pplx-embed-context-v1-0.6b-GGUF

!!!use this custom llama.cpp version to verify bitperfect compatibility with the original model: https://github.com/hellc/llama.cpp/commits/master/

argus-ai/pplx-embed-context-v1-0.6b-GGUF

This repository contains GGUF format quantized files for Perplexity's pplx-embed-context-v1-0.6b model. The YAML metadata at the top of this file ensures correct categorization as an embedding model on the Hugging Face Hub and within compatible local software like LM Studio.

Model Overview

pplx-embed-context-v1-0.6b is a 0.6 billion parameter text embedding model optimized for web-scale retrieval tasks. It is built on a diffusion-pretrained Qwen3 backbone using multi-stage contrastive learning.

Unlike autoregressive models, it uses bidirectional attention to capture comprehensive context. This specific "context" variant is designed specifically for encoding passages with respect to document-level context.

Key Specifications

• Architecture: Diffusion-pretrained Qwen3 (Bidirectional Attention)
• Parameters: 0.6 Billion
• Dimensions: 1024
• Context Window: 32,000 tokens
• Pooling: Mean
• Instruction Prefix: None required. You can embed the text directly without prompt engineering.

Intended Use

• Use this model for embedding document chunks in Retrieval-Augmented Generation (RAG) systems where the surrounding document context is important for accurately capturing the meaning of the chunk.
• Do not use this model for independent search queries or standalone sentences. For those tasks, the standard pplx-embed-v1-0.6b model is recommended.

Usage with Local Software

Note: Because this repository uses the feature-extraction pipeline tag, local inference servers (i.e. LM Studio) will recognize it as an embedding model rather than a text generation model.