntopk\n(\n5\n)\n# Print the result\nprint\n(\n\"\n\\n\nTop predictions:\n\\n\n\"\n)\nfor\nvalue\n,\nindex\nin\nzip\n(\nvalues\n,\nindices\n):\nprint\n(\nf\"\n{\ncifar100\n.\nclasses\n[\nindex\n]:>16s\n}\n:\n{\n100\n*\nvalue\n.\nitem\n():.2f\n}\n%\"\n)\nThe output will look like the following (the exact numbers may be slightly different depending on the compute device):\nTop predictions:\n\n           snake: 65.31%\n          turtle: 12.29%\n    sweet_pepper: 3.83%\n          lizard: 1.88%\n       crocodile: 1.75%\nNote that this example uses the\nencode_image()\nand\nencode_text()\nmethods that return the encoded features of given inputs.\nLinear-probe evaluation\nThe example below uses\nscikit-learn\nto perform logistic regression on image features.\nimport\nos\nimport\nclip\nimport\ntorch\nimport\nnumpy\nas\nnp\nfrom\nsklearn\n.\nlinear_model\nimport\nLogisticRegression\nfrom\ntorch\n.\nutils\n.\ndata\nimport\nDataLoader\nfrom\ntorchvision\n.\ndatasets\nimport\nCIFAR100\nfrom\ntqdm\nimport\ntqdm\n# Load the model\ndevice\n=\n\"cuda\"\nif\ntorch\n.\ncuda\n.\nis_available\n()\nelse\n\"cpu\"\nmodel\n,\npreprocess\n=\nclip\n.\nload\n(\n'ViT-B\/32'\n,\ndevice\n)\n# Load the dataset\nroot\n=\nos\n.\npath\n.\nexpanduser\n(\n\"~\/.cache\"\n)\ntrain\n=\nCIFAR100\n(\nroot\n,\ndownload\n=\nTrue\n,\ntrain\n=\nTrue\n,\ntransform\n=\npreprocess\n)\ntest\n=\nCIFAR100\n(\nroot\n,\ndownload\n=\nTrue\n,\ntrain\n=\nFalse\n,\ntransform\n=\npreprocess\n)\ndef\nget_features\n(\ndataset\n):\nall_features\n=\n[]\nall_labels\n=\n[]\nwith\ntorch\n.\nno_grad\n():\nfor\nimages\n,\nlabels\nin\ntqdm\n(\nDataLoader\n(\ndataset\n,\nbatch_size\n=\n100\n)):\nfeatures\n=\nmodel\n.\nencode_image\n(\nimages\n.\nto\n(\ndevice\n))\nall_features\n.\nappend\n(\nfeatures\n)\nall_labels\n.\nappend\n(\nlabels\n)\nreturn\ntorch\n.\ncat\n(\nall_features\n).\ncpu\n().\nnumpy\n(),\ntorch\n.\ncat\n(\nall_labels\n).\ncpu\n().\nnumpy\n()\n# Calculate the image features\ntrain_features\n,\ntrain_labels\n=\nget_features\n(\ntrain\n)\ntest_features\n,\ntest_labels\n=\nget_features\n(\ntest\n)\n# Perform logistic regression\nclassifier\n=\nLogisticRegression\n(\nrandom_state\n=\n0\n,\nC\n=\n0.316\n,\nmax_iter\n=\n1000\n,\nverbose\n=\n1\n)\nclassifier\n.\nfit\n(\ntrain_features\n,\ntrain_labels\n)\n# Evaluate using the logistic regression classifier\npredictions\n=\nclassifier\n.\npredict\n(\ntest_features\n)\naccuracy\n=\nnp\n.\nmean\n((\ntest_labels\n==\npredictions\n).\nastype\n(\nfloat\n))\n*\n100.\nprint\n(\nf\"Accuracy =\n{\naccuracy\n:.3f\n}\n\"\n)\nNote that the\nC\nvalue should be determined via a hyperparameter sweep using a validation split.\nSee Also\nOpenCLIP\n: includes larger and independently trained CLIP models up to ViT-G\/14\nHugging Face implementation of CLIP\n: for easier integration with the HF ecosystem","attrs_markdown":"[Skip to content](https:\/\/github.com\/openai\/CLIP#start-of-content)\n\n## Navigation Menu\nToggle navigation\n\n[Sign in](https:\/\/github.com\/login?return_to=https%3A%2F%2Fgithub.com%2Fopenai%2FCLIP)\n\nAppearance settings\n\n- Platform\n  \n  - AI CODE CREATION\n    \n    - [GitHub CopilotWrite better code with AI](https:\/\/github.com\/features\/copilot)\n    - [GitHub SparkBuild and deploy intelligent apps](https:\/\/github.com\/features\/spark)\n    - [GitHub ModelsManage and compare prompts](https:\/\/github.com\/features\/models)\n    - [MCP RegistryNewIntegrate external tools](https:\/\/github.com\/mcp)\n  - DEVELOPER WORKFLOWS\n    \n    - [ActionsAutomate any workflow](https:\/\/github.com\/features\/actions)\n    - [CodespacesInstant dev environments](https:\/\/github.com\/features\/codespaces)\n    - [IssuesPlan and track work](https:\/\/github.com\/features\/issues)\n    - [Code ReviewManage code changes](https:\/\/github.com\/features\/code-review)\n  - APPLICATION SECURITY\n    \n    - [GitHub Advanced SecurityFind and fix vulnerabilities](https:\/\/github.com\/security\/advanced-security)\n    - [Code securitySecure your code as you build](https:\/\/github.com\/security\/advanced-security\/code-security)\n    - [Secret protectionStop leaks before they start](https:\/\/github.com\/security\/advanced-security\/secret-protection)\n  - EXPLORE\n    \n    - [Why GitHub](https:\/\/github.com\/why-github)\n    - [Documentation](https:\/\/docs.github.com\/)\n    - [Blog](https:\/\/github.blog\/)\n    - [Changelog](https:\/\/github.blog\/changelog)\n    - [Marketplace](https:\/\/github.com\/marketplace)\n  \n  [View all features](https:\/\/github.com\/features)\n- Solutions\n  \n  - BY COMPANY SIZE\n    \n    - [Enterprises](https:\/\/github.com\/enterprise)\n    - [Small and medium teams](https:\/\/github.com\/team)\n    - [Startups](https:\/\/github.com\/enterprise\/startups)\n    - [Nonprofits](https:\/\/github.com\/solutions\/industry\/nonprofits)\n  - BY USE CASE\n    \n    - [App Modernization](https:\/\/github.com\/solutions\/use-case\/app-modernization)\n    - [DevSecOps](https:\/\/github.com\/solutions\/use-case\/devsecops)\n    - [DevOps](https:\/\/github.com\/solutions\/use-case\/devops)\n    - [CI\/CD](https:\/\/github.com\/solutions\/use-case\/ci-cd)\n    - [View all use cases](https:\/\/github.com\/solutions\/use-case)\n  - BY INDUSTRY\n    \n    - [Healthcare](https:\/\/github.com\/solutions\/industry\/healthcare)\n    - [Financial services](https:\/\/github.com\/solutions\/industry\/financial-services)\n    - [Manufacturing](https:\/\/github.com\/solutions\/industry\/manufacturing)\n    - [Government](https:\/\/github.com\/solutions\/industry\/government)\n    - [View all industries](https:\/\/github.com\/solutions\/industry)\n  \n  [View all solutions](https:\/\/github.com\/solutions)\n- Resources\n  \n  - EXPLORE BY TOPIC\n    \n    - [AI](https:\/\/github.com\/resources\/articles?topic=ai)\n    - [Software Development](https:\/\/github.com\/resources\/articles?topic=software-development)\n    - [DevOps](https:\/\/github.com\/resources\/articles?topic=devops)\n    - [Security](https:\/\/github.com\/resources\/articles?topic=security)\n    - [View all topics](https:\/\/github.com\/resources\/articles)\n  - EXPLORE BY TYPE\n    \n    - [Customer stories](https:\/\/github.com\/customer-stories)\n    - [Events & webinars](https:\/\/github.com\/resources\/events)\n    - [Ebooks & reports](https:\/\/github.com\/resources\/whitepapers)\n    - [Business insights](https:\/\/github.com\/solutions\/executive-insights)\n    - [GitHub Skills](https:\/\/skills.github.com\/)\n  - SUPPORT & SERVICES\n    \n    - [Documentation](https:\/\/docs.github.com\/)\n    - [Customer support](https:\/\/support.github.com\/)\n    - [Community forum](https:\/\/github.com\/orgs\/community\/discussions)\n    - [Trust center](https:\/\/github.com\/trust-center)\n    - [Partners](https:\/\/github.com\/partners)\n  \n  [View all resources](https:\/\/github.com\/resources)\n- Open Source\n  \n  - COMMUNITY\n    \n    - [GitHub SponsorsFund open source developers](https:\/\/github.com\/sponsors)\n  - PROGRAMS\n    \n    - [Security Lab](https:\/\/securitylab.github.com\/)\n    - [Maintainer Community](https:\/\/maintainers.github.com\/)\n    - [Accelerator](https:\/\/github.com\/accelerator)\n    - [GitHub Stars](https:\/\/stars.github.com\/)\n    - [Archive Program](https:\/\/archiveprogram.github.com\/)\n  - REPOSITORIES\n    \n    - [Topics](https:\/\/github.com\/topics)\n    - [Trending](https:\/\/github.com\/trending)\n    - [Collections](https:\/\/github.com\/collections)\n- Enterprise\n  \n  - ENTERPRISE SOLUTIONS\n    \n    - [Enterprise platformAI-powered developer platform](https:\/\/github.com\/enterprise)\n  - AVAILABLE ADD-ONS\n    \n    - [GitHub Advanced SecurityEnterprise-grade security features](https:\/\/github.com\/security\/advanced-security)\n    - [Copilot for BusinessEnterprise-grade AI features](https:\/\/github.com\/features\/copilot\/copilot-business)\n    - [Premium SupportEnterprise-grade 24\/7 support](https:\/\/github.com\/premium-support)\n- [Pricing](https:\/\/github.com\/pricing)\n\nSearch or jump to...