Task-oriented finetuning for better embeddings on neural search
Fine-tuning is an effective way to improve performance on neural search tasks. However, setting up and performing fine-tuning can be very time-consuming and resource-intensive.
Jina AI's Finetuner makes fine-tuning easier and faster by streamlining the workflow and handling all the complexity and infrastructure in the cloud. With Finetuner, you can easily enhance the performance of pre-trained models, making them production-ready without extensive labeling or expensive hardware.
🎏 Better embeddings: Create high-quality embeddings for semantic search, visual similarity search, cross-modal text<->image search, recommendation systems, clustering, duplication detection, anomaly detection, or other uses.
⏰ Low budget, high expectations: Bring considerable improvements to model performance, making the most out of as little as a few hundred training samples, and finish fine-tuning in as little as an hour.
📈 Performance promise: Enhance the performance of pre-trained models so that they deliver state-of-the-art performance on domain-specific applications.
🔱 Simple yet powerful: Easy access to 40+ mainstream loss functions, 10+ optimizers, layer pruning, weight freezing, dimensionality reduction, hard-negative mining, cross-modal models, and distributed training.
☁ All-in-cloud: Train using our GPU infrastructure, manage runs, experiments, and artifacts on Jina AI Cloud without worrying about resource availability, complex integration, or infrastructure costs.
| Model | Task | Metric | Pretrained | Finetuned | Delta | Run it! |
|---|---|---|---|---|---|---|
| BERT | Quora Question Answering | mRR | 0.835 | 0.967 | 15.8% | |
| Recall | 0.915 | 0.963 | 5.3% | |||
| ResNet | Visual similarity search on TLL | mAP | 0.110 | 0.196 | 78.2% | |
| Recall | 0.249 | 0.460 | 84.7% | |||
| CLIP | Deep Fashion text-to-image search | mRR | 0.575 | 0.676 | 17.4% | |
| Recall | 0.473 | 0.564 | 19.2% | |||
| M-CLIP | Cross market product recommendation (German) | mRR | 0.430 | 0.648 | 50.7% | |
| Recall | 0.247 | 0.340 | 37.7% | |||
| PointNet++ | ModelNet40 3D Mesh Search | mRR | 0.791 | 0.891 | 12.7% | |
| Recall | 0.154 | 0.242 | 57.1% |
All metrics were evaluated for k@20 after training for 5 epochs using the Adam optimizer with learning rates of 1e-4 for ResNet, 1e-7 for CLIP and 1e-5 for the BERT models, 5e-4 for PointNet++
Make sure you have Python 3.8+ installed. Finetuner can be installed via pip by executing:
pip install -U finetunerIf you want to encode local data with the finetuner.encode function, you need to install
"finetuner[full]". This includes a number of additional dependencies, which are necessary for encoding: Torch,
Torchvision and OpenCLIP:
pip install "finetuner[full]"
⚠️ Starting with version 0.5.0, Finetuner computing is performed on Jina AI Cloud. The last local version is0.4.1. This version is still available for installation viapip. See Finetuner git tags and releases.
Check out our published blogposts and tutorials to see Finetuner in action!
- Fine-tuning with Low Budget and High Expectations
- Hype and Hybrids: Search is more than Keywords and Vectors
- Improving Search Quality for Non-English Queries with Fine-tuned Multilingual CLIP Models
- How Much Do We Get by Finetuning CLIP?
- Use Discussions to talk about your use cases, questions, and support queries.
- Join our Discord community and chat with other community members about ideas.
- Join our Engineering All Hands meet-up to discuss your use case and learn Jina AI new features.
- When? The second Tuesday of every month
- Where? Zoom (see our public events calendar/.ical) and live stream on YouTube
- Subscribe to the latest video tutorials on our YouTube channel
Finetuner is backed by Jina AI and licensed under Apache-2.0.
We are actively hiring AI engineers and solution engineers to build the next generation of open-source AI ecosystems.