It would be nice to do a full-scale run with a large number of embeddings. We can try a larger subset of the coyo embeddings with Qwen: s3://qdrant--vectorforge/llava-onevision-1.5-mid-training-85m/qwen3-vl-embedding-2b/coyo/000/
its 1.4M embeddings:
nova inspect s3://qdrant--vectorforge/llava-onevision-1.5-mid-training-85m/qwen3-vl-embedding-2b/coyo/000/
path: s3://qdrant--vectorforge/llava-onevision-1.5-mid-training-85m/qwen3-vl-embedding-2b/coyo/000/**/*.parquet
files: 74
rows: 1,480,000
schema:
id VARCHAR
image STRUCT(bytes BLOB, path INTEGER)
caption VARCHAR
qwen_multimodal_embedding FLOAT[] (dim 2048)
It would be nice to do a full-scale run with a large number of embeddings. We can try a larger subset of the
coyoembeddings with Qwen:s3://qdrant--vectorforge/llava-onevision-1.5-mid-training-85m/qwen3-vl-embedding-2b/coyo/000/its 1.4M embeddings: