Configured more architectures to try and debug the fine-tuning issue each with different prompt styles

src/architectures.py

@@ -436,8 +436,10 @@ class HFInferenceEndpoint(ArchitectureComponent):
     A concrete pipeline component which sends the user text to a given llama chat based
     inference endpoint on HuggingFace
     """
-    def __init__(self, endpoint_url: str, model_name: str, system_prompt: str, max_new_tokens: int, temperature: float = 1.0):
+    def __init__(self, endpoint_url: str, model_name: str, system_prompt: str, max_new_tokens: int,
+                 temperature: float = 1.0, prompt_style: str = "full"):
         self.endpoint_url: str = endpoint_url
+        self.prompt_style = prompt_style
         self.model_name: str = model_name
         self.system_prompt: str = system_prompt
         self.max_new_tokens = max_new_tokens
@@ -466,8 +468,17 @@ class HFInferenceEndpoint(ArchitectureComponent):
             "Authorization": f"Bearer {self.api_token}",
             "Content-Type": "application/json"
         }
-        #return f"<s>[INST] <<SYS>>\n{sys_prompt}\n<</SYS>>\n\n{q}[/INST]{a}"
-        query_input = f"<s>[INST] <<SYS>>\n{self.system_prompt}\n<</SYS>>\n\n{request.request}[/INST] "
+
+        if self.prompt_style == "multi_line":
+            query_input = f"<s>[INST] <<SYS>>\n{self.system_prompt}\n<</SYS>>\n\n{request.request} [/INST] "
+        elif self.prompt_style == "multi_line_no_sys":
+            query_input = f"<s>[INST]\n{request.request} [/INST] "
+        elif self.prompt_style == "single_line_no_sys":
+            query_input = f"<s>[INST] {request.request} [/INST] "
+        elif self.prompt_style == "single_line":
+            query_input = f"<s>[INST] <<SYS>>\n{self.system_prompt}\n<</SYS>> {request.request} [/INST] "
+        else:
+            raise ValueError(f"Config error - Unknown prompt style: {self.prompt_style}")
         payload = {
             "inputs": query_input,
             "parameters": {