\n\n# Search code, repositories, users, issues, pull requests...\n[Search syntax tips](https:\/\/docs.github.com\/search-github\/github-code-search\/understanding-github-code-search-syntax)\n\n# Provide feedback\nCancel\n\nSubmit feedback\n\n# Saved searches\n## Use saved searches to filter your results more quickly\n\nCancel\n\nCreate saved search\n\n[Sign in](https:\/\/github.com\/login?return_to=https%3A%2F%2Fgithub.com%2Fopenai%2FCLIP)\n\n[Sign up](https:\/\/github.com\/signup?ref_cta=Sign+up&ref_loc=header+logged+out&ref_page=%2F%3Cuser-name%3E%2F%3Crepo-name%3E&source=header-repo&source_repo=openai%2FCLIP)\n\nAppearance settings\n\nResetting focus\n\nYou signed in with another tab or window. [Reload](https:\/\/github.com\/openai\/CLIP) to refresh your session. You signed out in another tab or window. [Reload](https:\/\/github.com\/openai\/CLIP) to refresh your session. You switched accounts on another tab or window. [Reload](https:\/\/github.com\/openai\/CLIP) to refresh your session.\n\nDismiss alert\n\n[openai](https:\/\/github.com\/openai)   \/  **[CLIP](https:\/\/github.com\/openai\/CLIP)**  Public\n\n- [Notifications](https:\/\/github.com\/login?return_to=%2Fopenai%2FCLIP)\n  You must be signed in to change notification settings\n- [Fork 4k](https:\/\/github.com\/login?return_to=%2Fopenai%2FCLIP)\n- [Star 33.1k](https:\/\/github.com\/login?return_to=%2Fopenai%2FCLIP)\n\n- [Code](https:\/\/github.com\/openai\/CLIP)\n- [Issues 251](https:\/\/github.com\/openai\/CLIP\/issues)\n- [Pull requests 21](https:\/\/github.com\/openai\/CLIP\/pulls)\n- [Actions](https:\/\/github.com\/openai\/CLIP\/actions)\n- [Security and quality 0](https:\/\/github.com\/openai\/CLIP\/security)\n- [Insights](https:\/\/github.com\/openai\/CLIP\/pulse)\n\nAdditional navigation options\n\n- [Code](https:\/\/github.com\/openai\/CLIP)\n- [Issues](https:\/\/github.com\/openai\/CLIP\/issues)\n- [Pull requests](https:\/\/github.com\/openai\/CLIP\/pulls)\n- [Actions](https:\/\/github.com\/openai\/CLIP\/actions)\n- [Security and quality](https:\/\/github.com\/openai\/CLIP\/security)\n- [Insights](https:\/\/github.com\/openai\/CLIP\/pulse)\n\n# openai\/CLIP\nmain\n\n[**4** Branches](https:\/\/github.com\/openai\/CLIP\/branches)\n\n[**0** Tags](https:\/\/github.com\/openai\/CLIP\/tags)\n\nGo to file\n\nCode\n\nOpen more actions menu\n\n## Folders and files\n| Name | Name | Last commit message | Last commit date |\n|---|---|---|---|\n| Latest commit [![hintz-openai](https:\/\/avatars.githubusercontent.com\/u\/201815723?v=4&size=40)](https:\/\/github.com\/hintz-openai)[hintz-openai](https:\/\/github.com\/openai\/CLIP\/commits?author=hintz-openai) [Pin GitHub Actions workflow references (](https:\/\/github.com\/openai\/CLIP\/commit\/d05afc436d78f1c48dc0dbf8e5980a9d471f35f6)[\\#537](https:\/\/github.com\/openai\/CLIP\/pull\/537)[)](https:\/\/github.com\/openai\/CLIP\/commit\/d05afc436d78f1c48dc0dbf8e5980a9d471f35f6) failure Mar 25, 2026 [d05afc4](https:\/\/github.com\/openai\/CLIP\/commit\/d05afc436d78f1c48dc0dbf8e5980a9d471f35f6) · Mar 25, 2026 History [58 Commits](https:\/\/github.com\/openai\/CLIP\/commits\/main\/) Open commit details 58 Commits |   |   |   |\n\n## Repository files navigation\n- [README](https:\/\/github.com\/openai\/CLIP)\n- [MIT license](https:\/\/github.com\/openai\/CLIP)\n\n# CLIP\n[\\[Blog\\]](https:\/\/openai.com\/blog\/clip\/) [\\[Paper\\]](https:\/\/arxiv.org\/abs\/2103.00020) [\\[Model Card\\]](https:\/\/github.com\/openai\/CLIP\/blob\/main\/model-card.md) [\\[Colab\\]](https:\/\/colab.research.google.com\/github\/openai\/clip\/blob\/master\/notebooks\/Interacting_with_CLIP.ipynb)\n\nCLIP (Contrastive Language-Image Pre-Training) is a neural network trained on a variety of (image, text) pairs. It can be instructed in natural language to predict the most relevant text snippet, given an image, without directly optimizing for the task, similarly to the zero-shot capabilities of GPT-2 and 3. We found CLIP matches the performance of the original ResNet50 on ImageNet “zero-shot” without using any of the original 1.28M labeled examples, overcoming several major challenges in computer vision.\n\n## Approach\n[![CLIP](https:\/\/github.com\/openai\/CLIP\/raw\/main\/CLIP.png)](https:\/\/github.com\/openai\/CLIP\/blob\/main\/CLIP.png)\n\n## Usage\nFirst, [install PyTorch 1.7.1](https:\/\/pytorch.org\/get-started\/locally\/) (or later) and torchvision, as well as small additional dependencies, and then install this repo as a Python package. On a CUDA GPU machine, the following will do the trick:\n```\n$ conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.0\n$ pip install ftfy regex tqdm\n$ pip install git+https:\/\/github.com\/openai\/CLIP.git\n```\nReplace `cudatoolkit=11.0` above with the appropriate CUDA version on your machine or `cpuonly` when installing on a machine without a GPU.\n```\nimport torch\nimport clip\nfrom PIL import Image\n\ndevice = \"cuda\" if torch.cuda.is_available() else \"cpu\"\nmodel, preprocess = clip.load(\"ViT-B\/32\", device=device)\n\nimage = preprocess(Image.open(\"CLIP.png\")).unsqueeze(0).to(device)\ntext = clip.tokenize([\"a diagram\", \"a dog\", \"a cat\"]).to(device)\n\nwith torch.no_grad():\n    image_features = model.encode_image(image)\n    text_features = model.encode_text(text)\n    \n    logits_per_image, logits_per_text = model(image, text)\n    probs = logits_per_image.softmax(dim=-1).cpu().numpy()\n\nprint(\"Label probs:\", probs)  # prints: [[0.9927937  0.00421068 0.00299572]]\n```\n## API\nThe CLIP module `clip` provides the following methods:\n\n#### `clip.available_models()`\nReturns the names of the available CLIP models.\n\n#### `clip.load(name, device=..., jit=False)`\nReturns the model and the TorchVision transform needed by the model, specified by the model name returned by `clip.available_models()`. It will download the model as necessary. The `name` argument can also be a path to a local checkpoint.\n\nThe device to run the model can be optionally specified, and the default is to use the first CUDA device if there is any, otherwise the CPU. When `jit` is `False`, a non-JIT version of the model will be loaded.\n\n#### `clip.tokenize(text: Union[str, List[str]], context_length=77)`\nReturns a LongTensor containing tokenized sequences of given text input(s). This can be used as the input to the model\n***\nThe model returned by `clip.load()` supports the following methods:\n\n#### `model.encode_image(image: Tensor)`\nGiven a batch of images, returns the image features encoded by the vision portion of the CLIP model.\n\n#### `model.encode_text(text: Tensor)`\nGiven a batch of text tokens, returns the text features encoded by the language portion of the CLIP model.