Update README.md

README.md

@@ -86,7 +86,7 @@ The `RWKVReRanker` utilizes the final hidden state produced by the main `Embeddi
 #### Workflow
 1.  **Format** Query and Document based on Online template.
 2.  Run the **Embedding Model** to generate the final State.
-3.  Feed the **Attention State** (`state[1]`) into the **ReRanker** to get a relevance score.
+3.  Feed the **TimeMixing State** (`state[1]`) into the **ReRanker** to get a relevance score.
 
 #### 📝 Online Mode Usage Example
 
@@ -132,7 +132,7 @@ for i in range(len(batch_tokens)):
 _, state = emb_model.forward(batch_tokens, None)
 
 # 6. Score with ReRanker
-# The ReRanker expects the Attention State: state[1]
+# The ReRanker expects the TimeMixing State: state[1]
 # state[1] shape: [Layers, Batch, Heads, HeadSize, HeadSize]
 logits = reranker.forward(state[1])
 scores = torch.sigmoid(logits) # Convert logits to probabilities (0-1)