Jina Code Embeddings 0.5B
Jina Code Embeddings 0.5B converts source code and technical text into search vectors. It supports text-to-code, code-to-code, code-to-text, code-to-completion and technical question-answer retrieval across more than 15 programming languages.
How the model works
Jina adapts Qwen2.5-Coder-0.5B for embeddings, using task-specific instruction prefixes and last-token pooling. The model accepts up to 32,768 tokens. Its default 896-dimensional output can be reduced to 64 dimensions through Matryoshka training. These are retrieval representations; the release does not generate a code completion merely because completion retrieval is one of its tasks.Jina Code Embeddings 0.5B model details
Use and access
The smaller 0.5B release offers a separate trained size from the 1.5B model. Its model card documents Transformers, Sentence Transformers and vLLM use. GGUF and MLX copies change the runtime format, not the task identity. The weights use CC BY-NC 4.0, with Jina also listing hosted access.Jina Code Embeddings 0.5B model detailsJina Code Embeddings 0.5B technical overview
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | API, Downloadable |
| Model identifiers | jina-code-embeddings-0.5bjinaai/jina-code-embeddings-0.5bjina-code-embeddings-0.5b-GGUFjina-code-embeddings-0.5b-mlx |
| Context window | 32,768 tokens Maximum input sequence in the model card; serving limits can be lower. Jina Code Embeddings 0.5B model details |
| Parameters | 0.5 billionJina Code Embeddings 0.5B model details |
| Embedding dimensions | 896; reducible to 64Jina Code Embeddings 0.5B model details |
| Weight license | CC BY-NC 4.0Jina Code Embeddings 0.5B model details |
| Base model | Qwen2.5-Coder-0.5BJina Code Embeddings 0.5B model details |
Related models
- Jina Code Embeddings 1.5B
Jina AI · Jina Code Embeddings
A larger code-search encoder with adjustable vectors for technical retrieval.
Launch date API · Downloadable - Jina Embeddings v2 Base Code
Jina AI · Jina Embeddings v2
A compact code-search encoder trained on source and documentation pairs.
Launch date Downloadable