\n\n#### `model(image: Tensor, text: Tensor)`\nGiven a batch of images and a batch of text tokens, returns two Tensors, containing the logit scores corresponding to each image and text input. The values are cosine similarities between the corresponding image and text features, times 100.\n\n## More Examples\n### Zero-Shot Prediction\nThe code below performs zero-shot prediction using CLIP, as shown in Appendix B in the paper. This example takes an image from the [CIFAR-100 dataset](https:\/\/www.cs.toronto.edu\/~kriz\/cifar.html), and predicts the most likely labels among the 100 textual labels from the dataset.\n```\nimport os\nimport clip\nimport torch\nfrom torchvision.datasets import CIFAR100\n\n# Load the model\ndevice = \"cuda\" if torch.cuda.is_available() else \"cpu\"\nmodel, preprocess = clip.load('ViT-B\/32', device)\n\n# Download the dataset\ncifar100 = CIFAR100(root=os.path.expanduser(\"~\/.cache\"), download=True, train=False)\n\n# Prepare the inputs\nimage, class_id = cifar100[3637]\nimage_input = preprocess(image).unsqueeze(0).to(device)\ntext_inputs = torch.cat([clip.tokenize(f\"a photo of a {c}\") for c in cifar100.classes]).to(device)\n\n# Calculate features\nwith torch.no_grad():\n    image_features = model.encode_image(image_input)\n    text_features = model.encode_text(text_inputs)\n\n# Pick the top 5 most similar labels for the image\nimage_features \/= image_features.norm(dim=-1, keepdim=True)\ntext_features \/= text_features.norm(dim=-1, keepdim=True)\nsimilarity = (100.0 * image_features @ text_features.T).softmax(dim=-1)\nvalues, indices = similarity[0].topk(5)\n\n# Print the result\nprint(\"\\nTop predictions:\\n\")\nfor value, index in zip(values, indices):\n    print(f\"{cifar100.classes[index]:>16s}: {100 * value.item():.2f}%\")\n```\nThe output will look like the following (the exact numbers may be slightly different depending on the compute device):\n```\nTop predictions:\n\n           snake: 65.31%\n          turtle: 12.29%\n    sweet_pepper: 3.83%\n          lizard: 1.88%\n       crocodile: 1.75%\n```\nNote that this example uses the `encode_image()` and `encode_text()` methods that return the encoded features of given inputs.\n\n### Linear-probe evaluation\nThe example below uses [scikit-learn](https:\/\/scikit-learn.org\/) to perform logistic regression on image features.\n```\nimport os\nimport clip\nimport torch\n\nimport numpy as np\nfrom sklearn.linear_model import LogisticRegression\nfrom torch.utils.data import DataLoader\nfrom torchvision.datasets import CIFAR100\nfrom tqdm import tqdm\n\n# Load the model\ndevice = \"cuda\" if torch.cuda.is_available() else \"cpu\"\nmodel, preprocess = clip.load('ViT-B\/32', device)\n\n# Load the dataset\nroot = os.path.expanduser(\"~\/.cache\")\ntrain = CIFAR100(root, download=True, train=True, transform=preprocess)\ntest = CIFAR100(root, download=True, train=False, transform=preprocess)\n\n\ndef get_features(dataset):\n    all_features = []\n    all_labels = []\n    \n    with torch.no_grad():\n        for images, labels in tqdm(DataLoader(dataset, batch_size=100)):\n            features = model.encode_image(images.to(device))\n\n            all_features.append(features)\n            all_labels.append(labels)\n\n    return torch.cat(all_features).cpu().numpy(), torch.cat(all_labels).cpu().numpy()\n\n# Calculate the image features\ntrain_features, train_labels = get_features(train)\ntest_features, test_labels = get_features(test)\n\n# Perform logistic regression\nclassifier = LogisticRegression(random_state=0, C=0.316, max_iter=1000, verbose=1)\nclassifier.fit(train_features, train_labels)\n\n# Evaluate using the logistic regression classifier\npredictions = classifier.predict(test_features)\naccuracy = np.mean((test_labels == predictions).astype(float)) * 100.\nprint(f\"Accuracy = {accuracy:.3f}\")\n```\nNote that the `C` value should be determined via a hyperparameter sweep using a validation split.\n\n## See Also\n- [OpenCLIP](https:\/\/github.com\/mlfoundations\/open_clip): includes larger and independently trained CLIP models up to ViT-G\/14\n- [Hugging Face implementation of CLIP](https:\/\/huggingface.co\/docs\/transformers\/model_doc\/clip): for easier integration with the HF ecosystem\n\n## About\nCLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image\n\n### Topics\n[machine-learning](https:\/\/github.com\/topics\/machine-learning \"Topic: machine-learning\")   [deep-learning](https:\/\/github.com\/topics\/deep-learning \"Topic: deep-learning\")\n\n### Resources\n[Readme](https:\/\/github.com\/openai\/CLIP#readme-ov-file)\n\n### License\n[MIT license](https:\/\/github.com\/openai\/CLIP#MIT-1-ov-file)\n\n### Uh oh\\!\nThere was an error while loading. [Please reload this page](https:\/\/github.com\/openai\/CLIP).\n\n[Activity](https:\/\/github.com\/openai\/CLIP\/activity)\n\n[Custom properties](https:\/\/github.com\/openai\/CLIP\/custom-properties)\n\n### Stars\n[**33\\.1k** stars](https:\/\/github.com\/openai\/CLIP\/stargazers)\n\n### Watchers\n[**327** watching](https:\/\/github.com\/openai\/CLIP\/watchers)\n\n### Forks\n[**4k** forks](https:\/\/github.com\/openai\/CLIP\/forks)\n\n[Report repository](https:\/\/github.com\/contact\/report-content?content_url=https%3A%2F%2Fgithub.com%2Fopenai%2FCLIP&report=openai+%28user%29)\n\n## [Used by 3\\.5k](https:\/\/github.com\/openai\/CLIP\/network\/dependents)\n[![@bcmi](https:\/\/avatars.githubusercontent.com\/u\/16733978?s=64&v=4) ![@LHM056awa](https:\/\/avatars.githubusercontent.com\/u\/153359572?s=64&v=4) ![@rkdonharvic-gif](https:\/\/avatars.githubusercontent.com\/u\/273623336?s=64&v=4) ![@wavey-ai](https:\/\/avatars.githubusercontent.com\/u\/124207121?s=64&v=4) ![@ethosphere-ai](https:\/\/avatars.githubusercontent.com\/u\/177072537?s=64&v=4) ![@robertpelloni](https:\/\/avatars.githubusercontent.com\/u\/673434?s=64&v=4) ![@p2plus](https:\/\/avatars.githubusercontent.com\/u\/100416519?s=64&v=4) ![@corticph](https:\/\/avatars.githubusercontent.com\/u\/16427447?s=64&v=4) + 3,532](https:\/\/github.com\/openai\/CLIP\/network\/dependents)\n\n## [Contributors](https:\/\/github.com\/openai\/CLIP\/graphs\/contributors)\n### Uh oh\\!\nThere was an error while loading. [Please reload this page](https:\/\/github.com\/openai\/CLIP).\n\n## Languages\n- [Jupyter Notebook 99.1%](https:\/\/github.com\/openai\/CLIP\/search?l=jupyter-notebook)\n- [Python 0.9%](https:\/\/github.com\/openai\/CLIP\/search?l=python)\n\n## Footer\n© 2026 GitHub, Inc.\n\n### Footer navigation\n- [Terms](https:\/\/docs.github.com\/site-policy\/github-terms\/github-terms-of-service)\n- [Privacy](https:\/\/docs.github.com\/site-policy\/privacy-policies\/github-privacy-statement)\n- [Security](https:\/\/github.com\/security)\n- [Status](https:\/\/www.githubstatus.com\/)\n- [Community](https:\/\/github.community\/)\n- [Docs](https:\/\/docs.github.com\/)\n- [Contact](https:\/\/support.github.com\/?tags=dotcom-footer)\n- Manage cookies\n- Do not share my personal information\n\nYou can’t perform that action at this time.","attrs_readable_markdown":"[\\[Blog\\]](https:\/\/openai.com\/blog\/clip\/) [\\[Paper\\]](https:\/\/arxiv.org\/abs\/2103.00020) [\\[Model Card\\]](https:\/\/github.com\/openai\/CLIP\/blob\/main\/model-card.md) [\\[Colab\\]](https:\/\/colab.research.google.com\/github\/openai\/clip\/blob\/master\/notebooks\/Interacting_with_CLIP.ipynb)\n\nCLIP (Contrastive Language-Image Pre-Training) is a neural network trained on a variety of (image, text) pairs. It can be instructed in natural language to predict the most relevant text snippet, given an image, without directly optimizing for the task, similarly to the zero-shot capabilities of GPT-2 and 3. We found CLIP matches the performance of the original ResNet50 on ImageNet “zero-shot” without using any of the original 1.28M labeled examples, overcoming several major challenges in computer vision.