skip to content
gigarouter gigarouter
models / embeddings · coming soon

Ivysaur

Mihaiii/Ivysaur

published Apr 2024 · updated Apr 2024

Ivysaur is an embedding model fine-tuned on QA pairs for semantic-autocomplete.

est. price
~$0.008
/ 1M tokens · estimated, set at launch
API providers
0
downloads / mo
232
license
mit

specs

TaskText Embedding
ArchitectureBERT
Parameters22.7M
Embedding Dimension384
Max Sequence Length512 tokens

about this model

Ivysaur is an embedding model fine-tuned from gte-tiny that produces 384-dimensional sentence embeddings, designed specifically for semantic-autocomplete applications.

The base model, gte-tiny (22.7M parameters), is distilled from gte-small and uses mean pooling with a maximum sequence length of 512 tokens. It follows the BERT architecture and offers performance comparable to its teacher at roughly half the model size. Ivysaur is further fine-tuned on the qa-assistant dataset (7.17k English question-answer pairs with relevance scores) using contrastive learning, which improves its ability to rank semantically similar query-response pairs for autocomplete suggestions.

Performance benchmarks

The teacher model gte-small achieves the following MTEB scores across 56 tasks (Ivysaur inherits similar quality with slightly reduced size):

TaskScore
Clustering44.89
Pair Classification83.54
Reranking57.70
Retrieval49.46
STS82.07
Summarization30.42
Classification72.31
Average61.36

Limitations

This model exclusively processes English text and truncates inputs longer than 512 tokens. It is released under the MIT license.

best for

FAQ

What is Ivysaur best used for?

It is designed for semantic-autocomplete, where it suggests completions based on meaning rather than exact match.

How does Ivysaur compare to gte-tiny?

Ivysaur is a fine-tune of gte-tiny on the qa-assistant dataset, improving performance on QA-like tasks while retaining the same 22.7M parameter size and 384-dimensional embeddings.

What input format does the model expect?

It accepts English text strings; long texts are truncated to 512 tokens. Use the sentence-transformers or HuggingFace Transformers library to generate embeddings.

How do I call Ivysaur via the gigarouter API?

Use the OpenAI-compatible endpoint with your API key and specify the model name "Mihaiii/Ivysaur" in your request.

not yet live

We're benchmarking and onboarding Ivysaur as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.

related embeddings models

compare all →