Add obisidian support

Author: abetlen

llama_cpp/llama_chat_format.py

@@ -2539,6 +2539,56 @@ def from_pretrained(
             **kwargs,
         )
 
+class ObsidianChatHandler(Llava15ChatHandler):
+    # Prompt Format
+    # The model followed ChatML format. However, with ### as the seperator
+
+    # <|im_start|>user
+    # What is this sign about?\n<image>
+    # ###
+    # <|im_start|>assistant
+    # The sign is about bullying, and it is placed on a black background with a red background.
+    # ###
+
+    CHAT_FORMAT = (
+        "{% for message in messages %}"
+        # System message
+        "{% if message.role == 'system' %}"
+        "<|im_start|>system\n"
+        "{{ message.content }}\n"
+        "###\n"
+        "{% endif %}"
+        # User message
+        "{% if message.role == 'user' %}"
+        "<|im_start|>user\n"
+        "{% if message.content is string %}"
+        "{{ message.content }}"
+        "{% endif %}"
+        "{% if message.content is iterable %}"
+        "{% for content in message.content %}"
+        "{% if content.type == 'text' %}"
+        "{{ content.text }}"
+        "{% endif %}"
+        "{% if content.type == 'image_url' %}"
+        "{{ content.image_url }}"
+        "{% endif %}"
+        "{% endfor %}"
+        "{% endif %}"
+        "###\n"
+        "{% endif %}"
+        # Assistant message
+        "{% if message.role == 'assistant' %}"
+        "<|im_start|>assistant\n"
+        "{{ message.content }}"
+        "###\n"
+        "{% endif %}"
+        "{% endfor %}"
+        # Generation prompt
+        "{% if add_generation_prompt %}"
+        "<|im_start|>assistant\n"
+        "{% endif %}"
+    )
+
 class MoondreamChatHanlder(Llava15ChatHandler):
     # Chat Format:
     # f"<image>\n\n{chat_history}Question: {question}\n\nAnswer:"

llama_cpp/server/model.py

@@ -84,6 +84,20 @@ def load_llama_from_model_settings(settings: ModelSettings) -> llama_cpp.Llama:
                 chat_handler = llama_cpp.llama_chat_format.Llava15ChatHandler(
                     clip_model_path=settings.clip_model_path, verbose=settings.verbose
                 )
+        elif settings.chat_format == "obsidian":
+            assert settings.clip_model_path is not None, "clip model not found"
+            if settings.hf_model_repo_id is not None:
+                chat_handler = (
+                    llama_cpp.llama_chat_format.ObsidianChatHandler.from_pretrained(
+                        repo_id=settings.hf_model_repo_id,
+                        filename=settings.clip_model_path,
+                        verbose=settings.verbose,
+                    )
+                )
+            else:
+                chat_handler = llama_cpp.llama_chat_format.ObsidianChatHandler(
+                    clip_model_path=settings.clip_model_path, verbose=settings.verbose
+                )
         elif settings.chat_format == "llava-1-6":
             assert settings.clip_model_path is not None, "clip model not found"
             if settings.hf_model_repo_id is not None: