Comments (5)
rust code for benching:
use dfdx::prelude::*;
use rand::{prelude::StdRng, SeedableRng};
use rand_distr::StandardNormal;
use std::time::{Duration, Instant};
fn main() {
let mut rng = StdRng::seed_from_u64(0);
let mut l: Linear<512, 256> = Default::default();
l.randomize(&mut rng, &StandardNormal);
let mut opt = Adam::default();
const N: usize = 10000;
let mut total = Duration::default();
for _ in 0..N {
let x: Tensor2D<32, 512> = Tensor2D::randn(&mut rng);
let y = l.forward(x.traced());
let loss = y.square().mean();
let start = Instant::now();
let gradients = loss.backward();
opt.update(&mut l, gradients);
total += start.elapsed();
}
println!("{:?} batch per s", N as f32 / total.as_secs_f32());
}
from dfdx.
Python code for benching:
from datetime import datetime, timedelta
import torch
torch.manual_seed(0)
l = torch.nn.Linear(512, 256)
opt = torch.optim.Adam(l.parameters())
total = timedelta()
N = 10000
for _ in range(N):
x = torch.randn(32, 512)
y = l(x)
loss = y.square().mean()
start = datetime.now()
opt.zero_grad()
loss.backward()
opt.step()
total += datetime.now() - start
print(N / total.total_seconds())
from dfdx.
Use https://crates.io/crates/criterion
from dfdx.
Both dfdx version and torch version should use flush to zero (#60)
from dfdx.
Closing - might do in future, but this will continue to be ad-hoc for now
from dfdx.
Related Issues (20)
- Alloc zero size memory on old model GPU may fail.
- Different results when CPU feature is on vs off HOT 2
- Unnecessary loss of precision when computing loss functions HOT 2
- trait TryConcatAlong not satisfied when using constants HOT 1
- Consider helpers for accessing tensors from tuples and input wrappers HOT 1
- Question / clarification regarding heap allocations HOT 2
- Examples or resources for autodiff with 2 networks?
- Bug: `Sequential` macro provide `forward_mut` as `forward`
- Replace explicit features and paths on generated code
- Send/Sync for Device HOT 1
- Add `OUTPUT_PADDING` to `ConvTrans2D`
- Split `TryConcatAlong` into different traits
- Add `Prodigy` optimizer HOT 1
- Run tests with miri HOT 1
- Reduce test sizes HOT 1
- Unclear how to handle error type in `dfdx::nn::LoadFromNpz::load`
- Add `nn::AdaptiveAvgPool2D`
- How does one update one model from another model? HOT 1
- Unable to build with old CUDA version (`CUDA_COMPUTE_CAP = 52`)
- OpenXLA Support HOT 3
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from dfdx.