Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Try to guess if code is real or GPT2-generated

doesnotexist.codes

Post reply on HN
Print view View original

11–20 of 108 posts

Re: Try to guess if code is real or GPT2-generated

#11

Post by moyix »

Hi, author here! Some details on the model:

* Trained 17GB of code from the top 10,000 most popular Debian packages. The source files were deduplicated using a process similar to the OpenWebText preprocessing (basically a locality-sensitive hash to detect near-duplicates).

* I used the [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) code for training. Training took about 1 month on 4x RTX8000 GPUs.

* You can download the trained model here: https://moyix.net/~moyix/csrc_final.zip and the dataset/BPE vocab here: https://moyix.net/~moyix/csrc_dataset_large.json.gz https://moyix.net/~moyix/csrc_vocab_large.zip

Happy to answer any questions!

Re: Try to guess if code is real or GPT2-generated

#12

Post by moyix »

nickysielicki wrote: post #10

This is difficult... because these models are just regurgitating after training on real code. Fun little site but I hope nobody reads too much into this.

I've tried searching for variable and function names and even bits of comments to see if they're copied from the training data. They're not!

Re: Try to guess if code is real or GPT2-generated

#13

Post by ivraatiems »

moyix wrote: post #11

Hi, author here! Some details on the model: * Trained 17GB of code from the top 10,000 most popular Debian packages. The source files were deduplicated using a process similar to the OpenWebText preprocessing (basically a locality-sensitive hash to detect near-duplicates). * I used the [Megatron-LM]( https://github.com/NVIDIA/Megatron-LM ) code for training. Training took about 1 month on 4x RTX8000 GPUs. * You can d…

Thanks for stopping by! This is impressive. I would be curious to know if my hunch below about potential weaknesses/tells was at all correct.

Did people find it to be as challenging when you showed it to them as some of us are here? Did you expect that level of complexity?

Re: Try to guess if code is real or GPT2-generated

#14

Post by et1337 »

This looks like overfitting to me. Some of the GPT samples were definitely real code, or largely real code. One looked like something from Xorg, another like it was straight from the COLLADA SDK. It’s really hard to define what “truly new code” is, if it’s just the same code copy pasted in different order. Blah blah Ship of Theseus etc.

Re: Try to guess if code is real or GPT2-generated

#16

Post by moyix »

ivraatiems wrote: post #13
moyix wrote: post #11

Hi, author here! Some details on the model: * Trained 17GB of code from the top 10,000 most popular Debian packages. The source files were deduplicated using a process similar to the OpenWebText preprocessing (basically a locality-sensitive hash to detect near-duplicates). * I used the [Megatron-LM]( https://github.com/NVIDIA/Megatron-LM ) code for training. Training took about 1 month on 4x RTX8000 GPUs. * You can d…

Thanks for stopping by! This is impressive. I would be curious to know if my hunch below about potential weaknesses/tells was at all correct. Did people find it to be as challenging when you showed it to them as some of us are here? Did you expect that level of complexity?

There are likely some "tells" but many fewer of them than I expected. I've seen it occasionally generate something malformed like "#includefrom", and like all GPT2 models it has a tendency to repeat things.

Yes, I think people definitely find it challenging. I'm keeping track of the correct and total guesses for each snippet, right now people are at almost exactly 50% accuracy:

     correct | total | pct 
    ---------+-------+-----    
        6529 | 12963 |  50

Re: Try to guess if code is real or GPT2-generated

#17

Post by moyix »

et1337 wrote: post #14

This looks like overfitting to me. Some of the GPT samples were definitely real code, or largely real code. One looked like something from Xorg, another like it was straight from the COLLADA SDK. It’s really hard to define what “truly new code” is, if it’s just the same code copy pasted in different order. Blah blah Ship of Theseus etc.

The generated snippets are prompted with 128 characters from real code (but not code from the training data), so they can often pick up on the name of the project etc.

Re: Try to guess if code is real or GPT2-generated

#18

Post by hntrader »

moyix wrote: post #16
ivraatiems wrote: post #13

Earlier quoted context omitted.

Thanks for stopping by! This is impressive. I would be curious to know if my hunch below about potential weaknesses/tells was at all correct. Did people find it to be as challenging when you showed it to them as some of us are here? Did you expect that level of complexity?

There are likely some "tells" but many fewer of them than I expected. I've seen it occasionally generate something malformed like "#includefrom", and like all GPT2 models it has a tendency to repeat things. Yes, I think people definitely find it challenging. I'm keeping track of the correct and total guesses for each snippet, right now people are at almost exactly 50% accuracy: correct | total | pct ---------+-------…

Are you presenting real samples and GPT2 samples to users with equal probabilities?

EDIT another poster guessed GPT2 each time and found the frequency was 80 percent

Re: Try to guess if code is real or GPT2-generated

#19

Post by lostmsu »

moyix wrote: post #11

Hi, author here! Some details on the model: * Trained 17GB of code from the top 10,000 most popular Debian packages. The source files were deduplicated using a process similar to the OpenWebText preprocessing (basically a locality-sensitive hash to detect near-duplicates). * I used the [Megatron-LM]( https://github.com/NVIDIA/Megatron-LM ) code for training. Training took about 1 month on 4x RTX8000 GPUs. * You can d…

Can you share the dataset too?
Post reply on HN