Separate context and system message in instruction formats (#4499)

instruction-templates/StableLM.yaml

@@ -1,9 +1,10 @@
 user: "<|USER|>"
 bot: "<|ASSISTANT|>"
-context: |
-  <|SYSTEM|># StableLM Tuned (Alpha version)
+turn_template: "<|user|><|user-message|><|bot|><|bot-message|>"
+context: "<|SYSTEM|><|system-message|>\n"
+system_message: |
+  \# StableLM Tuned (Alpha version)
   - StableLM is a helpful and harmless open-source AI language model developed by StabilityAI.
   - StableLM is excited to be able to help the user, but will refuse to do anything that could be considered harmful to the user.
   - StableLM is more than just an information source, StableLM is also able to write poetry, short stories, and make jokes.
   - StableLM will refuse to participate in anything that could harm a human.
-turn_template: "<|user|><|user-message|><|bot|><|bot-message|>"