Hugging Face's logo

perplexity-ai
/
pplx-embed-v1-0.6b

Feature Extraction
sentence-transformers
ONNX
Safetensors
multilingual
bidirectional_pplx_qwen3
sentence-similarity
mteb
custom_code
text-embeddings-inference
Model card Files
xet
 Community
13
bowang0911 commited on
Commit
fed698c
·
verified ·
1 Parent(s): 9408ebf

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +3 -1
README.md CHANGED
@@ -13,12 +13,14 @@ pipeline_tag: feature-extraction
13
  ## Models
14
 
15
  | Model | Dimensions | Context | MRL | Quantization | Instruction |
16
- |-------|------------|---------|-----|--------------|-------------|
17
  | `pplx-embed-1-0.6B` | 1024 | 32K tokens | Yes | INT8/BINARY | No |
18
  | `pplx-embed-1-4B` | 2560 | 32K tokens | Yes | INT8/BINARY | No |
19
  | `pplx-embed-1-context-0.6B` | 1024 | 32K tokens | Yes | INT8/BINARY | No |
20
  | `pplx-embed-1-context-4B` | 2560 | 32K tokens | Yes | INT8/BINARY | No |
 
21
  <sub>All models are built on diffusion continued pre-trained Qwen3 at Perplexity AI.</sub>
 
22
  <sub>Many modern embedding models rely on instruction tuning, where users prepend an instruction string to the text being embedded. This can yield a 2%-3% lift on benchmarks, but it also introduces prompt-selection overhead and can make indexing pipelines brittle (small instruction changes can shift embedding space). We deliberately **avoid** this requirement: you can embed the text you want to index directly, without having to choose or maintain an instruction prefix.</sub>
23
 
24
  ## Quick Start
 
13
  ## Models
14
 
15
  | Model | Dimensions | Context | MRL | Quantization | Instruction |
16
+ |:-----:|:----------:|:-------:|:---:|:------------:|:-----------:|
17
  | `pplx-embed-1-0.6B` | 1024 | 32K tokens | Yes | INT8/BINARY | No |
18
  | `pplx-embed-1-4B` | 2560 | 32K tokens | Yes | INT8/BINARY | No |
19
  | `pplx-embed-1-context-0.6B` | 1024 | 32K tokens | Yes | INT8/BINARY | No |
20
  | `pplx-embed-1-context-4B` | 2560 | 32K tokens | Yes | INT8/BINARY | No |
21
+
22
  <sub>All models are built on diffusion continued pre-trained Qwen3 at Perplexity AI.</sub>
23
+
24
  <sub>Many modern embedding models rely on instruction tuning, where users prepend an instruction string to the text being embedded. This can yield a 2%-3% lift on benchmarks, but it also introduces prompt-selection overhead and can make indexing pipelines brittle (small instruction changes can shift embedding space). We deliberately **avoid** this requirement: you can embed the text you want to index directly, without having to choose or maintain an instruction prefix.</sub>
25
 
26
  ## Quick Start