got consistent 3 fps

eyaler · Oct 20, 2024 · 16ca967 · 16ca967
1 parent 4d22af5
commit 16ca967
Show file tree

Hide file tree

Showing 10 changed files with 98 additions and 83 deletions.
diff --git a/README.md b/README.md
@@ -2,12 +2,12 @@
 
 ## This fork ([demo](https://eyaler.github.io/webnn-developer-preview/))
 1. Added sd-turbo [image-to-image](demos/sd-turbo/index.js) mode using the [vae encoder](https://huggingface.co/eyaler/sd-turbo-webnn/tree/main/vae_encoder) to initialize latents
-2. Added camera input option for the above (I can get up to 2 fps with an RTX 3070 TI laptop GPU on Windows)
+2. Added camera input option for the above (I can get up to 3 fps with an RTX 3070 Ti laptop GPU on Windows)
 3. Added a helper [script](demos/sd-turbo/fix_instance_norm.py) for casting instance-normalization nodes to float32 (which has critical numerical issues with float16)
 4. Removed sd-turbo safety checks ;)
 
 ### Known issues:
-1. Run times of VAE encoder and UNET are inconsistent and are sometimes slower by 10x
+1. Run times of VAE encoder and UNET are inconsistent and are sometimes slower by 15x
 
 ### Example:
 

diff --git a/assets/js/common_utils.js b/assets/js/common_utils.js
@@ -1,13 +1,14 @@
 const KNOWN_COMPATIBLE_CHROMIUM_VERSION = {
-  'stable-diffusion-1.5': '129.0.6617.0',
-  'sd-turbo': '129.0.6617.0',
-  'segment-anything': '129.0.6617.0',
-  'whisper-base': '129.0.6617.0',
-  'image-classification': '129.0.6617.0'
+  'stable-diffusion-1.5': ['129.0.6617.0'],
+  'sd-turbo': ['129.0.6617.0', '130.0.6723.59'],
+  'segment-anything': ['129.0.6617.0'],
+  'whisper-base': ['129.0.6617.0'],
+  'image-classification': ['129.0.6617.0']
 }
 
-export const showCompatibleChromiumVersion = (key) => {
-  const version = KNOWN_COMPATIBLE_CHROMIUM_VERSION[key]
+export const showCompatibleChromiumVersion = (key, branch) => {
+  const flag = (branch && !KNOWN_COMPATIBLE_ORT_VERSION[key][branch].startsWith('1.19') && KNOWN_COMPATIBLE_CHROMIUM_VERSION[key].length > 1) | 0
+  const version = KNOWN_COMPATIBLE_CHROMIUM_VERSION[key][flag]
   if (version) {
     const chromiumVersionElement = document.querySelector('#chromiumversion');
     chromiumVersionElement.innerHTML = `Known compatible Chromium version: 
@@ -123,13 +124,12 @@ export const getTime = () => {
 };
 
 const KNOWN_COMPATIBLE_ORT_VERSION = {
-  'stable-diffusion-1.5': { 'dev': '1.19.0-dev.20240804-ee2fe87e2d', 'stable': '1.20.0', 'test': 'test' },
-  'sd-turbo': { 'dev': '1.19.0-dev.20240804-ee2fe87e2d', 'stable': '1.20.0', 'test': 'test' },
-  //'sd-turbo': { 'dev': '1.19.2', 'stable': '1.20.0', 'test': 'test' },
-  //'sd-turbo': { 'dev': '1.20.0-dev.20240928-1bda91fc57', 'stable': '1.20.0', 'test': 'test' },
-  'segment-anything': { 'dev': '1.19.0-dev.20240804-ee2fe87e2d', 'stable': '1.20.0', 'test': 'test' },
-  'whisper-base': { 'dev': '1.19.0-dev.20240804-ee2fe87e2d', 'stable': '1.20.0', 'test': 'test' },
-  'image-classification': { 'dev': '1.19.0-dev.20240804-ee2fe87e2d', 'stable': '1.20.0', 'test': 'test' },
+  'stable-diffusion-1.5': { 'dev': '1.20.0-dev.20240919-bd60add8ce', 'stable': '1.20.0', 'test': 'test' },
+  'sd-turbo': { 'dev': '1.20.0-dev.20240919-bd60add8ce', 'stable': '1.20.0', 'test': 'test' },
+  //'sd-turbo': { 'dev': '1.21.0-dev.20241019-abad69b322', 'stable': '1.20.0', 'test': 'test' },
+  'segment-anything': { 'dev': '1.20.0-dev.20240919-bd60add8ce', 'stable': '1.20.0', 'test': 'test' },
+  'whisper-base': { 'dev': '1.20.0-dev.20240919-bd60add8ce', 'stable': '1.20.0', 'test': 'test' },
+  'image-classification': { 'dev': '1.20.0-dev.20240919-bd60add8ce', 'stable': '1.20.0', 'test': 'test' },
 };
 
