Aston Zhang

icon indicating a website link https://astonzhang.com/

icon company Meta icon location Menlo Park icon location Building large language models on the Llama Team at Meta Generative AI

Results 70 comments of Aston Zhang

Default value for "training" parameter for BatchNorm custom layer call method

Thanks. Have you checked the [refactored implementations](http://preview.d2l.ai.s3-website-us-west-2.amazonaws.com/d2l-en/master/chapter_convolutional-modern/batch-norm.html)?

The `devices` argument in the TF implementation of d2l.predict_seq2seq

Thanks. As long as the current implementation is correct (@terrytangyuan please double check), I'm ok with leaving it out in the argument list for now.

MXNet/PT Non Ideal Results NiN

@StevenJokes plz see my comments https://github.com/d2l-ai/d2l-en/pull/1176

Hope to implement the code of d2l based on paddlepaddle framework

Thanks! Is it this repo: https://github.com/tngt/Dive-into-DL-paddlepaddle/tree/master/docs Currently we've been revising the API for the entire book: http://preview.d2l.ai.s3-website-us-west-2.amazonaws.com/d2l-en/master/chapter_linear-networks/api.html Could you see if could adapt using the new API? Besides, we have...

TF implementation of RNN-scratch runs much slower than the other implementations

Thanks. RNN scratch does trains in one GPU for all the frameworks so I guess this is probably not the root cause. Maybe @terrytangyuan may help you with your PR...

[MXNet] Speedup SSD Scratch Implementation

Can you fix it?

[WIP] integrating changes to modern rnn chapter

Thanks! Per offline discussions we'll continue with our flexible approach :)

Jacobian matrix

Thanks. After checking we'll keep the original :)

Unify hyperparameters of all frameworks in DCGAN

@biswajitsahoo1111 Could you take a look at it? Thanks.

Sample weighting description + Example

Thanks. There are many topics that we can cover. As a book focusing on DL, currently we intentionally keep the first few foundational chapters *minimal* so readers can quickly get...