I am the only one who practices magic: I practice magic in the city

Chapter 159 Orange Large Model (Second Update)
To create an artificial intelligence, Fang Yu cannot just give Youzi instructions.

Perhaps due to safety considerations, Sekalis did not add independent innovation awareness to the ball when creating the grapefruit.

Only when the owner of the grapefruit gives it clear instructions, will it carry out the next step of information collection and calculation based on the instructions.

For a systematic task, Yuzi needs his master to break it down into independent and clear tasks.

If Fang Yu just told it: Youzi, create a general artificial intelligence.

It will be completely at a loss and will not know where to start. It will not proactively analyze what kind of prior knowledge is needed for the task, nor will it creatively perform the task.

In this case, Fang Yu himself needs to have a good understanding of the big artificial intelligence model first.

Give you a direction, break down the tasks, and let you execute them.

If the execution is successful, then it has been done. If it fails, it means that Fang Yu’s understanding of artificial intelligence is problematic and the direction is wrong, so he needs to change direction.

After all, there is a reason why Yuzu is so lazy!

"Now, make a preliminary deduction of the plan I just mentioned, and then tell me the results." Fang Yu was giving instructions to Youzi while typing away on the computer to revise the summary.

The stage summary written by Youzi was too advanced, so Fang Yu had to delete part of it and correct another part before handing it to Tong Yongshan.

Now Fang Yu has established the impression of being an expert in computer science in Tong Yongshan's mind. Therefore, Tong Yongshan did not look for any computer science experts to cooperate with in the first stage, but directly reached a cooperative research intention with Youzi Technology.

As for who owns Youzi Technology, Tong Yongshan doesn't care at all.

Even if this company belongs to this student, so what? Isn't it good to have a second-generation student who is willing to invest tens of millions in scientific research?

If other professors knew about this, they would be extremely jealous, right?
"Master, the combination of ResNet and DBN will not improve the ability to handle long-distance dependencies. We need to change the model combination." After a short period of calculation, Yuzu replied with silver-white starlight.

"Then replace them and see which models can create sparks." Fang Yu circled the model range on the server and gave it to Youzi to calculate.

Fuzzy calculation is one of the laws that grapefruit possesses. Its essence is actually the application of some basic theories of the prophecy school to grapefruit.

It is completely different from the way of extracting key factors in computer models for fuzzy calculation. In comparison, the accuracy of Yuzu's fuzzy calculation is much higher.

"Found it!" Not long after, Yuzu made an excited sound.

"The combination of the attention mechanism model mentioned in "Sequence to Sequence with Attention" with position encoding technology and neural networks such as LSTM, CNN, and FFN can form a new framework. After training, adding other models can form the most efficient large model in current measurements." Youzi quickly projected the corresponding content on Fang Yu's computer screen.

“I will put these models into my own framework and restructure them.”

As she spoke, the silver starlight on Yuzu's body gradually turned into a slowly flowing band of light. This was Yuzu using her body to help the array server perform auxiliary calculations, and then transmitting the results of the auxiliary calculations to the server network.

As long as the right path is found, the computing power of Yuzu is far beyond the reach of any supercomputer today.

In just three minutes, the silver light bands flowing on his body turned back into silver starlight.

"The framework has been reorganized, the underlying rules have been added, and the initial learning will be carried out after uploading to the server system." Yuzu's straightforward voice came, and it seemed like a ball had changed, just like the artificial intelligence in the movie.

Fang Yu rubbed his fingers unmoved: "Speak properly!"

Yuzi was visibly startled, and said in a flattering tone: "Master, it will take about 30 seconds for the new framework to be uploaded to the server, and then we can do model data training."

As he was speaking, the not-so-large framework model had been completely uploaded to the server unit, and the Yuzu clone in the server began to add the processed data to the new framework.

12 groups of GPU servers and 196 Tesla M60 computing chips are running at full capacity. The server's cooling fans rotate at a speed of revolutions, dissipating heat into a space of less than square meters.

"Currently, 40GB of data has been cleaned and we are now starting to use this data for the first training. This model will require 30 iterations to converge, which is expected to take 3 hours and 21 minutes."

Then, a countdown of 3 hours, 21 minutes and 20 seconds appeared on Fang Yu's screen. Fang Yu rubbed his hands and walked around the room twice, feeling both expectant and excited.

If this model can demonstrate certain analytical and processing capabilities and communication abilities after data training, it means that you have really created a usable large artificial intelligence model!
You said it was Yuzu who did it?
Bullshit, I’m the corresponding author, do you understand?

What I do is the work of a mentor: determining the research direction, dividing the research stages, and guiding the research content.

The project is completed, and the tutor has only done these things. Who dares to say that the project was not done by the tutor?

Do you understand scientific research? Can scientific research be called exploitation? Can it be called black-hearted?

This statement completely ignores the contributions of leading scientific research figures!

You said Tong Yongshan won’t give me a chance?
Hey! You black-hearted old thief! Academic tycoon! Shameless black-hearted tutor!
Fang Yu walked around twice, and in order to calm himself down, he simply ran back to the Westin and had another big fight with the four invaders from the pirate world who were about to escape.

The invaders of the pirate world are almost dying from the invasion.

When we checked out three hours later, all four of us had our legs splayed out.

When Fang Yu returned to the computer room again, 40G of model data had been trained.

"Master, 40G of data has been trained, of which 20G is Zhou Wen's data and 20G is Ang Wen's data. Judging from the convergence, this training is successful. We can do some tests on this model."

A simple dialog box appeared on Fang Yu's computer screen. This was a UI interface that Youzi had just made.

"Hello." Fang Yu entered two Zhou Wen in the dialog box.

"I'm not okay." Three words instantly appeared in the dialog box.

What the hell? Is there something wrong with your data cleaning? What data did you use to train it?

Yuzu was sweating. "Well, well, maybe there is something wrong with the training data. I'll clean it again."

The server ran frantically for a few minutes and the 40g data was cleaned up.

"Retraining will take about 1 hour and 20 minutes." Yuzu's voice was a little awkward.

Fang Yu was helpless. He looked at the sky and saw that it was already four o'clock in the afternoon.

I was originally thinking of going to the boxing club to show off and beat up the newbies, but it seems I don’t have the time anymore.

Just wait here.

Fang Yu is not a person who wastes time. One hour and twenty minutes is enough for him to review a subject he is about to take an exam on.

Finally, the countdown is over.

Fang Yu entered the word "Hello" in the dialog box again.

"Hello, how can I help you?" This response is much more normal.

"What's your name?" Fang Yu typed out another line of text.

The model replied in a standard way: "I don't have a name yet. I am a large model based on multiple neural networks developed by Youzi Technology. What do you want to call me?"

Fang Yu thought about it and smiled: "Since you were developed by Pomelo, I'll call you Orange."

(End of this chapter)