Implementation:Run llama Llama index OpenAIFinetuneEngine Get Finetuned Model

Overview

The get_finetuned_model method on OpenAIFinetuneEngine retrieves the completed finetuning job, extracts the finetuned model ID, and returns a fully configured OpenAI LLM instance ready for use in LlamaIndex pipelines. This method bridges the gap between a completed cloud training job and a usable local LLM object.

Source File

File: llama-index-finetuning/llama_index/finetuning/openai/base.py
Lines: 104-119
Import: Accessed as a method on OpenAIFinetuneEngine instances

Method Signature

def get_finetuned_model(self, **model_kwargs: Any) -> LLM:
    """Gets finetuned model."""

Parameters:

Parameter	Type	Description
`**model_kwargs`	`Any`	Keyword arguments passed to the `OpenAI` LLM constructor (e.g., `temperature`, `max_tokens`)

Returns: LLM -- An OpenAI LLM instance configured with the finetuned model ID

Raises:

ValueError -- If the job does not have a finetuned model ID ready (model still training or not started)
ValueError -- If the job status is not "succeeded"

Implementation Detail

def get_finetuned_model(self, **model_kwargs: Any) -> LLM:
    """Gets finetuned model."""
    current_job = self.get_current_job()

    job_id = current_job.id
    status = current_job.status
    model_id = current_job.fine_tuned_model

    if model_id is None:
        raise ValueError(
            f"Job {job_id} does not have a finetuned model id ready yet."
        )
    if status != "succeeded":
        raise ValueError(f"Job {job_id} has status {status}, cannot get model")

    return OpenAI(model=model_id, **model_kwargs)

The method performs the following steps:

Retrieve current job: Calls self.get_current_job() to fetch the latest job state from the OpenAI API
Extract metadata: Reads id, status, and fine_tuned_model from the job object
Validate model availability: Checks that fine_tuned_model is not None. This attribute is only populated when training completes successfully.
Validate job success: Confirms that the status is "succeeded". A job could theoretically have a model ID from a partial run but not be in a succeeded state.
Create LLM instance: Returns OpenAI(model=model_id, **model_kwargs), a standard LlamaIndex LLM wrapper that can be used anywhere an LLM is expected

Return Type

The returned object is an instance of llama_index.llms.openai.OpenAI, which implements the full LLM interface:

complete(prompt) -- Single completion
chat(messages) -- Chat completion
stream_complete(prompt) -- Streaming completion
stream_chat(messages) -- Streaming chat completion
acomplete(prompt) -- Async completion
achat(messages) -- Async chat completion

The finetuned model behaves identically to any other OpenAI model within LlamaIndex, just with domain-specific learned behavior from the training data.

Usage Example

from llama_index.finetuning import OpenAIFinetuneEngine

# After finetuning job has completed
engine = OpenAIFinetuneEngine(
    base_model="gpt-3.5-turbo",
    data_path="training_data.jsonl",
    start_job_id="ftjob-abc123",
)

# Get the finetuned model with custom inference parameters
ft_llm = engine.get_finetuned_model(temperature=0.3)

# Use it like any OpenAI model
response = ft_llm.complete("Explain vector similarity search.")
print(response)

# Use in chat mode
from llama_index.core.base.llms.types import ChatMessage
messages = [
    ChatMessage(role="user", content="What is semantic search?")
]
chat_response = ft_llm.chat(messages)
print(chat_response.message.content)

Knowledge Sources

LlamaIndex OpenAI Finetuning Engine Source

Metadata

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment