Xenova @xenovacom Twitter profile | Pikagi

Pikagi

Xenova

@xenovacom

6,322

Followers

286

Following

196

Media

845

Statuses

Bringing the power of machine learning to the web. Currently working on Transformers.js ( @huggingface 🤗)

https://t.co/n0HmaM7Wsb

Joined December 2015

Don't wanna be here? Send us removal request.

Pinned Tweet

@xenovacom

Xenova

1 year

I'm excited to announce that I have joined @huggingface ! 🤗 Continuing with Transformers.js, I aim to grow the community by bridging the gap between web development and machine learning. Thank you to everyone who has supported me so far, I can't wait to show you what's next! 🚀

29

28

302

Last Seen Profiles

@Grenhitya

@OvertakeCentral

@AsliEnverResmi

@t_mof34

@MadridD2i

@barcacenIre

@_matthewj27_

@WanaroonKandis

@A5hu_

@ThePrince_59

@Laiamape

@Y__D__D

@MichelleLe14143

@jomareewade

@jEJioodP1am8FGx

@ai90om

@LinnetsLadiesFC

@CersosimoN44178

@PereFontanals

@manoonxiv

@MarkLandler

@yonafraeufaeioe

@GrippedMagazine

@ho0782612645191

@abbyliv817

@malaikaterryy

@SAGU_Softball

@RAW_JEI

@CooperSusa67381

@Hahahah89245701

@UCSF_IR

@HoodCountyTX

@kldbnsln1

@witchslvt

@rinkaenbyou4TH

@SHEsus__Christ

@xenovacom

Xenova

11 months

Introducing Whisper Web: ML-powered speech recognition directly in your browser! 🚀 This comes with the release of Transformers.js v2.2.0, which now supports multilingual transcription and translation for over 100 different languages! 🤯 Check it out:

24

257

1K

@xenovacom

Xenova

22 days

Meta's Segment Anything Model (SAM) can now run in your browser w/ WebGPU (+ fp16), meaning up to 8x faster image encoding (10s → 1.25s)! 🤯⚡️ Video is not sped up! Everything runs 100% locally thanks to 🤗 Transformers.js and onnxruntime-web! 🔗 Demo:

17

194

1K

@xenovacom

Xenova

2 months

We just updated our in-browser Background Removal demo to use WebGPU and it's now ~50x faster! 🤯 ~9 seconds down to 180ms! ⚡️ Powered by @bria_ai_ 's RMBG-v1.4 model and 🤗 Transformers.js! ... and yes, the video is in real time! 🤯

22

113

667

@xenovacom

Xenova

3 months

New @karpathy video just dropped! 😍🔥 After watching, if you want to learn more about how different models (e.g., GPT4, Llama, T5, BERT) tokenize text, check out "The Tokenizer Playground": a web-app I built a few months ago with 🤗 Transformers.js! 🔗

@karpathy

Andrej Karpathy

3 months

New (2h13m 😅) lecture: "Let's build the GPT Tokenizer" Tokenizers are a completely separate stage of the LLM pipeline: they have their own training set, training algorithm (Byte Pair Encoding), and after training implement two functions: encode() from strings to tokens, and…

Tweet media one

385

2K

14K

3

106

640

@xenovacom

Xenova

6 months

Introducing Distil-Whisper Web: 49% smaller, 4.2x faster Whisper directly in your browser! 🚀 Here is a side-by-side comparison with OpenAI's original version! 🤯

16

88

573

@xenovacom

Xenova

2 months

WebGPU is the future! 😍🔥 Transformers.js can now perform real-time background removal, powered by MODNet! ⚡️ Development for Transformers.js v3 (which adds full WebGPU support) is well underway, and we're excited to continue sharing updates and demos! Try it out yourself! 👇

13

88

535

@xenovacom

Xenova

20 days

I know it just released, but I don't see many people talking about the Phi-3 tokenizer! 👀 Here's the full list of added special tokens... what do you notice? 🤯 <|assistant|> <|step|> <|function_output|> <|tag|> <|function_call|> <|system|> <|end|> <|raw|> <|continue|> <|user|>…

Tweet media one

18

71

515

@xenovacom

Xenova

8 months

Massive size reductions for Transformers.js models: - Whisper-tiny decoder: 50MB → 30MB (-40%) - NLLB decoder: 732MB → 476MB (-35%) - bloom: 819MB → 562MB (-31%) - gpt2: 167MB → 127MB (-24%) - bert-base: 134MB → 111MB (-17%) ... all this with ZERO drop in performance! 🤯

3

86

467

@xenovacom

Xenova

3 months

YOLOv9 just released, and now it's compatible with 🤗 Transformers.js! That's right... near real-time object detection running locally in your browser: no server required! 🤯 Try it out yourself! 👇 🔗 Demo:

