SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2103.00020 · 2021

Learning Transferable Visual Models From Natural Language Supervision

Alec Radford, Aditya Ramesh, Gretchen Krueger, Pamela Mishkin, Ilya Sutskever, Sandhini Agarwal, Amanda Askell, Jack Clark, Girish Sastry, Jong Kim, Gabriel Goh, Chris Hallacy

arXiv · PDF · Open in the Atlas

Code that ran

16 of 20 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 14 failures of our instrument; 4 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2103.00020.json · badge: /api/ran/2103.00020.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
copy not recorded community
not in the paper's declared repositories
7 ran · 3 no recorded run
YvanG/VQGAN-CLIP extension (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
filipbasara0/simple-clip reimplementation (classifier label)
classifier basis: classifier_2025
2 ran · 0 no recorded run
NYU-DICE-Lab/open_clip reimplementation (classifier label)
classifier basis: classifier_2025
1 ran · 1 no recorded run
moein-shariatnia/OpenAI-CLIP reimplementation (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
mainaksingha01/odg-clip extension (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
michi-3000/eyeclip extension (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
FunctionStatusWhere it lives
accuracy Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
this paper's copy was not recorded; identical code first harvested from KishoreP1/DetailCLIP
pointer only; the stored copy get_code serves, from KishoreP1/DetailCLIP, is pointer only (licence: NONE) · get_code("2e59707d16d21473")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

avg_entropy Ran
ran; our contract draft was wrong, not the code
this paper's copy was not recorded; identical code first harvested from azshue/TPT
pointer only · get_code("441ae80dd4f616f3")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

contrastive_loss Ran
ran; our contract draft was wrong, not the code
filipbasara0/simple-clip/simple_clip/clip.py
code served (permissive licence) · get_code("4ea4d130ce4030a1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function expects logits to be a square matrix where:
# - logits.size(0) gives the batch size
# - targets will be torch.arange(logits.size(0))
# - It computes cross_entropy on logits and logits.t() (transpose)
# For this to work properly, logits should be a square matrix (batch_size x batch_size)

batch_size = 2
logits = torch.randn(batch_size, batch_size)

ARGS = {
    'logits': logits
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
convert_to_custom_text_state_dict Ran
ran, violated the contract we drafted
this paper's copy was not recorded; identical code first harvested from deep-real/dcp
pointer only; the stored copy get_code serves, from deep-real/dcp, is pointer only (licence: NONE) · get_code("f19962ebb134b3d7")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

cross_entropy Ran
ran; our contract draft was wrong, not the code
moein-shariatnia/OpenAI-CLIP/CLIP.py
code served (permissive licence) · get_code("63e6e57b587c3aa8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned Tensor, ndim 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create predictions tensor (batch_size=2, num_classes=4)
# These are logits (pre-softmax values)
preds = torch.randn(2, 4)

# Create targets tensor (batch_size=2, num_classes=4)
# These should be probability distributions (sum to 1 along dim 1)
targets = torch.softmax(torch.randn(2, 4), dim=1)

# reduction parameter
reduction = 'none'

ARGS = {
    'preds': preds,
    'targets': targets,
    'reduction': reduction
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (2,) — shape (2,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
domain_text_loss Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mainaksingha01/odg-clip/pacs.py
code served (permissive licence) · get_code("7f39fa83f26c97c7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch.nn import functional as F

torch.manual_seed(0)

# The function expects:
# - diff_textfeatures: a tensor that can be indexed by integers (batch dimension)
# - domain: a list/tensor that can be indexed and compared for equality

# Looking at the function:
# - It iterates through indices 0 to len(domain)-2
# - It compares domain[i] with domain[i+1]
# - When they differ, it computes MSE loss between diff_textfeatures[i] and diff_textfeatures[i+1]
# - diff_textfeatures[i] should be tensors suitable for F.mse_loss

# Create diff_textfeatures as a batch of feature vectors
# Shape: (batch_size, feature_dim)
batch_size = 5
feature_dim = 4
diff_textfeatures = torch.randn(batch_size, feature_dim)

# Create domain as a list/tensor with some transitions
# We need at least 2 elements, and some should differ to trigger loss computation
domain = [0, 0, 1, 1, 2]  # Has transitions at indices 1->2 and 3->4