 const ORT_BASE_URL = 'https://www.npmjs.com/package/onnxruntime-web/v/';
@@ -187,7 +187,7 @@ export const setupORT = async (key, branch) => {
   ortVersionElement.innerHTML = versionHtml;
 };
 
-export const webNnStatus = async () => {
+export const getWebnnStatus = async () => {
   let result = {};
   try {
     const context = await navigator.ml.createContext();

diff --git a/demos/image-classification/index.js b/demos/image-classification/index.js
@@ -10,7 +10,7 @@ import {
   log,
   logError,
   getQueryValue,
-  webNnStatus,
+  getWebnnStatus,
   updateQueryStringParameter,
   getMedian,
   getAverage,
@@ -198,7 +198,7 @@ const main = async () => {
 };
 
 const checkWebNN = async () => {
-  let webnnStatus = await webNnStatus();
+  let webnnStatus = await getWebnnStatus();
 
   if (
     getQueryValue("provider") &&

diff --git a/demos/sd-turbo/fix_instance_norm.py b/demos/sd-turbo/fix_instance_norm.py
@@ -3,13 +3,16 @@
 import sys
 
 
-assert len(sys.argv) == 3, 'Syntax: python fix_instance_norm.py INPUT_TXT_OR_ONNX OUTPUT_TXT_OR_ONNX'
+assert len(sys.argv) >= 3, 'Syntax: python fix_instance_norm.py INPUT_TXT_OR_ONNX OUTPUT_TXT_OR_ONNX [location of onnx2text]'
 input_file = sys.argv[1]
 output_file = sys.argv[2]
+onnx2text = 'onnx2text'
+if len(sys.argv) > 3:
+    onnx2text = os.path.join(sys.argv[3], onnx2text)
 
 if os.path.splitext(input_file)[-1] == '.onnx':
     converted_input_file = input_file + '.txt'
-    os.system(f'onnx2text "{input_file}" "{converted_input_file}"')
+    os.system(f'{onnx2text} "{input_file}" "{converted_input_file}"')
     input_file = converted_input_file
 
 converted_output_file = None
@@ -49,7 +52,7 @@
 total = 0
 constants = []
 for i, part in enumerate(parts[1 : -1], start=1):
-    part, subs = re.subn('(.*?input: ")([^"]*?)(_output[^"]*)?(".*?output: ")([^"]*?)(_output[^"]*)?(".*op_type: "InstanceNormalization".*)', replace, part, flags=re.DOTALL)
+    part, subs = re.subn('(.*?input: ")([^"]*?)(_output[^"]*)?(".*?output: ")([^"]*?)(_output[^"]*)?((?<!_tocast)".*op_type: "InstanceNormalization".*)', replace, part, flags=re.DOTALL)
     if subs:
         constants += re.findall('".*constant.*"', part, flags=re.IGNORECASE)
         parts[i] = part
@@ -69,4 +72,4 @@
     f.write(''.join(parts))
 
 if converted_output_file:
-    os.system(f'onnx2text "{output_file}" "{converted_output_file}"')
+    os.system(f'{onnx2text} "{output_file}" "{converted_output_file}"')