7

81

457

@xenovacom

Xenova

8 months

When do you *really* need to use a vector database? 🤔 To try answer that question, I recreated my semantic image search application to run 100% in-browser with Transformers.js (no server). After loading the model and database, it only takes ~50ms to compute text embeddings and…

@karpathy

Andrej Karpathy

1 year

@sinclanich np.array people keep reaching for much fancier things way too fast these days

20

95

1K

33

61

432

@xenovacom

Xenova

1 year

Transformers.js v2.0 is finally here! 🔥 Run @huggingface transformers directly in your browser, with no need for a server! Some of the new features include: 🛠️ Complete ES6 rewrite 📄 Documentation + examples 🤗 Improved Hugging Face Hub integration

Tweet card media

GitHub - xenova/transformers.js: State-of-the-art Machine Learning for the web. Run 🤗 Transformers...

State-of-the-art Machine Learning for the web. Run 🤗 Transformers directly in your browser, with no need for a server! - xenova/transformers.js

7

91

405

@xenovacom

Xenova

3 months

Here's a side-by-side comparison of the GPT-4, Gemma, and Llama tokenizers, tested on "The Great Gatsby" (270k characters). As @karpathy points out, the Gemma and Llama tokenizers are very similar, with the main difference being vocabulary size. One interesting thing to see is…

Tweet media one

@karpathy

Andrej Karpathy

3 months

Seeing as I published my Tokenizer video yesterday, I thought it could be fun to take a deepdive into the Gemma tokenizer. First, the Gemma technical report [pdf]: says: "We use a subset of the SentencePiece tokenizer (Kudo and Richardson, 2018) of…

184

476

5K

13

68

390

@xenovacom

Xenova

4 months

Today we released 🤗 Transformers.js v2.14, which adds support for SAM (Segment Anything Model). This means you can now generate high-quality segmentation masks for objects in a scene, directly in your browser! 🤯 Demo (+ source code):

8

80

377

@xenovacom

Xenova

2 months

Grok-1 is finally out! 😍 But while everyone was focused on the weights, I decided to take a look at the tokenizer. I also added it to the Tokenizer Playground! Structurally, it looks quite similar to the Llama 2 tokenizer (BPE w/ byte-fallback), with a vocabulary size of 2¹⁷ =…

Tweet media one

8

77

358

@xenovacom

Xenova

8 months

WOW! 🤯 An in-browser version of ChatGPT (or HF Chat), built with 🤗 Transformers.js! Yes that's right, everything runs 100% locally in your browser, meaning no need for a server! Check it out! 🔗

@dhuynh95

Daniel Huynh

8 months

🚀Just released #BlindChat : an open-source & privacy-first ChatGPT alternative! 🔒BlindChat combines @huggingface transformers.js (from the great @xenovacom ) with #ChatUI for a private and fully in-browser experience Try it on HF:

10

41

225

10

70

353

@xenovacom

Xenova

5 months

Distil-Whisper small is finally here! 🔥 Over 10x smaller, 5x faster, and within 3% WER of large-v2. 🤯 Since it's only 166M params, it can even run locally in your browser with 🤗 Transformers.js! Check it out! 👉

9

61

356

@xenovacom

Xenova

3 months

Real-time object detection w/ 🤗 Transformers.js, running YOLOv9 directly in your browser! 🤯 This demo shows why on-device ML is so important: 1. Privacy - local inference means no user data is sent to the cloud 2. No server latency - empowers developers to build real-time…

9

63

347

@xenovacom

Xenova

11 months

Due to popular demand, we added microphone support to Whisper Web! 🎙️ This means you can now record and transcribe audio directly in your browser: no installation required. 😍 Demo: Source code:

7

81

334

@xenovacom

Xenova

7 months

We just released 🤗 Transformers.js v2.7.0, which adds supports for 🗣️ Text to Speech w/ speecht5. This means you can now synthesize human speech directly in your browser... no server required! 🔥 Check out the demo! 👇

15

74

322

@xenovacom

Xenova

3 months

Introducing Remove Background Web: In-browser background removal, powered by @bria_ai_ 's new RMBG-v1.4 model and 🤗 Transformers.js! That's right, everything runs 100% locally, meaning none of your images are uploaded to a server! 🤯 Try it out: 👇

10

64

320

@xenovacom

Xenova

6 months

Transformers.js v2.9.0 is now out! 😍 New features: 🎯 Zero-shot Object Detection w/ OwlViT 🕵️‍♂️ Depth Estimation w/ DPT and GLPN 📝 Optical Document Understanding w/ Nougat ... and you can get started in just a few lines of code! 🤯👇

Tweet media one

6

57

314

@xenovacom

Xenova

10 months

Calling all JS developers! We just released 2 example Next.js applications which show how to use Transformers.js for client-side (in-browser) or server-side (Node.js) inference. 🤗 Building full-stack AI applications has never been this easy! 🚀 Tutorial:…

5

71

312

@xenovacom

Xenova

7 months

Woah! 🤯 A new 20M parameter embeddings model that gives similar performance to OpenAI's text-embedding-ada-002, but is much smaller+faster! 🔥 I don't understand why >95% of developers still use closed-source embeddings models... 😅 + it's compatible with 🤗 Transformers.js!

