https://github.com/distributive-network/tfjs_util
https://github.com/distributive-network/tfjs_util
Last synced: 3 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/distributive-network/tfjs_util
- Owner: Distributive-Network
- Created: 2020-09-29T13:16:34.000Z (over 4 years ago)
- Default Branch: master
- Last Pushed: 2021-06-28T20:59:22.000Z (almost 4 years ago)
- Last Synced: 2025-01-10T19:47:54.111Z (4 months ago)
- Language: JavaScript
- Size: 14.3 MB
- Stars: 0
- Watchers: 4
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# TFJS Utils
## Serialization of TFJS model.json and bins
The following script will generate and publish a dcp module with the serialized tfjs model.
```bash
./bin/serializeModel.js -m $INPUT_MODEL.json -o $MODULENAME/$FILENAME -p 1.0.0 -d
```
If you remove the -d flag, you will have a node compliant module that will be saved at $OUTPUTFILE location.
Additionally, you can specify which identity key and bank keystore you will be using by using the following command:
```bash
./bin/serializeModel.js -I $PATHTOID.keystore --default-bank-account-file $PATHTOBANKACCOUNT.keystore -m $INPUT_MODEL.json -o $OUTPUTFILE.js/$MODULENAME -p 1.0.0 -d
```
Due to certain current restrictions in the package manager, it is important to also mark the version number using the `-p` command. If you have tried to publish before and are getting versioning errors, please increase the version number in the `-p` flag.
## Large Models
For large models, (any model larger than 60MB) can be uploaded using the sharding flag. It's important to note that sharding will always publish to dcp as sharding has no use in local files. The following command is an example of sharding:
```bash
./bin/serializeModel.js -m $INPUT_MODEL.json -o $MODULENAME/$FILENAME -p 1.0.0 -d -s
```
## Using the module
Once uploaded, here is how you'll require your model:
```js
let job = compute.for(INPUT_SET, async function(){
let tf = require('tfjs');
let { getModel } = require($FILENAME);
let model = await getModel();
...
});
job.requires('aistensorflow/tfjs');
job.requires('$MODULENAME/$FILENAME');
...
```