osmosis-ai

/

Osmosis-Structure-0.6B

like 360

Follow

osmosis 107

Hey! I noticed that you didn't specify a JSON schema in your system prompt. I replicated your prompt and added JSON schema, and here are the results:

(I'm using the ollama example as template)

from datetime import date
from ollama import chat
from pydantic import BaseModel


class RegistrationFee(BaseModel):
    early_bird_price: float
    regular_price: float
    early_bird_deadline: date


class ConferenceDetails(BaseModel):
    event_start_date: date
    event_end_date: date
    location: str
    registration_fees: RegistrationFee
    contact_email: str


reasoning_trace = """
The conference will be held on June 10-12, 2024 at the Grand Hotel in San Francisco.
Registration fee is $500 for early bird (before May 1) and $650 for regular registration.
Contact [email protected] for questions.
"""

response = chat(
    messages=[
        {
            "role": "system",
            "content": f"You are a helpful assistant that understands and translates text to JSON format according to the following schema. {ConferenceDetails.model_json_schema()}",
        },
        {
            "role": "user",
            "content": reasoning_trace,
        },
    ],
    model="hf.co/osmosis-ai/Osmosis-Structure-0.6B",
    format=ConferenceDetails.model_json_schema(),
)

answer = ConferenceDetails.model_validate_json(response.message.content)
print(answer.model_dump_json(indent=2))

Output:

{
  "event_start_date": "2024-05-10",
  "event_end_date": "2024-05-12",
  "location": "Grand Hotel, San Francisco",
  "registration_fees": {
    "early_bird_price": 500.0,
    "regular_price": 650.0,
    "early_bird_deadline": "2024-05-01"
  },
  "contact_email": "[email protected]"
}

Let me know if you still run into issues after specifying the schema.

woop woop , works perfect, i'm very happy

very good work here !