ollama - 💡(How to fix) Fix Request: Add jina-embeddings-v5-text-nano and jina-embeddings-v5-text-small [1 comments, 2 participants]

Fix Plan

To add support for the Jina AI embedding models to Ollama, we will integrate the jina-embeddings-v5-text-nano and jina-embeddings-v5-text-small models.

Steps

• Add the models to the list of supported embeddings in Ollama's configuration.
• Implement a function to load and use the models for embedding generation.
• Update the vector search application to utilize the new embedding models.

Example Code

import torch
from transformers import AutoModel, AutoTokenizer

# Load the models
nano_model = AutoModel.from_pretrained("jinaai/jina-embeddings-v5-text-nano")
small_model = AutoModel.from_pretrained("jinaai/jina-embeddings-v5-text-small")

# Define a function to generate embeddings
def generate_embedding(text, model):
    tokenizer = AutoTokenizer.from_pretrained(model)
    inputs = tokenizer(text, return_tensors="pt")
    with torch.no_grad():
        outputs = model(**inputs)
    return outputs.last_hidden_state[:, 0, :]

# Use the function to generate embeddings
nano_embedding = generate_embedding("Example text", "jinaai/jina-embeddings-v5-text-nano")
small_embedding = generate_embedding("Example text", "jinaai/jina-embeddings-v5-text-small")

Verification

To verify that the fix worked, test the embedding generation function with sample text and check that the output embeddings have the expected shape and values.

Extra Tips

• Ensure that the necessary dependencies, including transformers and torch, are installed and up-to-date.
• Consider adding error handling and logging to the embedding generation function to handle potential issues with model loading or input processing.