annoy
finetuner
annoy | finetuner | |
---|---|---|
40 | 36 | |
12,712 | 1,427 | |
0.9% | 1.2% | |
5.3 | 5.5 | |
3 months ago | about 2 months ago | |
C++ | Python | |
Apache License 2.0 | Apache License 2.0 |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
annoy
-
Do we think about vector dbs wrong?
The focus on the top 10 in vector search is a product of wanting to prove value over keyword search. Keyword search is going to miss some conceptual matches. You can try to work around that with tokenization and complex queries with all variations but it's not easy.
Vector search isn't all that new a concept. For example, the annoy library (https://github.com/spotify/annoy) has been around since 2014. It was one of the first open source approximate nearest neighbor libraries. Recommendations have always been a good use case for vector similarity.
Recommendations are a natural extension of search and transformers models made building the vectors for natural language possible. To prove the worth of vector search over keyword search, the focus was always on showing how the top N matches include results not possible with keyword search.
In 2023, there has been a shift towards acknowledging keyword search also has value and that a combination of vector + keyword search (aka hybrid search) operates in the sweet spot. Once again this is validated through the same benchmarks which focus on the top 10.
On top of all this, there is also the reality that the vector database space is very crowded and some want to use their performance benchmarks for marketing.
Disclaimer: I am the author of txtai (https://github.com/neuml/txtai), an open source embeddings database
-
Vector Databases 101
If you want to go larger you could still use some simple setup in conjunction with faiss, annoy or hnsw.
- I'm an undergraduate data science intern and trying to run kmodes clustering. Did this elbow method to figure out how many clusters to use, but I don't really see an "elbow". Tips on number of clusters?
-
Calculating document similarity in a special domain
I then use annoy to compare them. Annoy can use different measures for distance, like cosine, euclidean and more
-
Can Parquet file format index string columns?
Yes you can do this for equality predicates if your row groups are sorted . This blog post (that I didn't write) might add more color. You can't do this for any kind of text searching. If you need to do this with file based storage I'd recommend using a vector based text search and utilize a ANN index library like Annoy.
-
[D]: Best nearest neighbour search for high dimensions
If you need large scale (1000+ dimension, millions+ source points, >1000 queries per second) and accept imperfect results / approximate nearest neighbors, then other people have already mentioned some of the best libraries (FAISS, Annoy).
- Billion-Scale Approximate Nearest Neighbor Search [pdf]
-
[R] Unlimiformer: Long-Range Transformers with Unlimited Length Input
Would be possible to further speed up the process with using something like ANNOY? https://github.com/spotify/annoy
-
Faiss: A library for efficient similarity search
I like Faiss but I tried Spotify's annoy[1] for a recent project and was pretty impressed.
Since lots of people don't seem to understand how useful these embedding libraries are here's an example. I built a thing that indexes bouldering and climbing competition videos, then builds an embedding of the climber's body position per frame. I then can automatically match different climbers on the same problem.
It works pretty well. Since the body positions are 3D it works reasonably well across camera angles.
The biggest problem is getting the embedding right. I simplified it a lot above because I actually need to embed the problem shape itself because otherwise it matches too well: you get frames of people in identical positions but on different problems!
[1] https://github.com/spotify/annoy
-
How to find "k" nearest embeddings in a space with a very large number of N embeddings (efficiently)?
If you just want quick in memory search then pynndescent is a decent option: it's easy to install, and easy to get running. Another good option is Annoy; it's just as easy to install and get running with python, but it is a little less performant if you want to do a lot of queries, or get a knn-graph quickly.
finetuner
-
How do you think search will change with technology like ChatGPT, Bing’s new AI search engine and the upcoming Google Bard?
And all of that has something to do with finetuners. It basically fine-tunes AI models for specific use cases. With it can create a custom search experience that is tailored to their specific needs. I also wonder how this is going to be integrated into SEO tools soon since those tools are catered to traditional search engines.