\n\nApproach\n\n[![CLIP](https:\/\/github.com\/openai\/CLIP\/raw\/main\/CLIP.png)](https:\/\/github.com\/openai\/CLIP\/blob\/main\/CLIP.png)\n\nUsage\n\nFirst, [install PyTorch 1.7.1](https:\/\/pytorch.org\/get-started\/locally\/) (or later) and torchvision, as well as small additional dependencies, and then install this repo as a Python package. On a CUDA GPU machine, the following will do the trick:\n```\n$ conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.0\n$ pip install ftfy regex tqdm\n$ pip install git+https:\/\/github.com\/openai\/CLIP.git\n```\nReplace `cudatoolkit=11.0` above with the appropriate CUDA version on your machine or `cpuonly` when installing on a machine without a GPU.\n```\nimport torch\nimport clip\nfrom PIL import Image\n\ndevice = \"cuda\" if torch.cuda.is_available() else \"cpu\"\nmodel, preprocess = clip.load(\"ViT-B\/32\", device=device)\n\nimage = preprocess(Image.open(\"CLIP.png\")).unsqueeze(0).to(device)\ntext = clip.tokenize([\"a diagram\", \"a dog\", \"a cat\"]).to(device)\n\nwith torch.no_grad():\n    image_features = model.encode_image(image)\n    text_features = model.encode_text(text)\n    \n    logits_per_image, logits_per_text = model(image, text)\n    probs = logits_per_image.softmax(dim=-1).cpu().numpy()\n\nprint(\"Label probs:\", probs)  # prints: [[0.9927937  0.00421068 0.00299572]]\n```\nAPI\n\nThe CLIP module `clip` provides the following methods:\n\n`clip.available_models()`\n\nReturns the names of the available CLIP models.\n\n`clip.load(name, device=..., jit=False)`\n\nReturns the model and the TorchVision transform needed by the model, specified by the model name returned by `clip.available_models()`. It will download the model as necessary. The `name` argument can also be a path to a local checkpoint.\n\nThe device to run the model can be optionally specified, and the default is to use the first CUDA device if there is any, otherwise the CPU. When `jit` is `False`, a non-JIT version of the model will be loaded.\n\n`clip.tokenize(text: Union[str, List[str]], context_length=77)`\n\nReturns a LongTensor containing tokenized sequences of given text input(s). This can be used as the input to the model\n***\nThe model returned by `clip.load()` supports the following methods:\n\n`model.encode_image(image: Tensor)`\n\nGiven a batch of images, returns the image features encoded by the vision portion of the CLIP model.\n\n`model.encode_text(text: Tensor)`\n\nGiven a batch of text tokens, returns the text features encoded by the language portion of the CLIP model.\n\n`model(image: Tensor, text: Tensor)`\n\nGiven a batch of images and a batch of text tokens, returns two Tensors, containing the logit scores corresponding to each image and text input. The values are cosine similarities between the corresponding image and text features, times 100.\n\nMore Examples\n\nZero-Shot Prediction\n\nThe code below performs zero-shot prediction using CLIP, as shown in Appendix B in the paper. This example takes an image from the [CIFAR-100 dataset](https:\/\/www.cs.toronto.edu\/~kriz\/cifar.html), and predicts the most likely labels among the 100 textual labels from the dataset.\n```\nimport os\nimport clip\nimport torch\nfrom torchvision.datasets import CIFAR100\n\n# Load the model\ndevice = \"cuda\" if torch.cuda.is_available() else \"cpu\"\nmodel, preprocess = clip.load('ViT-B\/32', device)\n\n# Download the dataset\ncifar100 = CIFAR100(root=os.path.expanduser(\"~\/.cache\"), download=True, train=False)\n\n# Prepare the inputs\nimage, class_id = cifar100[3637]\nimage_input = preprocess(image).unsqueeze(0).to(device)\ntext_inputs = torch.cat([clip.tokenize(f\"a photo of a {c}\") for c in cifar100.classes]).to(device)\n\n# Calculate features\nwith torch.no_grad():\n    image_features = model.encode_image(image_input)\n    text_features = model.encode_text(text_inputs)\n\n# Pick the top 5 most similar labels for the image\nimage_features \/= image_features.norm(dim=-1, keepdim=True)\ntext_features \/= text_features.norm(dim=-1, keepdim=True)\nsimilarity = (100.0 * image_features @ text_features.T).softmax(dim=-1)\nvalues, indices = similarity[0].topk(5)\n\n# Print the result\nprint(\"\\nTop predictions:\\n\")\nfor value, index in zip(values, indices):\n    print(f\"{cifar100.classes[index]:>16s}: {100 * value.item():.2f}%\")\n```\nThe output will look like the following (the exact numbers may be slightly different depending on the compute device):\n```\nTop predictions:\n\n           snake: 65.31%\n          turtle: 12.29%\n    sweet_pepper: 3.83%\n          lizard: 1.88%\n       crocodile: 1.75%\n```\nNote that this example uses the `encode_image()` and `encode_text()` methods that return the encoded features of given inputs.\n\nLinear-probe evaluation\n\nThe example below uses [scikit-learn](https:\/\/scikit-learn.org\/) to perform logistic regression on image features.\n```\nimport os\nimport clip\nimport torch\n\nimport numpy as np\nfrom sklearn.linear_model import LogisticRegression\nfrom torch.utils.data import DataLoader\nfrom torchvision.datasets import CIFAR100\nfrom tqdm import tqdm\n\n# Load the model\ndevice = \"cuda\" if torch.cuda.is_available() else \"cpu\"\nmodel, preprocess = clip.load('ViT-B\/32', device)\n\n# Load the dataset\nroot = os.path.expanduser(\"~\/.cache\")\ntrain = CIFAR100(root, download=True, train=True, transform=preprocess)\ntest = CIFAR100(root, download=True, train=False, transform=preprocess)\n\n\ndef get_features(dataset):\n    all_features = []\n    all_labels = []\n    \n    with torch.no_grad():\n        for images, labels in tqdm(DataLoader(dataset, batch_size=100)):\n            features = model.encode_image(images.to(device))\n\n            all_features.append(features)\n            all_labels.append(labels)\n\n    return torch.cat(all_features).cpu().numpy(), torch.cat(all_labels).cpu().numpy()\n\n# Calculate the image features\ntrain_features, train_labels = get_features(train)\ntest_features, test_labels = get_features(test)\n\n# Perform logistic regression\nclassifier = LogisticRegression(random_state=0, C=0.316, max_iter=1000, verbose=1)\nclassifier.fit(train_features, train_labels)\n\n# Evaluate using the logistic regression classifier\npredictions = classifier.predict(test_features)\naccuracy = np.mean((test_labels == predictions).astype(float)) * 100.\nprint(f\"Accuracy = {accuracy:.3f}\")\n```\nNote that the `C` value should be determined via a hyperparameter sweep using a validation split.\n\nSee Also\n\n- [OpenCLIP](https:\/\/github.com\/mlfoundations\/open_clip): includes larger and independently trained CLIP models up to ViT-G\/14\n- [Hugging Face implementation of CLIP](https:\/\/huggingface.co\/docs\/transformers\/model_doc\/clip): for easier integration with the HF ecosystem","meta_canonical":null}