Tweet media one

@andersonbcdefg

Ben (e/sqlite)

@andersonbcdefg

7 months

New embeddings model, gte-tiny, is published! Distilled from gte-small, offering slightly-worse performance with half the layers. (Alternatively, same size but better performance compared to all-MiniLM-L6-v2.) ONNX models also available. Check it out! (link below)

Tweet media one

7

33

231

7

45

309

@xenovacom

Xenova

4 months

Depth Anything is now available in 🤗 Transformers.js! At just 25M parameters, the small version of the model runs great locally. Here's a demo I created which performs monocular depth estimation directly in your browser (no server needed)! 🤯 🔗 Demo:

3

54

302

@xenovacom

Xenova

4 months

TinyLlama is finally here: a 1.1B Llama model trained on 3 trillion tokens! 🤯 It's also compatible with 🤗 Transformers.js (see code below)! 👇 What a way to end the year! 🥳 🔗

Tweet media one

8

46

297

@xenovacom

Xenova

20 days

Using 🤗 Transformers.js, you can now run CLIP directly in your browser at over 20fps w/ WebGPU (@ fp16) for real-time zero-shot image classification! 🤯 As always, everything runs 100% locally, meaning no calls made to an API! 🔥 Try it out! 👇 🔗 Demo:

6

45

282

@xenovacom

Xenova

9 months

Introducing Chat with YouTube, an AI-powered browser extension that lets you chat with YouTube videos! 🚀 This project shows how easy it is to build conversational browser extensions using 🤗 Inference Endpoints and @Vercel 's AI SDK. + it's open source!

13

66

266

@xenovacom

Xenova

2 months

Claude 3 just released and although the weights aren't open, the tokenizer is! 🔥 If you want to calculate how many tokens you're sending to the API, check out The Tokenizer Playground, which we recently updated to include the Claude 3 tokenizer! 👇 🔗

3

54

273

@xenovacom

Xenova

1 month

Introducing MusicGen Web: AI-powered music generation directly in your browser, built with 🤗 Transformers.js! 🎵 Everything runs 100% locally, meaning no calls to an API! 🤯 Served as a static website... this costs $0 to host and run! 🔥 Try it out yourself! 👇

7

58

272

@xenovacom

Xenova

6 months

We just released 🤗 Transformers.js v2.8.0, which adds a ton of new features, including: 🖼️ Super-resolution and image restoration w/ Swin2SR ✍️ Optical character recognition w/ TrOCR 💬 Text-generation w/ Mistral and Falcon (<1B params) More details in 🧵👇

Tweet media one

5

53

269

@xenovacom

Xenova

1 year

Introducing Transformers.js – run @HuggingFace transformers directly in your browser! We currently support BERT, DistilBERT, T5, and GPT2 models, for a variety of tasks such as translation, text generation, and sentiment analysis. … and it's open-source!

10

53

258

@xenovacom

Xenova

10 months

We just released Transformers.js v2.4.0, which adds support for generating word-level timestamps w/ whisper! 🤯 That's right, you can now perfectly caption videos directly in your browser. 🔥 I can't wait to see what you build with this! Get started in just a few lines of code:

Tweet media one

10

53

253

@xenovacom

Xenova

2 months

⚡️ Now with WebGPU support! ⚡️ Run depth estimation w/ Depth Anything in under 200ms, thanks to Transformers.js and WebGPU! Try it out yourself! 👇

@xenovacom

Xenova

4 months

Depth Anything is now available in 🤗 Transformers.js! At just 25M parameters, the small version of the model runs great locally. Here's a demo I created which performs monocular depth estimation directly in your browser (no server needed)! 🤯 🔗 Demo:

3

54

302

7

41

252

@xenovacom

Xenova

9 months

Two annoying things about OpenAI's tokenizer playground: (1) it's capped at 50k characters, and (2) it doesn't support GPT-4 or GPT-3.5... So, I built my own version w/ Transformers.js! It can tokenize the entire "Great Gatsby" (269k chars) in 200ms! 🚀

Tweet media one

Tweet media two

8

35

252

@xenovacom

Xenova

2 months

New features in 🤗 Transformers.js v2.16.1: 🔥 New models: APISR for Anime Super-Resolution and EfficientNet for image classification 🖼️ New pipeline: Image Feature Extraction 💬 Improved chat templating support: C4AI Command-R tool and RAG prompt generation See 🧵 for more info

5

33

211

@xenovacom

Xenova

7 months

A new open-source embeddings model with 8K context length that matches the performance of text-embedding-ada-002! 🤯 This is a game changer! 🔥 And now it's compatible with 🤗 Transformers.js, meaning you can generate embeddings in your browser, Node.js, or even Deno!