-
Combining multiple lists into one, meaningfully
Combining multiple lists into one is tough, but it's doable if you have the right approach. Fine-tuning GPT-3 might help, but finding enough examples is tough. You could use existing text data or manually label a set of training examples. A finetuner could be help too. It's a platform-agnostic toolkit that can fine-tune pre-trained models and it's customizable to do lots of tasks.
-
speech_recognition not able to convert the full live audio to text. Please help me to fine-tune it.
You can adjust the pause threshold a little longer for pauses between and phrases. You can also use the phrase detection mode, which sets a time limit for the entire phrase instead of ending the transcription prematurely. If your microphone sensitivity is low, you can also try adjusting the energy threshold. If you want, you can use finetuners.
-
Questions about fine-tuned results. Should the completion results be identical to fine-tune examples?
It's possible that completion results may be identical to fine-tuned examples, but not guaranteed. Even with the same prompt, slight variations in output are expected due to the nature of probabilistic language models. You can experiment with different settings and parameters, including those with finetuners like these.
-
How can I create a dataset to refine Whisper AI from old videos with subtitles?
You can try creating your own dataset. Get some audio data that you want, preprocess it, and then create a custom dataset you can use to fine tune. You could use finetuners like these if you want as well.
-
A Guide to Using OpenTelemetry in Jina for Monitoring and Tracing Applications
We derived the dataset by pre-processing the deepfashion dataset using Finetuner. The image label generated by Finetuner is extracted and formatted to produce the text attribute of each product.
-
[D] Looking for an open source Downloadable model to run on my local device.
You can either use Hugging Face Transformers as they have a lot of pre-trained models that you can customize. Or Finetuners like this one: which is a toolkit for fine-tuning multiple models.
-
Improving Search Quality for Non-English Queries with Fine-tuned Multilingual CLIP Models
Very recently, a few non-English and multilingual CLIP models have appeared, using various sources of training data. In this article, we’ll evaluate a multilingual CLIP model’s performance in a language other than English, and show how you can improve it even further using Jina AI’s Finetuner.
-
Is there a way I can feed the gpt3 model database object like tables? I know we can create fine tune model but not sure about the completion part. Please help!
I think you can convert your data into text and fine-tune the model on it. But that might not be the ideal way to go since you kind of base that on the model. Try transfer learning or finetuning with a finetuner.
-
Classification using prompt or fine tuning?
you can try prompt-based classification or fine-tuning with a Finetuner. Prompts work well for simple tasks but fine-tuning may give better results for complex ones. Althouigh it's going to need more resources, but try both and see what works best for you.
What are some alternatives?
faiss - A library for efficient similarity search and clustering of dense vectors.
gpt_index - LlamaIndex (GPT Index) is a project that provides a central interface to connect your LLM's with external data. [Moved to: https://github.com/jerryjliu/llama_index]
hnswlib - Header-only C++/python library for fast approximate nearest neighbors
Jina AI examples - Jina examples and demos to help you get started
implicit - Fast Python Collaborative Filtering for Implicit Feedback Datasets
RWKV-LM - RWKV is an RNN with transformer-level LLM performance. It can be directly trained like a GPT (parallelizable). So it's combining the best of RNN and transformer - great performance, fast inference, saves VRAM, fast training, "infinite" ctx_len, and free sentence embedding.
Milvus - A cloud-native vector database, storage for next generation AI applications
jina - ☁️ Build multimodal AI applications with cloud-native stack
TensorRec - A TensorFlow recommendation algorithm and framework in Python.
Promptify - Prompt Engineering | Prompt Versioning | Use GPT or other prompt based models to get structured output. Join our discord for Prompt-Engineering, LLMs and other latest research
fastFM - fastFM: A Library for Factorization Machines
pysot - SenseTime Research platform for single object tracking, implementing algorithms like SiamRPN and SiamMask.