huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-16 10:38:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Print some per-step timings in stable-diffusion. (#520)

Browse Source 
* Skeleton files for neon support of quantization.

* SIMD version for q4 vecdot.

* Also simdify the q6k multiplication.

* Add some timings to stable-diffusion.
This commit is contained in:
Laurent Mazare
2023-08-20 05:45:12 +01:00
committed by GitHub
parent 82410995a2
commit aa207f2dd9
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
candle-examples/examples/stable-diffusion/main.rs
View File

@ -313,8 +313,9 @@ fn run(args: Args) -> Result<()> {
// scale the initial noise by the standard deviation required by the scheduler
latents = (latents * scheduler.init_noise_sigma())?;
println!("starting sampling");
for (timestep_index, &timestep) in scheduler.timesteps().iter().enumerate() {
println!("Timestep {timestep_index}/{n_steps}");
let start_time = std::time::Instant::now();
let latent_model_input = Tensor::cat(&[&latents, &latents], 0)?;
let latent_model_input = scheduler.scale_model_input(latent_model_input, timestep)?;
@ -325,6 +326,8 @@ fn run(args: Args) -> Result<()> {
let noise_pred =
(noise_pred_uncond + ((noise_pred_text - noise_pred_uncond)? * GUIDANCE_SCALE)?)?;
latents = scheduler.step(&noise_pred, timestep, &latents)?;
let dt = start_time.elapsed().as_secs_f32();
println!("step {}/{n_steps} done, {:.2}s", timestep_index + 1, dt);
if args.intermediary_images {
let image = vae.decode(&(&latents / 0.18215)?)?;