LLM Templates - Created vision_gpt_explain_image llm application

examples/vision_gpt_explain_image/app.py

@@ -0,0 +1,56 @@
+import agenta as ag
+from openai import OpenAI
+from typing import List, Dict
+
+
+client = OpenAI()
+
+ag.init(app_name="explain_image", base_name="app")
+ag.config.default(
+    model=ag.MultipleChoiceParam("gpt-4-vision-preview", []),
+    max_tokens=ag.IntParam(300, -1, 4000),
+)
+
+
+def replace_image_url(
+    messages: List[Dict[str, str]], image_one: str, image_two: str
+) -> Dict[str, str]:
+    new_message = {}
+    for message in messages:
+        for key, value in message.items():
+            if key == "content":
+                new_content = []
+                for content in value:
+                    if content["type"] == "image_url":
+                        content["image_url"] = (
+                            {"url": image_two}
+                            if content["image_url"] == image_one
+                            else {"url": image_one}
+                        )
+                    new_content.append(content)
+                new_message[key] = new_content
+            else:
+                new_message[key] = value
+    return new_message
+
+
+@ag.entrypoint
+def explain(
+    image_one: ag.FileInputURL,
+    image_two: ag.FileInputURL,
+    inputs: ag.DictInput = ag.DictInput(default_keys=["role"]),
+    messages: ag.MessagesInput = ag.MessagesInput(
+        [
+            {"type": "text", "text": "What are in these image?"},
+        ]
+    ),
+) -> str:
+    messages = [inputs] + [{"content": messages}]
+    new_messages = replace_image_url(messages, image_one, image_two)
+    max_tokens = ag.config.max_tokens if ag.config.max_tokens != -1 else None
+    chat_completion = client.chat.completions.create(
+        model=ag.config.model,
+        messages=[new_messages],
+        max_tokens=max_tokens,
+    )
+    return chat_completion.choices[0].message.content

examples/vision_gpt_explain_image/requirements.txt

@@ -0,0 +1,2 @@
+agenta
+openai