Perplexitypplx-embed v1
pplx-embed-v1-late-0.6b
This is a token-level retrieval adaptation of pplx-embed-v1-0.6b.
What the release provides
Continued training optimizes MaxSim similarity. Each token receives a 128-dimensional vector. The published model uses MIT terms.pplx-embed-v1-late-0.6b official model card
How to use it well
Practical guidance. Budget index size for token vectors. Compare retrieval precision after matching query tokens to document tokens, rather than averaging the vectors into one embedding.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | perplexity-ai/pplx-embed-v1-late-0.6b |