Fine tuning classification example

Mar 10, 2022
Open in Github

We will fine-tune an ada classifier to distinguish between the two sports: Baseball and Hockey.

from sklearn.datasets import fetch_20newsgroups
import pandas as pd
import openai

categories = ['', '']
sports_dataset = fetch_20newsgroups(subset='train', shuffle=True, random_state=42, categories=categories)

Data exploration

The newsgroup dataset can be loaded using sklearn. First we will look at the data itself:

From: (Doug Bank)
Subject: Re: Info needed for Cleveland tickets
Organization: Motorola Land Mobile Products Sector
Distribution: usa
Lines: 17

In article <1993Apr1.234031.4950@leland.Stanford.EDU>, bohnert@leland.Stanford.EDU (matthew bohnert) writes:

|> I'm going to be in Cleveland Thursday, April 15 to Sunday, April 18.
|> Does anybody know if the Tribe will be in town on those dates, and
|> if so, who're they playing and if tickets are available?

The tribe will be in town from April 16 to the 19th.
There are ALWAYS tickets available! (Though they are playing Toronto,
and many Toronto fans make the trip to Cleveland as it is easier to
get tickets in Cleveland than in Toronto.  Either way, I seriously
doubt they will sell out until the end of the season.)

Doug Bank                       Private Systems Division          Motorola Communications Sector                   Schaumburg, Illinois       708-576-8207                    

len_all, len_baseball, len_hockey = len(, len([e for e in if e == 0]), len([e for e in if e == 1])
print(f"Total examples: {len_all}, Baseball examples: {len_baseball}, Hockey examples: {len_hockey}")
Total examples: 1197, Baseball examples: 597, Hockey examples: 600

One sample from the baseball category can be seen above. It is an email to a mailing list. We can observe that we have 1197 examples in total, which are evenly split between the two sports.

Data Preparation

We transform the dataset into a pandas dataframe, with a column for prompt and completion. The prompt contains the email from the mailing list, and the completion is a name of the sport, either hockey or baseball. For demonstration purposes only and speed of fine-tuning we take only 300 examples. In a real use case the more examples the better the performance.

import pandas as pd

labels = [sports_dataset.target_names[x].split('.')[-1] for x in sports_dataset['target']]
texts = [text.strip() for text in sports_dataset['data']]
df = pd.DataFrame(zip(texts, labels), columns = ['prompt','completion']) #[:300]
prompt completion
0 From: (Doug Bank)\nSubject:... baseball
1 From: (Gary L Dare)... hockey
2 From: (Rudy Wade)\nSubject: Re... baseball
3 From: (david... hockey
4 Subject: Let it be Known\nFrom: <ISSBTL@BYUVM.... baseball

Both baseball and hockey are single tokens. We save the dataset as a jsonl file.

df.to_json("sport2.jsonl", orient='records', lines=True)

Data Preparation tool

We can now use a data preparation tool which will suggest a few improvements to our dataset before fine-tuning. Before launching the tool we update the openai library to ensure we're using the latest data preparation tool. We additionally specify -q which auto-accepts all suggestions.

!pip install --upgrade openai
!openai tools fine_tunes.prepare_data -f sport2.jsonl -q

- Your file contains 1197 prompt-completion pairs
- Based on your data it seems like you're trying to fine-tune a model for classification
- For classification, we recommend you try one of the faster and cheaper models, such as `ada`
- For classification, you can estimate the expected model performance by keeping a held out dataset, which is not used for training
- There are 11 examples that are very long. These are rows: [134, 200, 281, 320, 404, 595, 704, 838, 1113, 1139, 1174]
For conditional generation, and for classification the examples shouldn't be longer than 2048 tokens.
- Your data does not contain a common separator at the end of your prompts. Having a separator string appended to the end of the prompt makes it clearer to the fine-tuned model where the completion should begin. See for more detail and examples. If you intend to do open-ended generation, then you should leave the prompts empty
- The completion should start with a whitespace character (` `). This tends to produce better results due to the tokenization we use. See for more details

Based on the analysis we will perform the following actions:
- [Recommended] Remove 11 long examples [Y/n]: Y
- [Recommended] Add a suffix separator `\n\n###\n\n` to all prompts [Y/n]: Y
- [Recommended] Add a whitespace character to the beginning of the completion [Y/n]: Y
- [Recommended] Would you like to split into training and validation set? [Y/n]: Y

Your data will be written to a new JSONL file. Proceed [Y/n]: Y

Wrote modified files to `sport2_prepared_train.jsonl` and `sport2_prepared_valid.jsonl`
Feel free to take a look!

Now use that file when fine-tuning:
> openai api fine_tunes.create -t "sport2_prepared_train.jsonl" -v "sport2_prepared_valid.jsonl" --compute_classification_metrics --classification_positive_class " baseball"

After you’ve fine-tuned a model, remember that your prompt has to end with the indicator string `\n\n###\n\n` for the model to start generating completions, rather than continuing with the prompt.
Once your model starts training, it'll approximately take 30.8 minutes to train a `curie` model, and less for `ada` and `babbage`. Queue will approximately take half an hour per job ahead of you.

The tool helpfully suggests a few improvements to the dataset and splits the dataset into training and validation set.

A suffix between a prompt and a completion is necessary to tell the model that the input text has stopped, and that it now needs to predict the class. Since we use the same separator in each example, the model is able to learn that it is meant to predict either baseball or hockey following the separator. A whitespace prefix in completions is useful, as most word tokens are tokenized with a space prefix. The tool also recognized that this is likely a classification task, so it suggested to split the dataset into training and validation datasets. This will allow us to easily measure expected performance on new data.


