0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan-AI-php.cn

0Code fine-tuning of a large model costs less than $20(approximately RMB 144 yuan)?

The process is also very simple,only takes 5 steps.

Such as LLaMA, GPT, StableLM and other popular open source generative models can be solved.

0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan Picture

This isMonster API, a newly popular API platform.

Some people think that this new work in the open source field can rewrite the game rules of AI development and accelerate the speed of AI applications.

0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan Picture

Some people are excited to ask, will GPT-3/GPT-4 be accessed in the future?

0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan Picture

So, how is it implemented?

Five steps to get it done with zero code

Simple understanding, Monster API is to simplify the fine-tuning steps as much as possible, so that developers no longer need to manually perform a series of settings, while also providing cheap GPU resources and Memory optimization.

The specific process is as follows:

The first stepis to select a model for fine-tuning.

For example, LLaMA-7B, GPT-J-6B, StableLM-7B, etc., Monster API provides at least 10 basic large models.

0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan Picture

Step 2, select or create a fine-tuning task. For example, instruction fine-tuning, text classification, etc., or custom tasks.

0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan Picture

The third step, select a HuggingFace data set.

Monster API can seamlessly integrate the HuggingFace data set, with a wide range of choices; it can also recommend data sets based on task types.

And the format can be set automatically without having to do it manually.

0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan Picture

The fourth stepis to set the hyperparameters.

0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan Picture

Step 5, check and submit.

After setting up all the above steps, make sure there are no errors before submitting.

Monster API indicates that tasks can be monitored through logs on WandB.

wrote in his blog that using DataBricks Dolly 15k fine-tuned LLaMA-7B to complete 3 epouches cost less than 20 US dollars (equivalent to about 144 yuan in RMB).

The official website shows that 2,500 points will be given as a gift after registering as a user. Membership is divided into three levels, and the fees are respectively US$9/US$29/US$39 per month.

0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan Picture

Monster API not only provides fine-tuning functions, but also provides various generative AI API interfaces, and the cost is 80% lower than other solutions.

0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan Picture

The company behind it has received $1.1 million in financing

News shows that the company behind Monster API has won110 Ten thousand US dollars in pre-seed funding (pre-seed funding).

This AI start-up company has positioned itself to be the "Airbnb in the GPU field", flexibly scheduling the scattered GPU resources around the world so that developers can use it at a lower price.

0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan Picture

The founders are two brothers Gaurav Vij and Saurabh Vij.

Among them, Gaurav Vij also founded a CV company. It was precisely because CV companies needed to face huge cloud computing capital that they were inspired to build such a platform.

Saurabh Vij was previously a particle physicist at CERN and also studied distributed computing.

The brothers said that after multiple rounds of technology iterations, they optimized the performance of consumer-grade GPUs on machine learning tasks, reducing the cost of running the Whisper AI model by 90% compared to the AWS platform, so they thought Why not use this method to help tens of thousands of developers.

At the same time, they revealed that one of the company's customers has saved $300,000 by using their distributed GPU computing resources.

Reference link:
[1]https://blog.monsterapi.ai/no-code-fine-tuning-llm/
[2] https://www.enterpriseai.news/2023/06/09/monster-api-launches-the-airbnb-of-gpus-with-1-1m-pre-seed/

The above is the detailed content of 0 code fine-tuning of large models is popular, only 5 steps are needed, and the cost is as low as 150 yuan. For more information, please follow other related articles on the PHP Chinese website!