[Bug] Running custom workflows in evaluation

[snip2D]

Describe the bug
I customized a variant through the SDK and generated a test set in the playground. When I perform an evaluation, I encounter the following error message:
Traceback (most recent call last):
File "/app/agenta_backend/tasks/evaluations.py", line 297, in evaluate
type="text", value=app_output.result.value["data"]
KeyError: 'data'

variant project code
from pydantic import BaseModel, Field
import agenta as ag

class MyConfig(BaseModel):
prompt: str = Field(default="nihao")

@ag.route("/", config_schema=MyConfig)
def generate_text(input_text: str):
config = ag.ConfigManager.get_from_route(schema=MyConfig)
# Function implementation

output_text = f"input: {input_text} it's good! config: {config.prompt}"
return output_text

evaluate config

---

[mmabrouk]

Hello @snip2D ,

The code you are using looks good to me. I will try to reproduce the error. Can you please make sure to upgrade agenta locally to the latest version:
pip install -U agenta

then check which version are you using before serving the app again:

agenta --version
agenta variant serve app.py

Please let me know if that does not resolve the issue.

If the case persist:

• Can you please share with me the response of the application that you have created when called from the playground (you can go to the developer tools in your browser, and check the response to the request POST /generate).
• Can you use the application in the playground, or the issue is solely within evaluation.

As soon as we have that, and the problem is still unresolved, I will try to reproduce it from our side.

[snip2D]

Hi @mmabrouk I checked my agenta version, which was 0.29.0, and then I upgraded it to 0.31.0. After running the original code, a new error occurred：