pplx-embed-v2-context-9b-preview

Model release Hugging Face 模型 · perplexity-ai Source language: Chinese

官方机构模型仓库创建于 2026-09-26。仓库创建时间不等于正式发布日期。任务类型:feature-extraction

Full text

pplx-embed-v2-context-9b-preview

pplx-embed-v2-context-9b-preview is a contextual embedding model for document chunks in RAG systems. A document is passed as a list of chunks; the chunks are encoded together, so each chunk's embedding reflects its surrounding context, and one embedding is returned per chunk.

[!NOTE] This is a preview release, not a final model. Weights, embeddings, and the interface may change in later versions without backward compatibility, so embeddings produced with this preview should not be mixed with embeddings from a future release.

[!IMPORTANT] Queries and documents are encoded with different methods: use encode_queries for queries and encode for document chunks. The model is trained with separate query and document prefixes, and encoding queries with encode silently degrades retrieval quality.

[!IMPORTANT] Like pplx-embed-context-v1, the model natively produces *unnormalized* int8-quantized embeddings. Compare embeddings with *cosine similarity*, or pass normalize_embeddings=True and use the dot product.

Model

Model Dimensions MRL Quantization Instruction Pooling pplx-embed-v2-context-9b-preview 2048 1024, 2048 INT8 No (fixed query/document prefixes) Mean

Usage

Requires transformers>=5.4.0, torch, numpy, safetensors, and tqdm. The model uses custom code, so load it with trust_remote_code=True.

Options

encode(documents, ...) and encode_queries(queries, ...) accept:

Argument Default Description batch_size 32 Documents (or queries) per forward pass normalize_embeddings False L2-normalize outputs convert_to_numpy True Return NumPy arrays; False returns CPU tensors show_progress_bar False Show a progress bar device None Move the model to this device before encoding

Matryoshka dimensions

The model was trained with Matryoshka losses at 1024 and 2048 dimensions. To use 1024-dimensional embeddings, take the first 1024 values of each unnormalized embedding and normalize afterwards:

Other truncation sizes were not trained.

Sources

← Back to the daily brief