Hello!
I have two OpenAI API calls, 1 that feeds into the 2nd through a body parameter <apipt.1>, although I’m hitting the below issue:
Here is my 1st API call:
{
"model": "gpt-4-0125-preview",
"messages": [
{
"role": "system",
"content": "[instructions]\\nPlease format your response in JSON and in the following structure: Write the task titles you're analyzing in bold and underlined.\\nProvide an analysis of the project or workflow, preceed this with an underscore.\\nThen detail your response under the following labels, each on a new line and bolded in BBCode: '_[your title]', 'Description:', '_Assignee:', 'Due Date:' (no earlier than <todaysdate>), 'Priority:', '~Task ID: ~<taskID>~'. Tag insights as 'Type: -Addition', and improvements as 'Type: -Improvements'. Separate each with a line break and a backslash.\\n[/instructions]"
},
{
"role": "user",
"content": "Analyze tasks in workspace <workspace> and project <project>, considering <tasktitle>, <assignee>, <description>, <Duedate>, <Status>, <priority>, <Taskgroup>, <boardmembers> and <jobrole>."
}
],
"temperature": 0.3,
"max_tokens": 1000
}
And here is my 2nd API call which formats the first 1st call:
{
"model": "ft:davinci-002:personal::8vBItY4G",
"prompt": "Refine the analysis from <apipt.1> by strictly adhering to the detailed formatting and structural guidelines as outlined. Ensure the output matches the specified format to effectively address potential risks with actionable insights. Seperate each analysis with a line break and a backslash. Keep a very close eye on the use of underscores, tildes and backslashes, making sure you match their exact placement. Please apply the structural and formatting guidelines from the example to refine the initial analysis response <apipt.1>. The formatting must be exact, enhancing the analytical depth and insight into potential risks. Strictly adhere to the formatting guidelines, ensuring each task analysis is separated by a line break and backslash. The content should remain unchanged, except for formatting to meet the example's specifications.\n",
"temperature": 0,
"max_tokens": 256,
"top_p": 1,
"frequency_penalty": 0,
"presence_penalty": 0
}
Is there something happening with the fact that my second API call is for a fine tuned davinci model and not a gpt4 model as is the first API call?
Please help! Any insight is appreciated!
Thanks