fix prompt bug

.env

@@ -1,4 +1,4 @@
 LANGSMITH_TRACING=true
 LANGSMITH_ENDPOINT="https://api.smith.langchain.com"
 LANGSMITH_API_KEY="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"
-LANGSMITH_PROJECT="pr-internal-neglect-42"
+LANGSMITH_PROJECT="pr-internal-neglect-42".  

app.py

@@ -1,7 +1,7 @@
 import gradio as gr
 from services.analyse_image import ErrorResponse, analyze_image
-from utils.utils import image_to_base64_data_uri, read_image_as_data_url
 from dotenv import load_dotenv
+from utils.utils import image_to_base64_data_uri
 
 load_dotenv()
 
@@ -94,7 +94,7 @@ with gr.Blocks() as demo:
         fn=upload_and_analyse,
         inputs=image_input,
         outputs=[car_block, animal_block, flower_block, status],
-        show_progress=False  # 🛠️ Important when using async fn
+        show_progress=True  # 🛠️ Important when using async fn
     )
 
 demo.launch()

prompt/object_prompt.py

@@ -3,8 +3,7 @@ from langchain_core.messages import SystemMessage, HumanMessage, BaseMessage
 from pydantic import BaseModel
 from langchain.output_parsers import PydanticOutputParser
 from models.structured_model import StructuredModel
-
-
+from langchain.prompts import PromptTemplate
 from typing import Literal, TypedDict, Union, List
 
 
@@ -20,30 +19,32 @@ ChatMessages = List[ChatMessage]
 
 
 
-def get_system_intructions(structure: StructuredModel):
-    parser = PydanticOutputParser(pydantic_object=structure)
-    return parser.get_format_instructions()
+# def get_system_intructions(structure: StructuredModel):
+#     parser = PydanticOutputParser(pydantic_object=structure)
+#     return parser.get_format_instructions()
 
-def convert_langchain_messages(messages: List[BaseMessage]) -> ChatMessages:
-     return [{"role": m.type, "content": m.content} for m in messages]
+# def convert_langchain_messages(messages: List[BaseMessage]) -> ChatMessages:
+#      return [{"role": m.type, "content": m.content} for m in messages]
     
 
 def get_prompt_template(label: str, image_data_url: str, structure: StructuredModel):
-    
-    print(get_system_intructions(structure))
-    
-    
-    return convert_langchain_messages([
-        SystemMessage(content=(
-            "You are a vision-language model specialized in analyzing objects in images. "
-            "If you don't know the value of an attribute asked to be extracted, return null. "
-            f"The object in the image is {label}."
-            f"{get_system_intructions(structure)}"
-        )),
-        HumanMessage(content=[
-            {"type": "image_url", "image_url": {"url": image_data_url}}
-        ])
-    ])
+  parser = PydanticOutputParser(pydantic_object=structure)
+  prompt = PromptTemplate(
+        template="Answer the user query.\n{format_instructions}\n{query}\n",
+        input_variables=["query"],
+        partial_variables={"format_instructions": parser.get_format_instructions()},
+    )
+  prompt = prompt.format(query=f"Describe the {label}")
+  messages = [
+            {
+                "role": "user",
+                "content": [
+                    {"type" : "text", "text": prompt},
+                    {"type": "image_url", "image_url": {"url": image_data_url } }
+                ]
+            }
+    ]
+  return messages
     
 
     

pyproject.toml

@@ -16,6 +16,7 @@ dependencies = [
     "langchainhub>=0.1.21",
     "langgraph>=0.5.4",
     "llama-cpp-python>=0.3.14",
+    "pillow>=11.3.0",
     "python-multipart>=0.0.20",
     "text-generation>=0.7.0",
     "torch>=2.7.1",

utils/utils.py

@@ -2,16 +2,6 @@ import base64
 import io
 from PIL import Image
 
-def encode_image_as_data_url(image_bytes: bytes, mime_type: str = "image/jpeg") -> str:
-    image_b64 = base64.b64encode(image_bytes).decode("utf-8")
-    return f"data:{mime_type};base64,{image_b64}"
-
-
-def read_image_as_data_url(image_path: str, mime_type: str = "image/jpeg") -> str:
-    with open(image_path, "rb") as f:
-        image_bytes = f.read()
-    return encode_image_as_data_url(image_bytes, mime_type)
-
 
 def image_to_base64_data_uri(file_path, size=(224, 224)):
     # Open the image

uv.lock

@@ -1357,6 +1357,7 @@ dependencies = [
     { name = "langchainhub" },
     { name = "langgraph" },
     { name = "llama-cpp-python" },
+    { name = "pillow" },
     { name = "python-multipart" },
     { name = "text-generation" },
     { name = "torch" },
@@ -1377,6 +1378,7 @@ requires-dist = [
     { name = "langchainhub", specifier = ">=0.1.21" },
     { name = "langgraph", specifier = ">=0.5.4" },
     { name = "llama-cpp-python", specifier = ">=0.3.14", index = "https://abetlen.github.io/llama-cpp-python/whl/metal/" },
+    { name = "pillow", specifier = ">=11.3.0" },
     { name = "python-multipart", specifier = ">=0.0.20" },
     { name = "text-generation", specifier = ">=0.7.0" },
     { name = "torch", specifier = ">=2.7.1" },