3. Robots.txt Check

Query:

Response:

4. Spam/Ban Check

Query:

Response:

5. Seen Status Check

ℹ️ Skipped - page is already crawled

📄

INDEXABLE

✅

CRAWLED

4 days ago

🤖

ROBOTS ALLOWED

Page Info Filters

Filter	Status	Condition	Details
HTTP status	PASS	`download_http_code = 200`	HTTP 200
Age cutoff	PASS	`download_stamp > now() - 6 MONTH`	0.2 months ago (distributed domain, exempt)
History drop	PASS	`isNull(history_drop_reason)`	No drop reason
Spam/ban	PASS	`fh_dont_index != 1 AND ml_spam_score = 0`	ml_spam_score=0
Canonical	PASS	`meta_canonical IS NULL OR = '' OR = src_unparsed`	Not set

Page Details

Property	Value
URL	https://github.com/openai/CLIP
Last Crawled	2026-04-09 17:17:36 (4 days ago)
First Indexed	2021-01-05 21:21:15 (5 years ago)
HTTP Status Code	200
Meta Title	GitHub - openai/CLIP: CLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image · GitHub
Meta Description	CLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image - openai/CLIP
Meta Canonical	null
Boilerpipe Text	[Blog] [Paper] [Model Card] [Colab] CLIP (Contrastive Language-Image Pre-Training) is a neural network trained on a variety of (image, text) pairs. It can be instructed in natural language to predict the most relevant text snippet, given an image, without directly optimizing for the task, similarly to the zero-shot capabilities of GPT-2 and 3. We found CLIP matches the performance of the original ResNet50 on ImageNet “zero-shot” without using any of the original 1.28M labeled examples, overcoming several major challenges in computer vision. Approach Usage First, install PyTorch 1.7.1 (or later) and torchvision, as well as small additional dependencies, and then install this repo as a Python package. On a CUDA GPU machine, the following will do the trick: $ conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.0 $ pip install ftfy regex tqdm $ pip install git+https://github.com/openai/CLIP.git Replace cudatoolkit=11.0 above with the appropriate CUDA version on your machine or cpuonly when installing on a machine without a GPU. import torch import clip from PIL import Image device = "cuda" if torch . cuda . is_available () else "cpu" model , preprocess = clip . load ( "ViT-B/32" , device = device ) image = preprocess ( Image . open ( "CLIP.png" )). unsqueeze ( 0 ). to ( device ) text = clip . tokenize ([ "a diagram" , "a dog" , "a cat" ]). to ( device ) with torch . no_grad (): image_features = model . encode_image ( image ) text_features = model . encode_text ( text ) logits_per_image , logits_per_text = model ( image , text ) probs = logits_per_image . softmax ( dim = - 1 ). cpu (). numpy () print ( "Label probs:" , probs ) # prints: [[0.9927937 0.00421068 0.00299572]] API The CLIP module clip provides the following methods: clip.available_models() Returns the names of the available CLIP models. clip.load(name, device=..., jit=False) Returns the model and the TorchVision transform needed by the model, specified by the model name returned by clip.available_models() . It will download the model as necessary. The name argument can also be a path to a local checkpoint. The device to run the model can be optionally specified, and the default is to use the first CUDA device if there is any, otherwise the CPU. When jit is False , a non-JIT version of the model will be loaded. clip.tokenize(text: Union[str, List[str]], context_length=77) Returns a LongTensor containing tokenized sequences of given text input(s). This can be used as the input to the model The model returned by clip.load() supports the following methods: model.encode_image(image: Tensor) Given a batch of images, returns the image features encoded by the vision portion of the CLIP model. model.encode_text(text: Tensor) Given a batch of text tokens, returns the text features encoded by the language portion of the CLIP model. model(image: Tensor, text: Tensor) Given a batch of images and a batch of text tokens, returns two Tensors, containing the logit scores corresponding to each image and text input. The values are cosine similarities between the corresponding image and text features, times 100. More Examples Zero-Shot Prediction The code below performs zero-shot prediction using CLIP, as shown in Appendix B in the paper. This example takes an image from the CIFAR-100 dataset , and predicts the most likely labels among the 100 textual labels from the dataset. import os import clip import torch from torchvision . datasets import CIFAR100 # Load the model device = "cuda" if torch . cuda . is_available () else "cpu" model , preprocess = clip . load ( 'ViT-B/32' , device ) # Download the dataset cifar100 = CIFAR100 ( root = os . path . expanduser ( "~/.cache" ), download = True , train = False ) # Prepare the inputs image , class_id = cifar100 [ 3637 ] image_input = preprocess ( image ). unsqueeze ( 0 ). to ( device ) text_inputs = torch . cat ([ clip . tokenize ( f"a photo of a { c } " ) for c in cifar100 . classes ]). to ( device ) # Calculate features with torch . no_grad (): image_features = model . encode_image ( image_input ) text_features = model . encode_text ( text_inputs ) # Pick the top 5 most similar labels for the image image_features /= image_features . norm ( dim = - 1 , keepdim = True ) text_features /= text_features . norm ( dim = - 1 , keepdim = True ) similarity = ( 100.0 * image_features @ text_features . T ). softmax ( dim = - 1 ) values , indices = similarity [ 0 ]. topk ( 5 ) # Print the result print ( " \n Top predictions: \n " ) for value , index in zip ( values , indices ): print ( f" { cifar100 . classes [ index ]:>16s } : { 100 * value . item ():.2f } %" ) The output will look like the following (the exact numbers may be slightly different depending on the compute device): Top predictions: snake: 65.31% turtle: 12.29% sweet_pepper: 3.83% lizard: 1.88% crocodile: 1.75% Note that this example uses the encode_image() and encode_text() methods that return the encoded features of given inputs. Linear-probe evaluation The example below uses scikit-learn to perform logistic regression on image features. import os import clip import torch import numpy as np from sklearn . linear_model import LogisticRegression from torch . utils . data import DataLoader from torchvision . datasets import CIFAR100 from tqdm import tqdm # Load the model device = "cuda" if torch . cuda . is_available () else "cpu" model , preprocess = clip . load ( 'ViT-B/32' , device ) # Load the dataset root = os . path . expanduser ( "~/.cache" ) train = CIFAR100 ( root , download = True , train = True , transform = preprocess ) test = CIFAR100 ( root , download = True , train = False , transform = preprocess ) def get_features ( dataset ): all_features = [] all_labels = [] with torch . no_grad (): for images , labels in tqdm ( DataLoader ( dataset , batch_size = 100 )): features = model . encode_image ( images . to ( device )) all_features . append ( features ) all_labels . append ( labels ) return torch . cat ( all_features ). cpu (). numpy (), torch . cat ( all_labels ). cpu (). numpy () # Calculate the image features train_features , train_labels = get_features ( train ) test_features , test_labels = get_features ( test ) # Perform logistic regression classifier = LogisticRegression ( random_state = 0 , C = 0.316 , max_iter = 1000 , verbose = 1 ) classifier . fit ( train_features , train_labels ) # Evaluate using the logistic regression classifier predictions = classifier . predict ( test_features ) accuracy = np . mean (( test_labels == predictions ). astype ( float )) * 100. print ( f"Accuracy = { accuracy :.3f } " ) Note that the C value should be determined via a hyperparameter sweep using a validation split. See Also OpenCLIP : includes larger and independently trained CLIP models up to ViT-G/14 Hugging Face implementation of CLIP : for easier integration with the HF ecosystem
Markdown	[Skip to content](https://github.com/openai/CLIP#start-of-content) ## Navigation Menu Toggle navigation [Sign in](https://github.com/login?return_to=https%3A%2F%2Fgithub.com%2Fopenai%2FCLIP) Appearance settings - Platform - AI CODE CREATION - [GitHub CopilotWrite better code with AI](https://github.com/features/copilot) - [GitHub SparkBuild and deploy intelligent apps](https://github.com/features/spark) - [GitHub ModelsManage and compare prompts](https://github.com/features/models) - [MCP RegistryNewIntegrate external tools](https://github.com/mcp) - DEVELOPER WORKFLOWS - [ActionsAutomate any workflow](https://github.com/features/actions) - [CodespacesInstant dev environments](https://github.com/features/codespaces) - [IssuesPlan and track work](https://github.com/features/issues) - [Code ReviewManage code changes](https://github.com/features/code-review) - APPLICATION SECURITY - [GitHub Advanced SecurityFind and fix vulnerabilities](https://github.com/security/advanced-security) - [Code securitySecure your code as you build](https://github.com/security/advanced-security/code-security) - [Secret protectionStop leaks before they start](https://github.com/security/advanced-security/secret-protection) - EXPLORE - [Why GitHub](https://github.com/why-github) - [Documentation](https://docs.github.com/) - [Blog](https://github.blog/) - [Changelog](https://github.blog/changelog) - [Marketplace](https://github.com/marketplace) [View all features](https://github.com/features) - Solutions - BY COMPANY SIZE - [Enterprises](https://github.com/enterprise) - [Small and medium teams](https://github.com/team) - [Startups](https://github.com/enterprise/startups) - [Nonprofits](https://github.com/solutions/industry/nonprofits) - BY USE CASE - [App Modernization](https://github.com/solutions/use-case/app-modernization) - [DevSecOps](https://github.com/solutions/use-case/devsecops) - [DevOps](https://github.com/solutions/use-case/devops) - [CI/CD](https://github.com/solutions/use-case/ci-cd) - [View all use cases](https://github.com/solutions/use-case) - BY INDUSTRY - [Healthcare](https://github.com/solutions/industry/healthcare) - [Financial services](https://github.com/solutions/industry/financial-services) - [Manufacturing](https://github.com/solutions/industry/manufacturing) - [Government](https://github.com/solutions/industry/government) - [View all industries](https://github.com/solutions/industry) [View all solutions](https://github.com/solutions) - Resources - EXPLORE BY TOPIC - [AI](https://github.com/resources/articles?topic=ai) - [Software Development](https://github.com/resources/articles?topic=software-development) - [DevOps](https://github.com/resources/articles?topic=devops) - [Security](https://github.com/resources/articles?topic=security) - [View all topics](https://github.com/resources/articles) - EXPLORE BY TYPE - [Customer stories](https://github.com/customer-stories) - [Events & webinars](https://github.com/resources/events) - [Ebooks & reports](https://github.com/resources/whitepapers) - [Business insights](https://github.com/solutions/executive-insights) - [GitHub Skills](https://skills.github.com/) - SUPPORT & SERVICES - [Documentation](https://docs.github.com/) - [Customer support](https://support.github.com/) - [Community forum](https://github.com/orgs/community/discussions) - [Trust center](https://github.com/trust-center) - [Partners](https://github.com/partners) [View all resources](https://github.com/resources) - Open Source - COMMUNITY - [GitHub SponsorsFund open source developers](https://github.com/sponsors) - PROGRAMS - [Security Lab](https://securitylab.github.com/) - [Maintainer Community](https://maintainers.github.com/) - [Accelerator](https://github.com/accelerator) - [GitHub Stars](https://stars.github.com/) - [Archive Program](https://archiveprogram.github.com/) - REPOSITORIES - [Topics](https://github.com/topics) - [Trending](https://github.com/trending) - [Collections](https://github.com/collections) - Enterprise - ENTERPRISE SOLUTIONS - [Enterprise platformAI-powered developer platform](https://github.com/enterprise) - AVAILABLE ADD-ONS - [GitHub Advanced SecurityEnterprise-grade security features](https://github.com/security/advanced-security) - [Copilot for BusinessEnterprise-grade AI features](https://github.com/features/copilot/copilot-business) - [Premium SupportEnterprise-grade 24/7 support](https://github.com/premium-support) - [Pricing](https://github.com/pricing) Search or jump to... # Search code, repositories, users, issues, pull requests... [Search syntax tips](https://docs.github.com/search-github/github-code-search/understanding-github-code-search-syntax) # Provide feedback Cancel Submit feedback # Saved searches ## Use saved searches to filter your results more quickly Cancel Create saved search [Sign in](https://github.com/login?return_to=https%3A%2F%2Fgithub.com%2Fopenai%2FCLIP) [Sign up](https://github.com/signup?ref_cta=Sign+up&ref_loc=header+logged+out&ref_page=%2F%3Cuser-name%3E%2F%3Crepo-name%3E&source=header-repo&source_repo=openai%2FCLIP) Appearance settings Resetting focus You signed in with another tab or window. [Reload](https://github.com/openai/CLIP) to refresh your session. You signed out in another tab or window. [Reload](https://github.com/openai/CLIP) to refresh your session. You switched accounts on another tab or window. [Reload](https://github.com/openai/CLIP) to refresh your session. Dismiss alert [openai](https://github.com/openai) / [CLIP](https://github.com/openai/CLIP) Public - [Notifications](https://github.com/login?return_to=%2Fopenai%2FCLIP) You must be signed in to change notification settings - [Fork 4k](https://github.com/login?return_to=%2Fopenai%2FCLIP) - [Star 33.1k](https://github.com/login?return_to=%2Fopenai%2FCLIP) - [Code](https://github.com/openai/CLIP) - [Issues 251](https://github.com/openai/CLIP/issues) - [Pull requests 21](https://github.com/openai/CLIP/pulls) - [Actions](https://github.com/openai/CLIP/actions) - [Security and quality 0](https://github.com/openai/CLIP/security) - [Insights](https://github.com/openai/CLIP/pulse) Additional navigation options - [Code](https://github.com/openai/CLIP) - [Issues](https://github.com/openai/CLIP/issues) - [Pull requests](https://github.com/openai/CLIP/pulls) - [Actions](https://github.com/openai/CLIP/actions) - [Security and quality](https://github.com/openai/CLIP/security) - [Insights](https://github.com/openai/CLIP/pulse) # openai/CLIP main [4 Branches](https://github.com/openai/CLIP/branches) [0 Tags](https://github.com/openai/CLIP/tags) Go to file Code Open more actions menu ## Folders and files \| Name \| Name \| Last commit message \| Last commit date \| \|---\|---\|---\|---\| \| Latest commit [![hintz-openai](https://avatars.githubusercontent.com/u/201815723?v=4&size=40)](https://github.com/hintz-openai)[hintz-openai](https://github.com/openai/CLIP/commits?author=hintz-openai) [Pin GitHub Actions workflow references (](https://github.com/openai/CLIP/commit/d05afc436d78f1c48dc0dbf8e5980a9d471f35f6)[\#537](https://github.com/openai/CLIP/pull/537)[)](https://github.com/openai/CLIP/commit/d05afc436d78f1c48dc0dbf8e5980a9d471f35f6) failure Mar 25, 2026 [d05afc4](https://github.com/openai/CLIP/commit/d05afc436d78f1c48dc0dbf8e5980a9d471f35f6) · Mar 25, 2026 History [58 Commits](https://github.com/openai/CLIP/commits/main/) Open commit details 58 Commits \| \| \| \| ## Repository files navigation - [README](https://github.com/openai/CLIP) - [MIT license](https://github.com/openai/CLIP) # CLIP [\[Blog\]](https://openai.com/blog/clip/) [\[Paper\]](https://arxiv.org/abs/2103.00020) [\[Model Card\]](https://github.com/openai/CLIP/blob/main/model-card.md) [\[Colab\]](https://colab.research.google.com/github/openai/clip/blob/master/notebooks/Interacting_with_CLIP.ipynb) CLIP (Contrastive Language-Image Pre-Training) is a neural network trained on a variety of (image, text) pairs. It can be instructed in natural language to predict the most relevant text snippet, given an image, without directly optimizing for the task, similarly to the zero-shot capabilities of GPT-2 and 3. We found CLIP matches the performance of the original ResNet50 on ImageNet “zero-shot” without using any of the original 1.28M labeled examples, overcoming several major challenges in computer vision. ## Approach [![CLIP](https://github.com/openai/CLIP/raw/main/CLIP.png)](https://github.com/openai/CLIP/blob/main/CLIP.png) ## Usage First, [install PyTorch 1.7.1](https://pytorch.org/get-started/locally/) (or later) and torchvision, as well as small additional dependencies, and then install this repo as a Python package. On a CUDA GPU machine, the following will do the trick: ``` $ conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.0 $ pip install ftfy regex tqdm $ pip install git+https://github.com/openai/CLIP.git ``` Replace `cudatoolkit=11.0` above with the appropriate CUDA version on your machine or `cpuonly` when installing on a machine without a GPU. ``` import torch import clip from PIL import Image device = "cuda" if torch.cuda.is_available() else "cpu" model, preprocess = clip.load("ViT-B/32", device=device) image = preprocess(Image.open("CLIP.png")).unsqueeze(0).to(device) text = clip.tokenize(["a diagram", "a dog", "a cat"]).to(device) with torch.no_grad(): image_features = model.encode_image(image) text_features = model.encode_text(text) logits_per_image, logits_per_text = model(image, text) probs = logits_per_image.softmax(dim=-1).cpu().numpy() print("Label probs:", probs) # prints: [[0.9927937 0.00421068 0.00299572]] ``` ## API The CLIP module `clip` provides the following methods: #### `clip.available_models()` Returns the names of the available CLIP models. #### `clip.load(name, device=..., jit=False)` Returns the model and the TorchVision transform needed by the model, specified by the model name returned by `clip.available_models()`. It will download the model as necessary. The `name` argument can also be a path to a local checkpoint. The device to run the model can be optionally specified, and the default is to use the first CUDA device if there is any, otherwise the CPU. When `jit` is `False`, a non-JIT version of the model will be loaded. #### `clip.tokenize(text: Union[str, List[str]], context_length=77)` Returns a LongTensor containing tokenized sequences of given text input(s). This can be used as the input to the model *** The model returned by `clip.load()` supports the following methods: #### `model.encode_image(image: Tensor)` Given a batch of images, returns the image features encoded by the vision portion of the CLIP model. #### `model.encode_text(text: Tensor)` Given a batch of text tokens, returns the text features encoded by the language portion of the CLIP model. #### `model(image: Tensor, text: Tensor)` Given a batch of images and a batch of text tokens, returns two Tensors, containing the logit scores corresponding to each image and text input. The values are cosine similarities between the corresponding image and text features, times 100. ## More Examples ### Zero-Shot Prediction The code below performs zero-shot prediction using CLIP, as shown in Appendix B in the paper. This example takes an image from the [CIFAR-100 dataset](https://www.cs.toronto.edu/~kriz/cifar.html), and predicts the most likely labels among the 100 textual labels from the dataset. ``` import os import clip import torch from torchvision.datasets import CIFAR100 # Load the model device = "cuda" if torch.cuda.is_available() else "cpu" model, preprocess = clip.load('ViT-B/32', device) # Download the dataset cifar100 = CIFAR100(root=os.path.expanduser("~/.cache"), download=True, train=False) # Prepare the inputs image, class_id = cifar100[3637] image_input = preprocess(image).unsqueeze(0).to(device) text_inputs = torch.cat([clip.tokenize(f"a photo of a {c}") for c in cifar100.classes]).to(device) # Calculate features with torch.no_grad(): image_features = model.encode_image(image_input) text_features = model.encode_text(text_inputs) # Pick the top 5 most similar labels for the image image_features /= image_features.norm(dim=-1, keepdim=True) text_features /= text_features.norm(dim=-1, keepdim=True) similarity = (100.0 * image_features @ text_features.T).softmax(dim=-1) values, indices = similarity[0].topk(5) # Print the result print("\nTop predictions:\n") for value, index in zip(values, indices): print(f"{cifar100.classes[index]:>16s}: {100 * value.item():.2f}%") ``` The output will look like the following (the exact numbers may be slightly different depending on the compute device): ``` Top predictions: snake: 65.31% turtle: 12.29% sweet_pepper: 3.83% lizard: 1.88% crocodile: 1.75% ``` Note that this example uses the `encode_image()` and `encode_text()` methods that return the encoded features of given inputs. ### Linear-probe evaluation The example below uses [scikit-learn](https://scikit-learn.org/) to perform logistic regression on image features. ``` import os import clip import torch import numpy as np from sklearn.linear_model import LogisticRegression from torch.utils.data import DataLoader from torchvision.datasets import CIFAR100 from tqdm import tqdm # Load the model device = "cuda" if torch.cuda.is_available() else "cpu" model, preprocess = clip.load('ViT-B/32', device) # Load the dataset root = os.path.expanduser("~/.cache") train = CIFAR100(root, download=True, train=True, transform=preprocess) test = CIFAR100(root, download=True, train=False, transform=preprocess) def get_features(dataset): all_features = [] all_labels = [] with torch.no_grad(): for images, labels in tqdm(DataLoader(dataset, batch_size=100)): features = model.encode_image(images.to(device)) all_features.append(features) all_labels.append(labels) return torch.cat(all_features).cpu().numpy(), torch.cat(all_labels).cpu().numpy() # Calculate the image features train_features, train_labels = get_features(train) test_features, test_labels = get_features(test) # Perform logistic regression classifier = LogisticRegression(random_state=0, C=0.316, max_iter=1000, verbose=1) classifier.fit(train_features, train_labels) # Evaluate using the logistic regression classifier predictions = classifier.predict(test_features) accuracy = np.mean((test_labels == predictions).astype(float)) * 100. print(f"Accuracy = {accuracy:.3f}") ``` Note that the `C` value should be determined via a hyperparameter sweep using a validation split. ## See Also - [OpenCLIP](https://github.com/mlfoundations/open_clip): includes larger and independently trained CLIP models up to ViT-G/14 - [Hugging Face implementation of CLIP](https://huggingface.co/docs/transformers/model_doc/clip): for easier integration with the HF ecosystem ## About CLIP (Contrastive Language-Image Pretraining), Predict the most relevant text snippet given an image ### Topics [machine-learning](https://github.com/topics/machine-learning "Topic: machine-learning") [deep-learning](https://github.com/topics/deep-learning "Topic: deep-learning") ### Resources [Readme](https://github.com/openai/CLIP#readme-ov-file) ### License [MIT license](https://github.com/openai/CLIP#MIT-1-ov-file) ### Uh oh\! There was an error while loading. [Please reload this page](https://github.com/openai/CLIP). [Activity](https://github.com/openai/CLIP/activity) [Custom properties](https://github.com/openai/CLIP/custom-properties) ### Stars [33\.1k stars](https://github.com/openai/CLIP/stargazers) ### Watchers [327 watching](https://github.com/openai/CLIP/watchers) ### Forks [4k forks](https://github.com/openai/CLIP/forks) [Report repository](https://github.com/contact/report-content?content_url=https%3A%2F%2Fgithub.com%2Fopenai%2FCLIP&report=openai+%28user%29) ## [Used by 3\.5k](https://github.com/openai/CLIP/network/dependents) [![@bcmi](https://avatars.githubusercontent.com/u/16733978?s=64&v=4) ![@LHM056awa](https://avatars.githubusercontent.com/u/153359572?s=64&v=4) ![@rkdonharvic-gif](https://avatars.githubusercontent.com/u/273623336?s=64&v=4) ![@wavey-ai](https://avatars.githubusercontent.com/u/124207121?s=64&v=4) ![@ethosphere-ai](https://avatars.githubusercontent.com/u/177072537?s=64&v=4) ![@robertpelloni](https://avatars.githubusercontent.com/u/673434?s=64&v=4) ![@p2plus](https://avatars.githubusercontent.com/u/100416519?s=64&v=4) ![@corticph](https://avatars.githubusercontent.com/u/16427447?s=64&v=4) + 3,532](https://github.com/openai/CLIP/network/dependents) ## [Contributors](https://github.com/openai/CLIP/graphs/contributors) ### Uh oh\! There was an error while loading. [Please reload this page](https://github.com/openai/CLIP). ## Languages - [Jupyter Notebook 99.1%](https://github.com/openai/CLIP/search?l=jupyter-notebook) - [Python 0.9%](https://github.com/openai/CLIP/search?l=python) ## Footer © 2026 GitHub, Inc. ### Footer navigation - [Terms](https://docs.github.com/site-policy/github-terms/github-terms-of-service) - [Privacy](https://docs.github.com/site-policy/privacy-policies/github-privacy-statement) - [Security](https://github.com/security) - [Status](https://www.githubstatus.com/) - [Community](https://github.community/) - [Docs](https://docs.github.com/) - [Contact](https://support.github.com/?tags=dotcom-footer) - Manage cookies - Do not share my personal information You can’t perform that action at this time.
