#fine tuning

1 messages · Page 1 of 1 (latest)

ember warren
#

I have a specific use case to evaluate the price of an NFT. I fine tuned a Babbage-002 model and it’s working really well, but I wonder if it’s overkill. Is there an open source alternative where I can train my own model totally from scratch to just do this specific thing

verbal willowBOT
#

Hi @ember warren. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:

#

Based on the information provided in the extracts, OpenAI offers fine-tuning capabilities for several models including Babbage-002. Fine-tuning allows you to customize the model to better suit your specific use case, which in this case is evaluating the price of an NFT.

However, the extracts do not provide information on open source alternatives where you can train your own model totally from scratch. Training a model from scratch requires significant computational resources and expertise in machine learning and natural language processing.

If you find that the Babbage-002 model is more than what you need, you might consider fine-tuning a smaller model. OpenAI supports fine-tuning for several models including davinci, curie, babbage, and ada (source (https://platform.openai.com/docs/guides/legacy-fine-tuning)).

Remember, the choice of model and whether to fine-tune or train from scratch depends on your specific requirements, including the complexity of the task, the amount of training data you have, and the computational resources available to you.

For more information on fine-tuning with OpenAI, you can refer to the fine-tuning guide (https://platform.openai.com/docs/guides/fine-tuning) on the OpenAI platform.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

#

@ember warren kapa.ai is still learning and improving, please let us know how it did by reacting below

ember warren
#

Needs to know when OpenAI is not the best solution. I make my own AI powered discord bots, I wonder why you guys chose to partner with this one specific company to make your discord bots @verbal willow

#

???

leaden obsidian
# ember warren ???

you can't edit the message with the tag, you'll have to repost the question with the tag 😉

ember warren
ember warren
leaden obsidian
#

best of both worlds so to speak , as i see it anyway 😄 🍻

odd solar
leaden knoll
#

@verbal willow will Babbage always be supported, or will it be deprecated? If so, when? I need another model similar to its price point for a project. Do you recommend something else from OpenAI?

verbal willowBOT
#

Let me check the knowledge base for you. One moment! :books:

#

Based on the information provided in the extracts, OpenAI has announced the deprecation of certain models, including the babbage model. The shutdown date for the babbage model is set for January 4, 2024 (source (https://platform.openai.com/docs/deprecations)). The recommended replacement for the babbage model is babbage-002.

In terms of pricing, the babbage-002 model has a similar price point to the babbage model. The training price for babbage-002 is $0.4 per 1M tokens, and the usage price is $1.6 per 1M tokens (source (https://community.openai.com/t/567567)).

If you're looking for a model with a similar price point to babbage, babbage-002 would be a suitable choice. However, the choice of model should also take into account the specific requirements of your project, including the complexity of the task and the amount of training data you have.

For more information on the models available from OpenAI and their capabilities, you can refer to the Models (https://platform.openai.com/docs/models) section of the OpenAI platform documentation.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

#

@leaden knoll kapa.ai is still learning and improving, please let us know how it did by reacting below

ember warren
#

@verbal willow summarize the chat

verbal willowBOT
#

Let me check the knowledge base for you. One moment! :books:

#

In this conversation, you asked about fine-tuning models for a specific use case of evaluating the price of an NFT. I explained that OpenAI offers fine-tuning capabilities for several models, which allows you to customize the model to better suit your specific use case. However, the extracts did not provide information on open source alternatives for training your own model from scratch.

You also asked about the future support for the Babbage model. Based on the extracts, OpenAI has announced the deprecation of the babbage model, with a shutdown date set for January 4, 2024. The recommended replacement for the babbage model is babbage-002, which has a similar price point.

Here are the key sources referenced:

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

#

@ember warren kapa.ai is still learning and improving, please let us know how it did by reacting below

ember warren
#

@verbal willow the completions mode says its Legacy, does that mean it will be deprecated soon?

verbal willowBOT
#

Let me check the knowledge base for you. One moment! :books:

#

Yes, the Completions API endpoint is considered legacy and will be deprecated. According to the information provided in the extracts, the Completions API endpoint received its final update in July 2023 and will be retired on January 4, 2024 (source (https://platform.openai.com/docs/deprecations)).

OpenAI recommends transitioning to the new chat completions endpoint or using the gpt-3.5-turbo model for improved performance and lower cost (source (https://platform.openai.com/docs/models)).

For more information on how to use the new chat completions endpoint, you can refer to the Text generation models guide (https://platform.openai.com/docs/guides/text-generation) on the OpenAI platform.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

#

@ember warren kapa.ai is still learning and improving, please let us know how it did by reacting below

ember warren
#

@verbal willow it seems that fine tuning gpt3.5 is like 20x more expensive than fine tuning Babbage-002. If completions go away what will be my options

verbal willowBOT
#

Let me check the knowledge base for you. One moment! :books:

#

Based on the information provided in the extracts, the Completions API endpoint is indeed being deprecated and will be retired on January 4, 2024 (source (https://platform.openai.com/docs/deprecations)).

However, OpenAI has provided alternatives to the Completions API. One of these is the gpt-3.5-turbo model, which is recommended for most tasks due to its better performance compared to the GPT base models (source (https://platform.openai.com/docs/guides/fine-tuning)).

If you're concerned about the cost of fine-tuning gpt-3.5-turbo, you might consider using the babbage-002 model, which is a newer version of the babbage model and has a similar price point. The babbage-002 model supports fine-tuning and is designed to ensure a smooth transition from the legacy babbage model (source (https://platform.openai.com/docs/guides/fine-tuning)).

Remember, the choice of model should take into account the specific requirements of your project, including the complexity of the task, the amount of training data you have, and your budget.

For more information on the models available from OpenAI and their capabilities, you can refer to the Models (https://platform.openai.com/docs/models) section of the OpenAI platform documentation.

Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:

#

@ember warren kapa.ai is still learning and improving, please let us know how it did by reacting below

ember warren
#

@verbal willow when will Babbage-002 be retired

verbal willowBOT
#

Let me check the knowledge base for you. One moment! :books:

#

@ember warren That's a great question, but unfortunately I am only capable of responding to a maximum of five questions per thread. Don't hesitate to create a new thread for any further questions you have though. I'm here to help!

ember warren
odd solar
#

Having to account for message edits and then checking if a ping has occurred for a bot while being across multiple servers sounds like something that'll hit Discord rate limits pretty easily, especially if the bot is in multiple large servers derpcat

ember warren
# odd solar Having to account for message edits and then checking if a ping has occurred for...

no not really... there's two different events. MESSAGE_CREATE and MESSAGE_UPDATE. if you just handle the MESSAGE_UPDATE as if it's a MESSAGE_CREATE it can re-evaluate and decide if it should respond, or even edit it's previous response. it just takes smarter programming. also discord bots can handle tens of thousands of servers with sharding, even if there's 100,000 users on them. they dont need to load all members.

#

so since we can't ask it any more questions in this thread i guess we can discuss the misinformation it provided. completions end point did not get shut off yesterday

#

i understand it's new, testing, beta, whatever. that's fine. i hope they can use this to make it better

leaden obsidian
drowsy mist
#

@verbal willow summarize the weather of Iceland

verbal willowBOT
#

Let me check the knowledge base for you. One moment! :books: