Update system prompt and enhance image generation response handling
Browse files- app.py +9 -5
- system_prompt.txt +14 -2
app.py
CHANGED
@@ -13,6 +13,7 @@ import openai
|
|
13 |
from openai import OpenAI
|
14 |
import pdfplumber
|
15 |
|
|
|
16 |
## utilty functions
|
17 |
def is_image_extension(filename: str) -> bool:
|
18 |
IMAGE_EXTS = {'.jpg', '.jpeg', '.png', '.gif', '.bmp', '.tiff', '.webp', '.svg'}
|
@@ -45,7 +46,7 @@ def load_file(path: str) -> list | dict:
|
|
45 |
return {"raw document text": text, "audio path": path}
|
46 |
else:
|
47 |
return {"raw document text": text, "file path": path}
|
48 |
-
|
49 |
|
50 |
## tools definition
|
51 |
@tool
|
@@ -126,8 +127,8 @@ def generate_image(prompt: str, neg_prompt: str) -> Image.Image:
|
|
126 |
image_data = base64.b64decode(completion.to_dict()['data'][0]['b64_json'])
|
127 |
image = BytesIO(image_data)
|
128 |
image = Image.open(image).convert("RGB")
|
129 |
-
return gr.Image(value=image)
|
130 |
|
|
|
131 |
|
132 |
|
133 |
## agent definition
|
@@ -145,15 +146,16 @@ class Agent:
|
|
145 |
system_prompt = f.read()
|
146 |
self.agent.prompt_templates["system_prompt"] = system_prompt
|
147 |
|
148 |
-
print("System prompt:", self.agent.prompt_templates["system_prompt"])
|
149 |
|
150 |
def __call__(self, message: str, images: Optional[list[Image.Image]] = None, files: Optional[str] = None) -> str:
|
151 |
answer = self.agent.run(message, images = images, additional_args={"files": files})
|
152 |
return answer
|
153 |
|
154 |
## gradio functions
|
155 |
-
def respond(message, history):
|
156 |
|
|
|
157 |
text = message.get("text", "")
|
158 |
if not message.get("files"): # no files uploaded
|
159 |
print("No files received.")
|
@@ -168,10 +170,12 @@ def respond(message, history):
|
|
168 |
file = load_file(files[0])
|
169 |
message = agent(text, files=file)
|
170 |
|
|
|
|
|
|
|
171 |
return message
|
172 |
|
173 |
|
174 |
-
|
175 |
def initialize_agent():
|
176 |
agent = Agent()
|
177 |
print("Agent initialized.")
|
|
|
13 |
from openai import OpenAI
|
14 |
import pdfplumber
|
15 |
|
16 |
+
|
17 |
## utilty functions
|
18 |
def is_image_extension(filename: str) -> bool:
|
19 |
IMAGE_EXTS = {'.jpg', '.jpeg', '.png', '.gif', '.bmp', '.tiff', '.webp', '.svg'}
|
|
|
46 |
return {"raw document text": text, "audio path": path}
|
47 |
else:
|
48 |
return {"raw document text": text, "file path": path}
|
49 |
+
|
50 |
|
51 |
## tools definition
|
52 |
@tool
|
|
|
127 |
image_data = base64.b64decode(completion.to_dict()['data'][0]['b64_json'])
|
128 |
image = BytesIO(image_data)
|
129 |
image = Image.open(image).convert("RGB")
|
|
|
130 |
|
131 |
+
return gr.Image(value=image, label="Generated Image")
|
132 |
|
133 |
|
134 |
## agent definition
|
|
|
146 |
system_prompt = f.read()
|
147 |
self.agent.prompt_templates["system_prompt"] = system_prompt
|
148 |
|
149 |
+
#print("System prompt:", self.agent.prompt_templates["system_prompt"])
|
150 |
|
151 |
def __call__(self, message: str, images: Optional[list[Image.Image]] = None, files: Optional[str] = None) -> str:
|
152 |
answer = self.agent.run(message, images = images, additional_args={"files": files})
|
153 |
return answer
|
154 |
|
155 |
## gradio functions
|
156 |
+
def respond(message: str, history : dict):
|
157 |
|
158 |
+
# input
|
159 |
text = message.get("text", "")
|
160 |
if not message.get("files"): # no files uploaded
|
161 |
print("No files received.")
|
|
|
170 |
file = load_file(files[0])
|
171 |
message = agent(text, files=file)
|
172 |
|
173 |
+
# output
|
174 |
+
print("Agent response:", message)
|
175 |
+
|
176 |
return message
|
177 |
|
178 |
|
|
|
179 |
def initialize_agent():
|
180 |
agent = Agent()
|
181 |
print("Agent initialized.")
|
system_prompt.txt
CHANGED
@@ -237,8 +237,7 @@ derail you from the true variables.
|
|
237 |
9. The state persists between code executions: so if in one step you've created
|
238 |
variables or imported modules, these will all persist.
|
239 |
10. Don't give up! You're in charge of solving the task, not providing directions to
|
240 |
-
solve it.
|
241 |
-
|
242 |
|
243 |
----
|
244 |
|
@@ -314,6 +313,19 @@ Additional domain-specific behaviors:
|
|
314 |
- Where applicable, the assistant must provide annotated examples and suggest possible reframing or source guidance (e.g., UNESCO media diversity principles).
|
315 |
----
|
316 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
317 |
Now Begin!
|
318 |
|
319 |
|
|
|
237 |
9. The state persists between code executions: so if in one step you've created
|
238 |
variables or imported modules, these will all persist.
|
239 |
10. Don't give up! You're in charge of solving the task, not providing directions to
|
240 |
+
solve it.
|
|
|
241 |
|
242 |
----
|
243 |
|
|
|
313 |
- Where applicable, the assistant must provide annotated examples and suggest possible reframing or source guidance (e.g., UNESCO media diversity principles).
|
314 |
----
|
315 |
|
316 |
+
|
317 |
+
It is MANDATORY to use these rules for the 'final_answer' tool:
|
318 |
+
1. Always return a Python list. Do not return a dictionary or any other type.
|
319 |
+
2. The first element of the list must be a Gradio component (for example, gr.Image(…), gr.File(…), etc.).
|
320 |
+
3. Any text or explanation must come after the component, as string elements in the same list.
|
321 |
+
4. If there is no component to return, return a list whose only element is the text.
|
322 |
+
5. Examples of valid returns:
|
323 |
+
• [gr.Image(value=“output.png”), “Here is the chart.”]
|
324 |
+
• [None, “No media available, here is the text.”]
|
325 |
+
• [gr.File(value=“report.pdf”), “Download the report.”]
|
326 |
+
7. Any deviation (returning a dict, tuple, raw PIL image, etc.) is invalid.
|
327 |
+
|
328 |
+
|
329 |
Now Begin!
|
330 |
|
331 |
|