improve embeddings API results

README.md

@@ -91,6 +91,9 @@ body.json:
 
 ## Embeddings API Example
 
+Note: v1.4.0 has improved the quality of embeddings API. The generated results are not compatible
+with previous versions. If you are using embeddings API to generate knowledge bases or similar, please regenerate.
+
 If you are using langchain, just use `OpenAIEmbeddings(openai_api_base="http://127.0.0.1:8000", openai_api_key="sk-")`
 
 ```python

README_JA.md

@@ -91,7 +91,11 @@ body.json:
 
 ## 埋め込み API の例
 
-LangChain を使用している場合は、`OpenAIEmbeddings(openai_api_base="http://127.0.0.1:8000", openai_api_key="sk-")`を使用してください
+Note: v1.4.0 has improved the quality of embeddings API. The generated results are not compatible
+with previous versions. If you are using embeddings API to generate knowledge bases or similar, please regenerate.
+
+LangChain を使用している場合は、`OpenAIEmbeddings(openai_api_base="http://127.0.0.1:8000", openai_api_key="sk-")`
+を使用してください
 
 ```python
 import numpy as np

README_ZH.md

@@ -89,6 +89,8 @@ body.json:
 
 ## Embeddings API 示例
 
+注意: 1.4.0 版本对embeddings API质量进行了改善，生成结果与之前的版本不兼容，如果你正在使用此API生成知识库等，请重新生成
+
 如果你在用langchain, 直接使用 `OpenAIEmbeddings(openai_api_base="http://127.0.0.1:8000", openai_api_key="sk-")`
 
 ```python

backend-python/utils/rwkv.py

@@ -69,7 +69,7 @@ class AbstractRWKV(ABC):
             self.model_state = None
             self.model_tokens = []
             _, token_len = self.run_rnn(self.fix_tokens(self.pipeline.encode(input)))
-            embedding = self.model_state[-5].tolist()
+            embedding = self.model_state[-11].tolist()
         embedding = (embedding / np.linalg.norm(embedding)).tolist()
         return embedding, token_len