You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+3-3Lines changed: 3 additions & 3 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -83,7 +83,7 @@ Besides supporting WebGPU, this project also provides the harness for other kind
83
83
./scripts/prep_deps.sh
84
84
```
85
85
86
-
The last thing to do is setting up the site with
86
+
The last thing to do is setting up the site with:
87
87
```shell
88
88
./scripts/local_deploy_site.sh
89
89
```
@@ -97,7 +97,7 @@ Besides supporting WebGPU, this project also provides the harness for other kind
97
97
98
98
## How
99
99
100
-
The key technology here is machine learning compilation (MLC). Our solution builds on the shoulders of the open source ecosystem, including Hugging Face, model variants from LLaMA and Vicuna, wasm and WebGPU. The main flow builds on Apache TVM Unity, an exciting ongoing development in the [Apache TVM Community](https://github.com/apache/tvm/)
100
+
The key technology here is machine learning compilation (MLC). Our solution builds on the shoulders of the open source ecosystem, including Hugging Face, model variants from LLaMA and Vicuna, wasm and WebGPU. The main flow builds on Apache TVM Unity, an exciting ongoing development in the [Apache TVM Community](https://github.com/apache/tvm/).
101
101
102
102
- We bake a language model's IRModule in TVM with native dynamic shape support, avoiding the need of padding to max length and reducing both computation amount and memory usage.
103
103
- Each function in TVM’s IRModule can be further transformed and generate runnable code that can be deployed universally on any environment that is supported by minimum tvm runtime (JavaScript being one of them).
@@ -129,7 +129,7 @@ WebGPU works by translating WGSL shaders to native shaders. We observed that the
129
129
130
130
Some of the current gaps are caused by Chrome's WebGPU implementation inserts bound clips for all array index access, such that `a[i]` becomes `a[min(i, a.size)]`. This can be optimized out as the WebGPU support continues to mature.
131
131
132
-
You can get around this by using a special flag to launch Chrome (thanks to Dawn developers forproviding the pointers), by exiting Chrome completely, thenincommand line, type
132
+
You can get around this by using a special flag to launch Chrome (thanks to Dawn developers forproviding the pointers), by exiting Chrome completely, thenincommand line, type:
0 commit comments