Update README.md

README.md

@@ -529,4 +529,36 @@ print(tokenizer.apply_chat_template(chat, tokenize=False))
             raise RuntimeError("bad format: " + line)
           context[parts.group(1)] = method_map[parts.group(2)](parts.group(3), **context)
   ```
+</details>
+
+<details>
+  <summary>Create character cards</summary>
+
+  Included in the cinematika dataset, you can create YAML formatted character cards easily, e.g.:
+
+  ```text
+  Create a character card for Audrey, a woman who is the owner of a derelict building and is fiercely protective of her property. She should be portrayed as brave and resourceful, with a healthy skepticism towards the supernatural claims made by others. Audrey is determined to protect her family's legacy and the secrets it holds, often using intimidation and her practical approach to problem-solving to maintain control over her environment.
+  ```
+</details>
+
+<details>
+  <summary>Conversational memory creation</summary>
+
+  Also part of cinematika dataset, you can use a summarization style prompt to create memories from previous chat turns, which can then be used in a RAG system to populate your prompts when context becomes too long.
+
+  ```text
+  BEGININPUT
+  {chat}
+  ENDINPUT
+  BEGININSTRUCTION
+  Create a JSON formatted memory of the conversation with the following fields:
+  sentiment: Overall sentiment of the conversation, which must be "negative", "positive", "neutral", or "mixed".
+  emotions: List of most important/relevant emotions expressed within the conversation, if any.
+  impact: The importance and emotional impact of the conversation on a scale of 1 to 10, 10 being extremely important/emotional, and 1 being general chit-chat without anything of particular value.
+  topics: List of topics discussed.
+  personal_info: List of strings containing key personality traits, physical descriptions, preferences, quirks, interests, job, education, life goals, hobbies, pet names, or any other type of personal information that is shared.
+  title: Very brief title, which will be useful in quickly identifying or searching for memories.
+  summary: Summary of the conversation.
+  ENDINSTRUCTION
+  ```
 </details>