Readable Markdown	[\[Blog\]](https://openai.com/blog/clip/) [\[Paper\]](https://arxiv.org/abs/2103.00020) [\[Model Card\]](https://github.com/openai/CLIP/blob/main/model-card.md) [\[Colab\]](https://colab.research.google.com/github/openai/clip/blob/master/notebooks/Interacting_with_CLIP.ipynb) CLIP (Contrastive Language-Image Pre-Training) is a neural network trained on a variety of (image, text) pairs. It can be instructed in natural language to predict the most relevant text snippet, given an image, without directly optimizing for the task, similarly to the zero-shot capabilities of GPT-2 and 3. We found CLIP matches the performance of the original ResNet50 on ImageNet “zero-shot” without using any of the original 1.28M labeled examples, overcoming several major challenges in computer vision. Approach [![CLIP](https://github.com/openai/CLIP/raw/main/CLIP.png)](https://github.com/openai/CLIP/blob/main/CLIP.png) Usage First, [install PyTorch 1.7.1](https://pytorch.org/get-started/locally/) (or later) and torchvision, as well as small additional dependencies, and then install this repo as a Python package. On a CUDA GPU machine, the following will do the trick: ``` $ conda install --yes -c pytorch pytorch=1.7.1 torchvision cudatoolkit=11.0 $ pip install ftfy regex tqdm $ pip install git+https://github.com/openai/CLIP.git ``` Replace `cudatoolkit=11.0` above with the appropriate CUDA version on your machine or `cpuonly` when installing on a machine without a GPU. ``` import torch import clip from PIL import Image device = "cuda" if torch.cuda.is_available() else "cpu" model, preprocess = clip.load("ViT-B/32", device=device) image = preprocess(Image.open("CLIP.png")).unsqueeze(0).to(device) text = clip.tokenize(["a diagram", "a dog", "a cat"]).to(device) with torch.no_grad(): image_features = model.encode_image(image) text_features = model.encode_text(text) logits_per_image, logits_per_text = model(image, text) probs = logits_per_image.softmax(dim=-1).cpu().numpy() print("Label probs:", probs) # prints: [[0.9927937 0.00421068 0.00299572]] ``` API The CLIP module `clip` provides the following methods: `clip.available_models()` Returns the names of the available CLIP models. `clip.load(name, device=..., jit=False)` Returns the model and the TorchVision transform needed by the model, specified by the model name returned by `clip.available_models()`. It will download the model as necessary. The `name` argument can also be a path to a local checkpoint. The device to run the model can be optionally specified, and the default is to use the first CUDA device if there is any, otherwise the CPU. When `jit` is `False`, a non-JIT version of the model will be loaded. `clip.tokenize(text: Union[str, List[str]], context_length=77)` Returns a LongTensor containing tokenized sequences of given text input(s). This can be used as the input to the model *** The model returned by `clip.load()` supports the following methods: `model.encode_image(image: Tensor)` Given a batch of images, returns the image features encoded by the vision portion of the CLIP model. `model.encode_text(text: Tensor)` Given a batch of text tokens, returns the text features encoded by the language portion of the CLIP model. `model(image: Tensor, text: Tensor)` Given a batch of images and a batch of text tokens, returns two Tensors, containing the logit scores corresponding to each image and text input. The values are cosine similarities between the corresponding image and text features, times 100. More Examples Zero-Shot Prediction The code below performs zero-shot prediction using CLIP, as shown in Appendix B in the paper. This example takes an image from the [CIFAR-100 dataset](https://www.cs.toronto.edu/~kriz/cifar.html), and predicts the most likely labels among the 100 textual labels from the dataset. ``` import os import clip import torch from torchvision.datasets import CIFAR100 # Load the model device = "cuda" if torch.cuda.is_available() else "cpu" model, preprocess = clip.load('ViT-B/32', device) # Download the dataset cifar100 = CIFAR100(root=os.path.expanduser("~/.cache"), download=True, train=False) # Prepare the inputs image, class_id = cifar100[3637] image_input = preprocess(image).unsqueeze(0).to(device) text_inputs = torch.cat([clip.tokenize(f"a photo of a {c}") for c in cifar100.classes]).to(device) # Calculate features with torch.no_grad(): image_features = model.encode_image(image_input) text_features = model.encode_text(text_inputs) # Pick the top 5 most similar labels for the image image_features /= image_features.norm(dim=-1, keepdim=True) text_features /= text_features.norm(dim=-1, keepdim=True) similarity = (100.0 * image_features @ text_features.T).softmax(dim=-1) values, indices = similarity[0].topk(5) # Print the result print("\nTop predictions:\n") for value, index in zip(values, indices): print(f"{cifar100.classes[index]:>16s}: {100 * value.item():.2f}%") ``` The output will look like the following (the exact numbers may be slightly different depending on the compute device): ``` Top predictions: snake: 65.31% turtle: 12.29% sweet_pepper: 3.83% lizard: 1.88% crocodile: 1.75% ``` Note that this example uses the `encode_image()` and `encode_text()` methods that return the encoded features of given inputs. Linear-probe evaluation The example below uses [scikit-learn](https://scikit-learn.org/) to perform logistic regression on image features. ``` import os import clip import torch import numpy as np from sklearn.linear_model import LogisticRegression from torch.utils.data import DataLoader from torchvision.datasets import CIFAR100 from tqdm import tqdm # Load the model device = "cuda" if torch.cuda.is_available() else "cpu" model, preprocess = clip.load('ViT-B/32', device) # Load the dataset root = os.path.expanduser("~/.cache") train = CIFAR100(root, download=True, train=True, transform=preprocess) test = CIFAR100(root, download=True, train=False, transform=preprocess) def get_features(dataset): all_features = [] all_labels = [] with torch.no_grad(): for images, labels in tqdm(DataLoader(dataset, batch_size=100)): features = model.encode_image(images.to(device)) all_features.append(features) all_labels.append(labels) return torch.cat(all_features).cpu().numpy(), torch.cat(all_labels).cpu().numpy() # Calculate the image features train_features, train_labels = get_features(train) test_features, test_labels = get_features(test) # Perform logistic regression classifier = LogisticRegression(random_state=0, C=0.316, max_iter=1000, verbose=1) classifier.fit(train_features, train_labels) # Evaluate using the logistic regression classifier predictions = classifier.predict(test_features) accuracy = np.mean((test_labels == predictions).astype(float)) * 100. print(f"Accuracy = {accuracy:.3f}") ``` Note that the `C` value should be determined via a hyperparameter sweep using a validation split. See Also - [OpenCLIP](https://github.com/mlfoundations/open_clip): includes larger and independently trained CLIP models up to ViT-G/14 - [Hugging Face implementation of CLIP](https://huggingface.co/docs/transformers/model_doc/clip): for easier integration with the HF ecosystem
Shard	174 (laksa)
Root Hash	6325672905007345774
Unparsed URL	com,github!/openai/CLIP s443