Datasetpublic

raynardj/amz-product-title-emb

Embedding on Product Title This was using gemini embedding text-embedding-005 to embed this product title dataset. Mostly using batch job. Each embedding vector has 768 fp16 floats. (usually discount original fp32 to fp16 for gemini embedding model can put little damage on accuracy) You can stream this dataset, so having fast boost start on training script.

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes500downloads

Dataset Card

Embedding on Product Title

This was using gemini embedding text-embedding-005 to embed this product title dataset. Mostly using batch job.

Each embedding vector has 768 fp16 floats. (usually discount original fp32 to fp16 for gemini embedding model can put little damage on accuracy)

You can stream this dataset, so having fast boost start on training script.