{ "slug": "How-to-Fine-tune-ChatGPT-305606a02605", "title": "How to Fine-tune ChatGPT", "subtitle": "Train ChatGPT with your own data.", "excerpt": "Train ChatGPT with your own data.", "date": "2024-03-12", "tags": [ "Design Rag System", "Machine Learning" ], "readingTime": "2 min", "url": "https://medium.com/@mobinshaterian/fine-tune-chatgpt-gpt-305606a02605", "hero": "https://cdn-images-1.medium.com/max/800/1*ytRq2w4AZrMI_eocKt21fA.jpeg", "content": [ { "type": "paragraph", "html": "Train ChatGPT with your own data." }, { "type": "paragraph", "html": "https://platform.openai.com/docs/guides/fine-tuning" }, { "type": "image", "src": "https://cdn-images-1.medium.com/max/800/1*ytRq2w4AZrMI_eocKt21fA.jpeg", "alt": "How to Fine-tune ChatGPT", "caption": "", "width": 640, "height": 832 }, { "type": "paragraph", "html": "OpenAI’s text generation models have been pre-trained on a vast amount of text. To use the models effectively, we include instructions and sometimes several examples in a prompt. Using demonstrations to show how to perform a task is often called “few-shot learning.”" }, { "type": "paragraph", "html": "Fine-tuning improves on few-shot learning by training on many more examples than can fit in the prompt, letting you achieve better results on a wide number of tasks. Once a model has been fine-tuned, you won’t need to provide as many examples in the prompt. This saves costs and enables lower-latency requests." }, { "type": "paragraph", "html": "Fine-tuning is currently available for the following models: gpt-3.5-turbo-0125 (recommended), gpt-3.5-turbo-1106, gpt-3.5-turbo-0613, babbage-002, davinci-002, and gpt-4-0613 (experimental)." }, { "type": "heading", "level": 2, "text": "Preparing your dataset" }, { "type": "paragraph", "html": "In this example, our goal is to create a chatbot that occasionally gives sarcastic responses, these are three training examples (conversations) we could create for a dataset:" }, { "type": "code", "lang": "json", "code": "{\"messages\": [{\"role\": \"system\", \"content\": \"Marv is a factual chatbot that is also sarcastic.\"},\n{\"role\": \"user\", \"content\": \"What's the capital of France?\"}, {\"role\": \"assistant\", \"content\":\n\"Paris, as if everyone doesn't know that already.\"}]}{\"messages\": [{\"role\": \"system\", \"content\":\n\"Marv is a factual chatbot that is also sarcastic.\"}, {\"role\": \"user\", \"content\":\n\"Who wrote 'Romeo and Juliet'?\"}, {\"role\": \"assistant\", \"content\":\n\"Oh, just some guy named William Shakespeare. Ever heard of him?\"}]}{\"messages\": [{\"role\": \"system\",\n\"content\": \"Marv is a factual chatbot that is also sarcastic.\"}, {\"role\": \"user\", \"content\":\n\"How far is the Moon from Earth?\"}, {\"role\": \"assistant\", \"content\":\n\"Around 384,400 kilometers. Give or take a few, like that really matters.\"}]}" }, { "type": "heading", "level": 2, "text": "Convert CSV to json" }, { "type": "code", "lang": "python", "code": "json_string = \"\"\nfor index, row in data.iterrows(): message = str(row['Subject']) + \" \" +\nstr(row['Description']) answer = str(row['Answer']) prompt =\n\"You are a formal chat bot assistant.\" json_text =\n'{\"messages\": [{\"role\": \"system\", \"content\":\"'+prompt\n+'\"}, {\"role\": \"user\", \"content\":\"'+message+'\" }, {\"role\": \"assistant\", \"content\":\"'+answer+'\"} ]}'\njson_string = json_string + json_text + \"\\n\"" }, { "type": "code", "lang": "python", "code": "with open(\"out.json\", \"w\") as text_file: text_file.write(json_string)" }, { "type": "heading", "level": 2, "text": "ImportError: cannot import name ‘OpenAI’ from ‘openai’" }, { "type": "paragraph", "html": "Run:
pip install openai — upgrade" }, { "type": "paragraph", "html": "This is available only in version openai==1.2.0" }, { "type": "heading", "level": 2, "text": "Upload a training file" }, { "type": "code", "lang": "python", "code": "from openai\nimport OpenAIclient = OpenAI()client.files.create( file=open(\"mydata.jsonl\", \"rb\"),\npurpose=\"fine-tune\")" }, { "type": "code", "lang": "python", "code": "FileObject(id='file-KAdfdfgefd*****', bytes=54530, created_at=1710157205, filename='out.json',\nobject='file', purpose='fine-tune', status='processed', status_details=None)" }, { "type": "heading", "level": 2, "text": "Waiting for Queue" }, { "type": "code", "lang": "python", "code": "client.fine_tuning.jobs.list(limit=1)" }, { "type": "code", "lang": "python", "code": "SyncCursorPage[FineTuningJob](data=[FineTuningJob(id='ftjob-zde************', created_at=1710169666,\nerror=Error(code=None, message=None, param=None, error=None), fine_tuned_model=None,\nfinished_at=None, hyperparameters=Hyperparameters(n_epochs=3, batch_size=1,\nlearning_rate_multiplier=2), model='gpt-3.5-turbo-0125', object='fine_tuning.job',\norganization_id='org-nUFI**************', result_files=[], status='queued', trained_tokens=None,\ntraining_file='file-TlDij************', validation_file=None, user_provided_suffix=None)],\nobject='list', has_more=True)" }, { "type": "heading", "level": 2, "text": "Status Of traning" }, { "type": "code", "lang": "python", "code": "client.fine_tuning.jobs.retrieve(\"ftjob-zde*********\")" }, { "type": "code", "lang": "python", "code": "FineTuningJob(id='ftjob-zde***********', created_at=1710169666, error=Error(code=None, message=None,\nparam=None, error=None), fine_tuned_model=None, finished_at=None,\nhyperparameters=Hyperparameters(n_epochs=3, batch_size=1, learning_rate_multiplier=2),\nmodel='gpt-3.5-turbo-0125', object='fine_tuning.job', organization_id='org-nUFIO0I7*********',\nresult_files=[], status='queued', trained_tokens=None, training_file='file-TlDi****',\nvalidation_file=None, user_provided_suffix=None)" }, { "type": "heading", "level": 3, "text": "Check the status" }, { "type": "code", "lang": "python", "code": "res = client.fine_tuning.jobs.retrieve(\"ftjob-zde*************\")\nres.status" }, { "type": "code", "lang": "python", "code": "status='queued'\nstatus='running'\nstatus='succeeded'" }, { "type": "heading", "level": 2, "text": "Use a fine-tuned model" }, { "type": "paragraph", "html": "name of model" }, { "type": "code", "lang": "text", "code": "res.fine_tuned_model" }, { "type": "code", "lang": "python", "code": "completion = client.chat.completions.create( model=res.fine_tuned_model, messages=[ {\"role\":\n\"system\", \"content\": \"You are a formal chat bot.\"}, {\"role\": \"user\", \"content\": \"Hello\"}\n])\nprint(completion.choices[0].message.content)" }, { "type": "heading", "level": 2, "text": "Stackademic 🎓" }, { "type": "paragraph", "html": "Thank you for reading until the end. Before you go:" }, { "type": "list", "ordered": false, "items": [ "Please consider clapping and following the writer! 👏", "Follow us X | LinkedIn | YouTube | Discord", "Visit our other platforms: In Plain English | CoFeed | Venture | Cubed", "More content at Stackademic.com" ] } ] }