-
Notifications
You must be signed in to change notification settings - Fork 3.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[BYORTL][Verilator] update ops and add MobileNet #7972
Conversation
Thanks @vegaluisjose . just want to make sure that the mobile net simulation won't cost a long time to finish. otherwise it would be still good to have unit test models that can finish in a reasonable amount of time. We can move mobile net tests to nightly category later once we support nightlies |
Thank you @vegaluisjose the changes LGTM. To alleviate @tqchen 's concerns can you comment on how long the simulations take on MobileNet? We're only simulating bias_add on a small RTL design so I presume the runtime won't be as long as some of the other TSIM tests on full VTA designs offloading entire conv2ds but let's make sure we keep tabs on runtime. Thanks! |
It takes about the same time as running the pure software version. |
I tested the two tests on my local machine, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks @vegaluisjose ; I left a nit that would be best to address if you have the bandwidth.
predictions = np.squeeze(res) | ||
prediction = np.argmax(predictions) | ||
# 387 is the elephant | ||
tvm.testing.assert_allclose(prediction, 387, rtol=1e-5, atol=1e-5) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
In this case can't we just assert that prediction is equal to 387?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Fixed
Weird, it looks like we got a CI error, but probably unrelated to the changes. Could you re-trigger @vegaluisjose ? |
Thank you @vegaluisjose the PR has been merged. |
* update * update vta submodule * cpp fmt * python fmt * skip if tflite is not available * fmt * change assertion * update comment
* update * update vta submodule * cpp fmt * python fmt * skip if tflite is not available * fmt * change assertion * update comment
The following PR adds support to test end-to-end(sw/hw) MobileNet using the Verilator backend. This is build upon this commit.
@tmoreau89