Replies: 1 comment 1 reply
-
@trnq-eu I facing same issue. Did you get any solution ? |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Hi.
I want to reply this tutoria using Mixtral 8 instead of GPT 3.5: https://haystack.deepset.ai/tutorials/28_structured_output_with_loop
Problem is that I keep getting an error because it seems I get an invalid json from the language model.
Errors are always like: Error from OutputValidator: Unterminated string starting at: line 5 column 15 (char 37)
OutputValidator at Iteration 14: Invalid JSON from LLM - Let's try again.
Output from LLM:
"cities": [
{
"name": "Berl
Error from OutputValidator: Unterminated string starting at: line 5 column 15 (char 37)
OutputValidator at Iteration 15: Invalid JSON from LLM - Let's try again.
Looks like Mixtral is unable to correctly produce a json output from this very simple string of text, but it seems improbable to me since I already managed to get a json output from Mixtral (using a retriever, for example), even though I never manage to validate the output against a json schema.
Here's a chunk of my code. The HuggingFaceTGIGenerator seems to be working fine
from haystack.components.generators import HuggingFaceTGIGenerator
from getpass import getpass
HF_TOKEN = getpass('HuggingFace Token:')
generator = HuggingFaceTGIGenerator(model="mistralai/Mixtral-8x7B-Instruct-v0.1", token=HF_TOKEN)
from haystack import Pipeline
pipeline = Pipeline(max_loops_allowed=5)
pipeline.add_component(instance=prompt_builder, name="prompt_builder")
pipeline.add_component(instance=generator, name="llm")
pipeline.add_component(instance=output_validator, name="output_validator")
pipeline.connect("prompt_builder", "llm")
pipeline.connect("llm", "output_validator")
pipeline.connect("output_validator.invalid_replies", "prompt_builder.invalid_replies")
pipeline.connect("output_validator.error_message", "prompt_builder.error_message")
Beta Was this translation helpful? Give feedback.
All reactions