ARGS = {
    'diff_textfeatures': diff_textfeatures,
    'domain': domain
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_cast_dtype Ran
ran; our contract draft was wrong, not the code
this paper's copy was not recorded; identical code first harvested from zsxkib/PuLID
pointer only · get_code("dcd422d66b0581d8")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_input_dtype Ran
ran; our contract draft was wrong, not the code
this paper's copy was not recorded; identical code first harvested from deep-real/dcp
pointer only; the stored copy get_code serves, from deep-real/dcp, is pointer only (licence: NONE) · get_code("b476c8cfbf0f1f47")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

lanczos Ran
ran; our contract draft was wrong, not the code
YvanG/VQGAN-CLIP/CLIP_VQGAN.py
pointer only (licence: NONE) · get_code("1c4b58a58ab65b12")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, YvanG/VQGAN-CLIP when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

mean_per_class Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
this paper's copy was not recorded; identical code first harvested from KishoreP1/DetailCLIP
pointer only; the stored copy get_code serves, from KishoreP1/DetailCLIP, is pointer only (licence: NONE) · get_code("f0eb8193f86d363b")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

misc_measures Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
michi-3000/eyeclip/zero_shot.py
pointer only (licence: NONE) · get_code("1b17b45e4d751423")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, michi-3000/eyeclip when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ramp Ran
ran, honoured the contract we drafted
YvanG/VQGAN-CLIP/CLIP_VQGAN.py
pointer only (licence: NONE) · get_code("8c580a6820115461")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, YvanG/VQGAN-CLIP when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

select_confident_samples Ran
ran; our contract draft was wrong, not the code
this paper's copy was not recorded; identical code first harvested from azshue/TPT
pointer only · get_code("3ce4d11e58de02ad")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

siglip_loss Ran
ran; our contract draft was wrong, not the code
filipbasara0/simple-clip/simple_clip/clip.py
code served (permissive licence) · get_code("67c4ef5004396e02")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# siglip_loss expects logits of shape (n, n) where n is the batch size
# The function creates labels as 2 * eye(n) - 1, so logits should be square
n = 2
logits = torch.randn(n, n)

ARGS = {
    'logits': logits
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
sinc Ran
ran; our contract draft was wrong, not the code
YvanG/VQGAN-CLIP/CLIP_VQGAN.py
pointer only (licence: NONE) · get_code("86a64cd8e1771cc5")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, YvanG/VQGAN-CLIP when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

trace_model Ran
ran; our contract draft was wrong, not the code
NYU-DICE-Lab/open_clip/src/open_clip/model.py
pointer only (licence: NOASSERTION) · get_code("8e95157106d1df8c")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NYU-DICE-Lab/open_clip (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

build_model_from_openai_state_dict Not yet run
unverified (harvested, no recorded run)
NYU-DICE-Lab/open_clip/src/open_clip/model.py
pointer only (licence: NOASSERTION) · get_code("903b104e9c495329")
How it ran

The body of this sample is withheld: GitHub could not classify the licence of this paper's repository, NYU-DICE-Lab/open_clip (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

has_ddp_wrapper Not yet run
unverified (harvested, no recorded run)
this paper's copy was not recorded; identical code first harvested from wenquanlu/noisy_dinov2
pointer only · get_code("1cd2728a84872357")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

remove_ddp_wrapper Not yet run
unverified (harvested, no recorded run)
this paper's copy was not recorded; identical code first harvested from wenquanlu/noisy_dinov2
pointer only · get_code("f565e9c6316d8c64")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

test_time_tuning Not yet run
unverified (harvested, no recorded run)
this paper's copy was not recorded; identical code first harvested from azshue/TPT
pointer only · get_code("631c70c26c0fcedf")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Also present in other papers' code

