Hugging Face's logo

litert-community
/
Qwen3-Reranker-0.6B-LiteRT

Text Ranking
LiteRT
LiteRT
qwen3
qwen3-reranker
reranker
reranking
cross-encoder
rag
retrieval
on-device
gpu
android
Model card Files
xet
 Community
mlboydaisuke commited on
Commit
880b615
·
verified ·
1 Parent(s): 0513a34

Reference hero in card

Browse files
Files changed (1) hide show
  1. README.md +2 -0
README.md CHANGED
@@ -16,6 +16,8 @@ reranking half of an on-device RAG pipeline.
16
  Pairs with [`litert-community/Qwen3-Embedding-0.6B-LiteRT`](https://huggingface.co/litert-community/Qwen3-Embedding-0.6B-LiteRT):
17
  **embed → retrieve top-k → rerank**, all on-device, no server.
18
 
 
 
19
  Like the embedder it is a **single forward pass** (no generation, no KV cache) → a plain `.tflite`, not
20
  a `.litertlm`. Verified on a **Pixel 8a / Tensor G3**: all nodes on the GPU delegate, `P(yes)` parity
21
  **ref 0.9995 / dev 0.9994** vs the HF fp32 reference.
 
16
  Pairs with [`litert-community/Qwen3-Embedding-0.6B-LiteRT`](https://huggingface.co/litert-community/Qwen3-Embedding-0.6B-LiteRT):
17
  **embed → retrieve top-k → rerank**, all on-device, no server.
18
 
19
+ ![On-device RAG reranking on a Pixel 8a](hero.png)
20
+
21
  Like the embedder it is a **single forward pass** (no generation, no KV cache) → a plain `.tflite`, not
22
  a `.litertlm`. Verified on a **Pixel 8a / Tensor G3**: all nodes on the GPU delegate, `P(yes)` parity
23
  **ref 0.9995 / dev 0.9994** vs the HF fp32 reference.