e0157817eb
Extract test utilities to separate file
2023-04-19 20:47:58 +02:00
abbbe3fab5
Fix `bert_prepare_model_inputs` when tokenizer returns array
2023-04-19 20:47:01 +02:00
a3ed4e3f18
Update `GenerationConfig` documentation
2023-04-19 17:48:00 +02:00
47b256a92a
Remove unnecessary inheritance from `Function` class
2023-04-18 17:10:09 +02:00
aba491e942
[version] Set version to 1.5.0.dev
2023-04-18 17:09:29 +02:00
4f158a5ad5
Move math utility functions to `math_utils.js`
2023-04-17 21:43:30 +02:00
cfa37d3c92
Merge `max` and `indexOfMax` functions, and return tuple of results
2023-04-17 21:27:19 +02:00
9bfd126c84
Update package-lock.json
2023-04-17 15:33:53 +02:00
d36ba315ec
Merge branch 'main' into docs
2023-04-17 15:28:57 +02:00
8330870752
Remove redundant `resolve.fallback` to webpack config
...
This is already handled by the "browser" option in package.json
2023-04-15 22:33:31 +02:00
87e121c76e
Use `sharp.js` instead of `node-canvas` for resizing images with node.js ( Fixes #85 )
...
node-canvas has problems with installation, and doesn't support bilinear/bicubic interpolation.
sharp.js is also much faster: https://github.com/ivanoff/images-manipulation-performance
2023-04-15 22:30:49 +02:00
421aa331cd
Replace CommonJS imports/exports with ES6
2023-04-13 21:01:44 +02:00
1b9fe0f27c
Replace `Math.max` with iterative version in `softmax` function ( #83 )
...
TODO: In future, replace all min/max/indexOfMin/indexOfMax functions with functions that return index and value of the minimum/maximum at the same time (like PyTorch).
2023-04-12 03:35:06 +02:00
ee1056bb71
[version] Update to v1.4.2
2023-04-11 02:48:24 +02:00
3b9ebf0a08
Add documented support for `image-segmentation` task
2023-04-11 02:44:30 +02:00
c4c0df5696
Replace `Jimp` with `canvas` for loading images
...
For the following reasons:
1. Size - Jimp is a large package, making it less suitable for use in the browser. Most of the functions aren't even used by this library.
2. Speed - According to https://github.com/ivanoff/images-manipulation-performance , `node-canvas` is around 9.3x faster than Jimp.
3. `Image` class - Their `Image` class did not support easily converting between colour formats (RGB vs. RGBA vs L), which is needed for mask images (for example).
The main downside is that when resizing, the interpolation method isn't able to be specified (e.g., CLIP using bicubic). We aim to fix this in the future.
2023-04-11 02:09:43 +02:00
fdfdd625a8
Move `interpolate` and `transpose` method logic from tensor_utils.js to math_utils.js
2023-04-11 01:32:58 +02:00
c6f5f16722
Allow filePath to be a URL
2023-04-11 01:30:17 +02:00
b1162f310a
Exclude imports that use the `node:` protocol when building for the web
2023-04-09 23:39:10 +02:00
6cf12df3c2
Create image segmentation test
2023-04-09 23:34:03 +02:00
46c66b81b7
Create `ImageSegmentationPipeline`
2023-04-09 23:33:28 +02:00
a19b498f7e
Set return type of `get` function
2023-04-09 23:30:49 +02:00
3b3711643e
Allow untyped array to be used in Tensor constructor
2023-04-09 23:30:27 +02:00
34ed29e493
Add `post_process_panoptic_segmentation` function to `DetrFeatureExtractor`
2023-04-09 23:29:18 +02:00
b986d41bc2
Expose `Tensor` class from transformers.js
2023-04-09 21:58:53 +02:00
b3efaefb1e
Add `interpolate` and `sigmoid` functions to tensor_utils.js
2023-04-09 21:56:32 +02:00
3348904138
Add image segmentation model and output classes
2023-04-09 20:44:41 +02:00
f529ada1ab
Add `facebook/detr-resnet-50-panoptic` to supported models
2023-04-09 20:32:09 +02:00
d862626bf0
Merge pull request #72 from ekolve/node-readable-stream
...
Define ReadableStream from node:stream/web if it doesn't exist
2023-04-08 17:18:12 +02:00
f0e47a8536
Fix `ClipTokenizer` ( Closes #74 )
2023-04-08 13:19:19 +02:00
c3f853b488
Use custom build of catharsis for parsing JSDoc
...
Fixes a few issues in their parser
Fork:
https://github.com/xenova/catharsis
Original:
https://github.com/hegemonic/catharsis
2023-04-07 23:31:58 +02:00
eb574bb64f
Add `npm run docs` command to generate docs
2023-04-07 16:57:26 +02:00
a38a77d8c2
Update types for tokenizers.js
2023-04-07 16:53:53 +02:00
b127ceed74
Update `Tensor` JSDoc
2023-04-07 16:53:21 +02:00
606de46947
Update `seq2seqLoadModel` JSDoc
2023-04-07 16:52:50 +02:00
c76374fee8
Add `jsdoc-to-markdown` as dev dependency
2023-04-07 16:51:59 +02:00
b7d964d4e9
Update `post_process_object_detection` JSDoc
2023-04-07 16:51:33 +02:00
47c367458c
Update list of supported text2text-generation models
2023-04-07 14:13:56 +02:00
f611db0917
Update src/utils.js
...
Co-authored-by: Joshua Lochner <admin@xenova.com>
2023-04-06 15:05:10 -07:00
05bfddba74
defining ReadableStream from node:stream/web if it doesn't exist
2023-04-06 14:31:49 -07:00
db26ba1dea
[version] Update to v1.4.1
2023-04-06 22:13:56 +02:00
ab8b510e33
Add documented support for token classification
2023-04-06 21:52:30 +02:00
915e5fbde9
Do not process special tokens in token classification pipeline
2023-04-06 21:48:20 +02:00
97720512eb
[demo] Create token-classification demo (named entity recognition)
2023-04-06 21:47:31 +02:00
41c7519046
Fix Bert pretokenizer regex
2023-04-06 21:40:01 +02:00
12163dd28b
Fix tokenization for multilingual Helsinki-NLP models ( #70 )
2023-04-06 17:36:52 +02:00
682f7a1832
[demo] Update worker.js comments
2023-04-05 21:46:41 +02:00
db7678c3a8
[demo] Store promise to pipeline in factory
2023-04-05 21:37:18 +02:00
f3f87eba5d
Add supported token classification models
2023-04-05 21:27:11 +02:00
b6483cc7f5
Extract supported marian models to extra folder
2023-04-05 21:24:11 +02:00
Joshua Lochner