The tool suggests we run the following command to train the dataset. Since this is a classification task, we would like to know what the generalization performance on the provided validation set is for our classification use case. The tool suggests to add --compute_classification_metrics --classification_positive_class " baseball" in order to compute the classification metrics.

We can simply copy the suggested command from the CLI tool. We specifically add -m ada to fine-tune a cheaper and faster ada model, which is usually comperable in performance to slower and more expensive models on classification use cases.

!openai api fine_tunes.create -t "sport2_prepared_train.jsonl" -v "sport2_prepared_valid.jsonl" --compute_classification_metrics --classification_positive_class " baseball" -m ada
Upload progress: 100%|████████████████████| 1.52M/1.52M [00:00<00:00, 1.81Mit/s]
Uploaded file from sport2_prepared_train.jsonl: file-Dxx2xJqyjcwlhfDHpZdmCXlF
Upload progress: 100%|███████████████████████| 388k/388k [00:00<00:00, 507kit/s]
Uploaded file from sport2_prepared_valid.jsonl: file-Mvb8YAeLnGdneSAFcfiVcgcN
Created fine-tune: ft-2zaA7qi0rxJduWQpdvOvmGn3
Streaming events until fine-tuning is complete...

(Ctrl-C will interrupt the stream, but not cancel the fine-tune)
[2021-07-30 13:15:50] Created fine-tune: ft-2zaA7qi0rxJduWQpdvOvmGn3
[2021-07-30 13:15:52] Fine-tune enqueued. Queue number: 0
[2021-07-30 13:15:56] Fine-tune started
[2021-07-30 13:18:55] Completed epoch 1/4
[2021-07-30 13:20:47] Completed epoch 2/4
[2021-07-30 13:22:40] Completed epoch 3/4
[2021-07-30 13:24:31] Completed epoch 4/4
[2021-07-30 13:26:22] Uploaded model: ada:ft-openai-2021-07-30-12-26-20
[2021-07-30 13:26:27] Uploaded result file: file-6Ki9RqLQwkChGsr9CHcr1ncg
[2021-07-30 13:26:28] Fine-tune succeeded

Job complete! Status: succeeded 🎉
Try out your fine-tuned model:

openai api completions.create -m ada:ft-openai-2021-07-30-12-26-20 -p <YOUR_PROMPT>

The model is successfully trained in about ten minutes. We can see the model name is ada:ft-openai-2021-07-30-12-26-20, which we can use for doing inference.

!openai api fine_tunes.results -i ft-2zaA7qi0rxJduWQpdvOvmGn3 > result.csv
results = pd.read_csv('result.csv')
step elapsed_tokens elapsed_examples training_loss training_sequence_accuracy training_token_accuracy classification/accuracy classification/precision classification/recall classification/auroc classification/auprc classification/f1.0 validation_loss validation_sequence_accuracy validation_token_accuracy
929 930 3027688 3720 0.044408 1.0 1.0 0.991597 0.983471 1.0 1.0 1.0 0.991667 NaN NaN NaN

The accuracy reaches 99.6%. On the plot below we can see how accuracy on the validation set increases during the training run.

image generated by notebook
test = pd.read_json('sport2_prepared_valid.jsonl', lines=True)
prompt completion
0 From: (Gary L Dare)... hockey
1 From: (Ron Morris ... hockey
2 From: (Geral... hockey
3 From: (Kim Krattig... baseball
4 From: (Doug Dolven)\nSub... baseball

We need to use the same separator following the prompt which we used during fine-tuning. In this case it is \n\n###\n\n. Since we're concerned with classification, we want the temperature to be as low as possible, and we only require one token completion to determine the prediction of the model.

ft_model = 'ada:ft-openai-2021-07-30-12-26-20'
res = openai.Completion.create(model=ft_model, prompt=test['prompt'][0] + '\n\n###\n\n', max_tokens=1, temperature=0)
' hockey'

To get the log probabilities, we can specify logprobs parameter on the completion request

res = openai.Completion.create(model=ft_model, prompt=test['prompt'][0] + '\n\n###\n\n', max_tokens=1, temperature=0, logprobs=2)
<OpenAIObject at 0x7fe114e435c8> JSON: {
  " baseball": -7.6311407,
  " hockey": -0.0006307676

We can see that the model predicts hockey as a lot more likely than baseball, which is the correct prediction. By requesting log_probs, we can see the prediction (log) probability for each class.


Interestingly, our fine-tuned classifier is quite versatile. Despite being trained on emails to different mailing lists, it also successfully predicts tweets.

sample_hockey_tweet = """Thank you to the 
 and all you amazing Caniacs that have been so supportive! You guys are some of the best fans in the NHL without a doubt! Really excited to start this new chapter in my career with the 
res = openai.Completion.create(model=ft_model, prompt=sample_hockey_tweet + '\n\n###\n\n', max_tokens=1, temperature=0, logprobs=2)
' hockey'
sample_baseball_tweet="""BREAKING: The Tampa Bay Rays are finalizing a deal to acquire slugger Nelson Cruz from the Minnesota Twins, sources tell ESPN."""
res = openai.Completion.create(model=ft_model, prompt=sample_baseball_tweet + '\n\n###\n\n', max_tokens=1, temperature=0, logprobs=2)
' baseball'