Add table with results in README and overall clarify it #652
Conversation
|
@gabrielilharco looking much cleaner One point of confusion for me, you pushed up a new B/32 to https://huggingface.co/laion/CLIP-ViT-B-32-DataComp.XL-s13B-b90K And there is also this one https://huggingface.co/laion/CLIP-ViT-B-32-256x256-DataComp-s34B-b86K (that I measured to be 72.7) However the former one, 13B samples, looks like it's ~69 in the csv, so that means the highlight table should have 34B samples for the 256x256 B/32 yes? (currently says 13B) |
README.md
Outdated
|
|
||
| This codebase is work in progress, and we invite all to contribute in making it more accessible and useful. In the future, we plan to add support for TPU training and release larger models. We hope this codebase facilitates and promotes further research in contrastive image-text learning. Please submit an issue or send an email if you have any other requests or suggestions. | ||
| Using this codebase, we have trained several models on a variety of data sources and compute budgets, ranging from [small-scale experiments](docs/LOW_ACC.md) to larger runs including models trained on datasets such as | ||
| [LAION-400M](https://arxiv.org/abs/2111.02114), [LAION-2B](https://laion.ai/blog/laion-5b/) and [DataComp-1B](https://arxiv.org/abs/2304.14108). |
There was a problem hiding this comment.
Let's maybe put the laion5b paper here for consistency
| | ViT-B/32 | DataComp-1B | 256px | 13B | 72.8% | | ||
| | ViT-B/16 | DataComp-1B | 224px | 13B | 73.5% | | ||
| | ViT-L/14 | LAION-2B | 224px | 13B | 75.3% | | ||
| | ViT-L/14 | DataComp-1B | 224px | 13B | 79.2% | |
There was a problem hiding this comment.
Maybe let's add H too since it's widely used
|
Nice! I left a few comments |
|
Thanks @rwightman and @rom1504! Added H-14 to the main table and updated the LAION-5B ref. @rwightman good catch re. the B/32, it should indeed be 34B samples seen for that one. Any thoughts on the 0.1 accuracy difference? |
README.md
Outdated
| | ConvNext-XXLarge | LAION-2B | 256px | 34B | 79.5% | | ||
| | ViT-B/32 | DataComp-1B | 256px | 34B | 72.8% | | ||
| | ViT-B/16 | DataComp-1B | 224px | 13B | 73.5% | | ||
| | ViT-L/14 | LAION-2B | 224px | 13B | 75.3% | |
There was a problem hiding this comment.
are you sure that's on 13B samples seen ? I thought that was on 32B. Did you check again the reproducibility paper ?
There was a problem hiding this comment.
indeed, updated
rom1504
changed the title
|
Great, thank you @gabrielilharco ! |
…s#652) * README revamp * update docs * update readme * update readme * update readme * Add results table * update table * update readme * update readme * update readme
No description provided.