Of 20 samples lifted from this paper's linked repositories, 20 carry a function identity and 0 do not. Of the 20 distinct functions with an identity, 17 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 0 the same month, 17 later) and 3 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
get_cast_dtype 66 other papers
66 later
2210.08402 (2022-10)
LAION-5B: An open large-scale dataset for training next generation image-text models
2212.07143 (2022-12)
Reproducible scaling laws for contrastive language-image learning
2301.07094 (2023-01)
Learning Customized Visual Models with Retrieval-Augmented Knowledge
and 63 more not named here
convert_to_custom_text_state_dict 49 other papers
49 later
2210.08402 (2022-10)
LAION-5B: An open large-scale dataset for training next generation image-text models
2212.07143 (2022-12)
Reproducible scaling laws for contrastive language-image learning
2301.07094 (2023-01)
Learning Customized Visual Models with Retrieval-Augmented Knowledge
and 46 more not named here
get_input_dtype 29 other papers
29 later
2210.08402 (2022-10)
LAION-5B: An open large-scale dataset for training next generation image-text models
2212.07143 (2022-12)
Reproducible scaling laws for contrastive language-image learning
2312.12856 (2023-12)
SkyScript: A Large and Semantically Diverse Vision-Language Dataset for Remote Sensing
and 26 more not named here
accuracy 13 other papers
3 dated earlier, 10 later
1706.02677 (2017-06)
Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
1911.00068 (2019-11)
Confident Learning: Estimating Uncertainty in Dataset Labels
2011.09157 (2020-11)
Dense Contrastive Learning for Self-Supervised Visual Pre-Training
and 10 more not named here
select_confident_samples 8 other papers
8 later · 1 only as an assembly brick
2209.07511 (2022-09)
Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models
2305.18010 (2023-05)
Test-Time Adaptation with CLIP Reward for Zero-Shot Generalization in Vision-Language Models
2401.15657 (2024-01)
Data-Free Generalized Zero-Shot Learning
and 5 more not named here
trace_model 8 other papers
8 later
2210.04150 (2022-10)
Open-Vocabulary Semantic Segmentation with Mask-adapted CLIP
2302.10281 (2023-02)
LiT Tuned Models for Efficient Species Detection
2303.07240 (2023-03)
PMC-CLIP: Contrastive Language-Image Pre-training using Biomedical Documents
and 5 more not named here
avg_entropy 7 other papers
7 later · 2 only as an assembly brick
2209.07511 (2022-09)
Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models
2305.18010 (2023-05)
Test-Time Adaptation with CLIP Reward for Zero-Shot Generalization in Vision-Language Models
2401.15657 (2024-01)
Data-Free Generalized Zero-Shot Learning
and 4 more not named here
cross_entropy 6 other papers
6 later
2203.03897 (2022-03)
Geodesic Multi-Modal Mixup for Robust Fine-Tuning
2306.01859 (2023-06)
Spatially Resolved Gene Expression Prediction from H&E Histology Images via Bi-modal Contrastive Learning
2404.04538 (2024-04)
Soft-Prompting with Graph-of-Thought for Multi-modal Representation Learning
and 3 more not named here
has_ddp_wrapper 5 other papers
5 later
2309.16588 (2023-09)
Vision Transformers Need Registers
2407.10366 (2024-07)
Accessing Vision Foundation Models via ImageNet-1K
2505.12191 (2025-05)
Ditch the Denoiser: Emergence of Noise Robustness in Self-Supervised Learning from Data Curriculum
and 2 more not named here
mean_per_class 5 other papers
5 later
2112.12750 (2021-12)
SLIP: Self-supervision meets Language-Image Pre-training
136860514 (2022 (year only))
SLIP: Self-supervision meets Language-Image Pre-training
and 2 more not named here
remove_ddp_wrapper 4 other papers
4 later
2309.16588 (2023-09)
Vision Transformers Need Registers
2407.10366 (2024-07)
Accessing Vision Foundation Models via ImageNet-1K
2505.12191 (2025-05)
Ditch the Denoiser: Emergence of Noise Robustness in Self-Supervised Learning from Data Curriculum
and 1 more not named here
lanczos 2 other papers
1 dated earlier, 1 later
1809.11096 (2018-09)
Large Scale GAN Training for High Fidelity Natural Image Synthesis
2204.08583 (2022-04)
VQGAN-CLIP: Open Domain Image Generation and Editing with Natural Language Guidance
ramp 2 other papers
1 dated earlier, 1 later
1809.11096 (2018-09)
Large Scale GAN Training for High Fidelity Natural Image Synthesis
2204.08583 (2022-04)
VQGAN-CLIP: Open Domain Image Generation and Editing with Natural Language Guidance
sinc 2 other papers
1 dated earlier, 1 later
1809.11096 (2018-09)
Large Scale GAN Training for High Fidelity Natural Image Synthesis
2204.08583 (2022-04)
VQGAN-CLIP: Open Domain Image Generation and Editing with Natural Language Guidance
test_time_tuning 2 other papers
2 later
2209.07511 (2022-09)
Test-Time Prompt Tuning for Zero-Shot Generalization in Vision-Language Models
2412.09706 (2024-12)
Diffusion-Enhanced Test-time Adaptation with Text and Image Augmentation
build_model_from_openai_state_dict 1 other paper
1 later
2302.10281 (2023-02)
LiT Tuned Models for Efficient Species Detection
siglip_loss 1 other paper
1 later
2303.15343 (2023-03)
Sigmoid Loss for Language Image Pre-Training

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

More are linked than shown here.

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

State-of-the-art computer vision systems are trained to predict a fixed set of predetermined object categories. This restricted form of supervision limits their generality and usability since additional labeled data is needed to specify any other visual concept. Learning directly from raw text about images is a promising alternative which leverages a much broader source of supervision. We demonstrate that the simple pre-training task of predicting which caption goes with which image is an efficient and scalable way to learn SOTA image representations from scratch on a dataset of 400 million (image, text) pairs collected from the internet. After pre-training, natural language is used to reference learned visual concepts (or describe new ones) enabling zero-shot transfer of the model to downstream tasks. We study the performance of this approach by benchmarking on over 30 different existing computer vision datasets, spanning tasks such as OCR, action recognition in videos, geo-localization, and many types of fine-grained object classification. The model transfers non-trivially to most tasks and is often competitive with a fully supervised baseline without the need for any dataset specific training. For instance, we match the accuracy of the original ResNet-50 on ImageNet zero-shot without needing to use any of the 1.28 million training examples it was trained on. We release our code and pre-trained model weights at https://github.com/OpenAI/CLIP.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2103.00020")
get_code_for_paper("2103.00020")
have("2103.00020")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2103.00020.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2103.00020.svg)](https://syntology.ai/paper/2103.00020)

Connect an agent — have() is free.