Failure with deepset/translation prompt

[dkbs12]

Hello,
When I found the deepset/translation prompt in PromptHub, I decided to try it.
And I tried to translate English to French, but I failed to translate it.
I set the code as below(just as the example in PromptHub);

import os
from haystack.nodes import PromptNode, PromptTemplate
translation_template = PromptTemplate("deepset/translation")
prompt_node = PromptNode(model_name_or_path="text-davinci-003", api_key=api_key, model_kwargs={'temperature':0}, use_gpu=True)
prompt_node.prompt(prompt_template=translation_template, documents="I am a boy.", target_language="french")

and the result was as below;

['Je',
'Je veux aller à la plage.\nJe veux aller à la plage. Je veux aller à la plage.',
'a: Salut!',
'm = m',
'Je veux aller à la plage.\nJe veux aller à la plage. Je veux aller à la plage.',
'a: Salut!',
'Je veux aller à la plage.\nJe veux aller à la plage. Je veux aller à la plage.',
'b) Je veux aller à la plage.',
'Oui',
'y',
'Je suis heureux de vous rencontrer.']

Please tell me if there is any mistake with my coding.
If not, let me know how to get a correct performance.
Thanks.

[julian-risch]

Hi @dkbs12 it looks like the problem is that you pass a string "I am a boy" to the documents parameter of the PromptNode. It should be a list of documents instead. Try the following:

import os

from haystack.nodes import PromptNode, PromptTemplate
from haystack.schema import Document

translation_template = PromptTemplate("deepset/translation")

prompt_node = PromptNode(model_name_or_path="text-davinci-003", api_key=os.environ.get("OPENAI_API_KEY"))

prompt_node.prompt(prompt_template=translation_template, documents=[Document(content="I am a boy")], target_language="french")

You can also experiment with "gpt-3.5-turbo-instruct" instead of "text-davinci-003".

[dkbs12]

Hello,
Thank you for your help.
I tried to test with this set of API for many times and found that almost every response was done correctly.
However, the downside seems to be that the response time takes a long more than approximatelt 5 seconds.
Thank you again.