# Breaking the Limits of Open-Weight CLIP: An Optimization Framework for Self-supervised Fine-tuning of CLIP

arXiv: https://arxiv.org/abs/2601.09859 · HTML: https://mlsanity.com/paper/2601.09859/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 49 of its 63 references that are in the map.
- First posted on arXiv 14 Jan 2026.

## Builds on (most cited first, 30 of 49)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [Sigmoid Loss for Language Image Pre-Training](https://mlsanity.com/paper/2303.15343/) (2023)
- [Learning to Prompt for Vision-Language Models](https://mlsanity.com/paper/2109.01134/) (2021)
- [Identity Mappings in Deep Residual Networks](https://mlsanity.com/paper/1603.05027/) (2016)
- [Conditional Prompt Learning for Vision-Language Models](https://mlsanity.com/paper/2203.05557/) (2022)
- [CoCa: Contrastive Captioners are Image-Text Foundation Models](https://mlsanity.com/paper/2205.01917/) (2022)
- [Conceptual 12M: Pushing Web-Scale Image-Text Pre-Training To Recognize Long-Tail Visual Concepts](https://mlsanity.com/paper/2102.08981/) (2021)
- [Reproducible scaling laws for contrastive language-image learning](https://mlsanity.com/paper/2212.07143/) (2022)
- [EVA-CLIP: Improved Training Techniques for CLIP at Scale](https://mlsanity.com/paper/2303.15389/) (2023)
- [Robust fine-tuning of zero-shot models](https://mlsanity.com/paper/2109.01903/) (2021)
- [Fine-Tuning can Distort Pretrained Features and Underperform Out-of-Distribution](https://mlsanity.com/paper/2202.10054/) (2022)
- [ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders](https://mlsanity.com/paper/2301.00808/) (2023)
- [MaPLe: Multi-modal Prompt Learning](https://mlsanity.com/paper/2210.03117/) (2022)
- [DataComp: In search of the next generation of multimodal datasets](https://mlsanity.com/paper/2304.14108/) (2023)
- [How Much Can CLIP Benefit Vision-and-Language Tasks?](https://mlsanity.com/paper/2107.06383/) (2021)
- [Scaling Language-Image Pre-training via Masking](https://mlsanity.com/paper/2212.00794/) (2022)
- [On the Stability of Fine-tuning BERT: Misconceptions, Explanations, and Strong Baselines](https://mlsanity.com/paper/2006.04884/) (2020)
- [Data Filtering Networks](https://mlsanity.com/paper/2309.17425/) (2023)
- [GestureDiffuCLIP: Gesture Diffusion Model with CLIP Latents](https://mlsanity.com/paper/2303.14613/) (2023)
- [Finetune like you pretrain: Improved finetuning of zero-shot vision models](https://mlsanity.com/paper/2212.00638/) (2022)
- [Show and Tell: Lessons learned from the 2015 MSCOCO Image Captioning Challenge](https://mlsanity.com/paper/1609.06647/) (2016)
- [Stable and low-precision training for large-scale vision-language models](https://mlsanity.com/paper/2304.13013/) (2023)
- [Don't Stop Learning: Towards Continual Learning for the CLIP Model](https://mlsanity.com/paper/2207.09248/) (2022)
- [GeoDE: a Geographically Diverse Evaluation Dataset for Object Recognition](https://mlsanity.com/paper/2301.02560/) (2023)
- [Provable Stochastic Optimization for Global Contrastive Learning: Small Batch Does Not Harm Performance](https://mlsanity.com/paper/2202.12387/) (2022)
- [MedCLIP-SAMv2: Towards Universal Text-Driven Medical Image Segmentation](https://mlsanity.com/paper/2409.19483/) (2024)
- [CLIP Itself is a Strong Fine-tuner: Achieving 85.7% and 88.0% Top-1 Accuracy with ViT-B and ViT-L on ImageNet](https://mlsanity.com/paper/2212.06138/) (2022)

## Built on by (most cited first, 1 of 1)

- [Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation](https://mlsanity.com/paper/2606.06712/) (2026)

## Abstract (arXiv)

CLIP has become a cornerstone of multimodal representation learning, yet improving its performance typically requires a prohibitively costly process of training from scratch on billions of samples. We ask a different question: Can we improve the performance of open-weight CLIP models across various downstream tasks using only existing self-supervised datasets? Unlike supervised fine-tuning, which adapts a pretrained model to a single downstream task, our setting seeks to improve general performance across various tasks. However, as both our experiments and prior studies reveal, simply applying standard training protocols starting from an open-weight CLIP model often fails, leading to performance degradation. In this paper, we introduce TuneCLIP, a self-supervised fine-tuning framework that overcomes the performance degradation. TuneCLIP has two key components: (1) a warm-up stage of recovering optimization statistics to reduce cold-start bias, inspired by theoretical analysis, and (2) a fine-tuning stage of optimizing a new contrastive loss to mitigate the penalization on false negative pairs. Our extensive experiments show that TuneCLIP consistently improves performance across model architectures and scales. Notably, it elevates leading open-weight models like SigLIP (ViT-B/16), achieving gains of up to +2.5% on ImageNet and related out-of-distribution benchmarks, and +1.2% on the highly competitive DataComp benchmark, setting a new strong baseline for efficient post-pretraining adaptation.