Tweet media one

@JinaAI_

Jina AI

7 months

Introducing jina-embeddings-v2, the world's first open-source model boasting an 8K context length. Matching the prowess of OpenAI's proprietary models, now accessible on @huggingface , signaling a significant milestone in the landscape of text embeddings.

15

102

495

2

27

200

@xenovacom

Xenova

27 days

The competition in AI music generation is heating up, with Suno and Udio leading the way. Unfortunately, neither are open source... 😢 Luckily, MusicGen is! 😍 The quality is amazing and you can even run it locally in your browser with Transformers.js! 🎵 For example: prompt:…

6

39

200

@xenovacom

Xenova

6 months

Yes, that's right... the new Distil-Whisper models from @huggingface are fully compatible with Transformers.js! 🤗 This means you can generate high-quality transcripts directly in JavaScript: in-browser, Node, or even Deno! 🤯🚀 Get started in just 3 lines of code:

Tweet media one

@xenovacom

Xenova

6 months

Introducing Distil-Whisper Web: 49% smaller, 4.2x faster Whisper directly in your browser! 🚀 Here is a side-by-side comparison with OpenAI's original version! 🤯

16

88

573

5

25

176

@xenovacom

Xenova

2 months

Introducing the 🤗 Transformers.js WebGPU Embedding Benchmark! ⚡️ How much does WebGPU speed up ML models running locally in your browser? Try it out and share your results! 🚀👇 🔗

7

28

174

@xenovacom

Xenova

10 months

AI code completion running 100% locally inside your browser, thanks to @BigCodeProject 's StarCoder models and 🤗 Transformers.js! We also got their new 1B model running at ~20 tokens per second in Node.js (CPU). 🚀 Check out the demo!

7

35

165

@xenovacom

Xenova

25 days

Meta's Llama 3 is here, with a brand new tokenizer! 🦙 I've added it to the Tokenizer Playground, so you can experiment with it in your browser: For those interested, here are the key differences over Llama 2: 1. 4x larger vocabulary (32K -> 128K). This…

Tweet media one

3

32

169

@xenovacom

Xenova

5 months

🤗 Transformers.js v2.13 - Holiday update! ☃️ In this version, we added: 1. SegFormer for semantic segmentation and image classification. 2. VITS for multilingual text-to-speech (>1000 languages). 3. CLIPSeg for zero-shot image segmentation. 4. Table Transformer for table…

Tweet media one

5

30

160

@xenovacom

Xenova

6 months

This is an absolute game changer! 🤯 @threejs but for Gaussian Splatting! 🔥 I can’t wait to see what the web-dev community builds with this! 🤗 cc @mrdoob

@dylan_ebert_

dylan

6 months

🚀 Introducing gsplat.js - a JavaScript Gaussian Splatting library 🚄 Speed: Optimized WebAssembly ☂️ Versatility: Modular, swap out the renderer 🤗 Open Source: MIT License Code: Repo:

18

273

1K

1

32

156

@xenovacom

Xenova

6 months

Transformers.js just hit 1 million total requests on @jsDelivr , with 52% of them coming in the past 30 days alone! 🤯 We have a ton of exciting updates coming soon, so stay tuned! I'm excited to show you what's next... 🚀🔥

Tweet media one

8

24

150

@xenovacom

Xenova

4 months

Image-to-LaTeX in 3 lines of JavaScript code, with 🤗 Transformers.js! This is made possible thanks to @vikparuchuri 's amazing texify2 model, which we converted to ONNX so it can run in the browser! 🔥 💡 Project idea: browser extension to convert PDFs/screenshots to LaTeX!

Tweet media one

3

26

143

@xenovacom

Xenova

5 months

To showcase the power of in-browser machine learning for real-time data visualization, I built a semantic music search app, powered by 🤗 Transformers.js and Deepscatter. Users can search over 50k songs with natural language, all running client-side (no server)! 🤯

9

19

142

@xenovacom

Xenova

3 months

Nomic Embed v1.5 is out, the first open model with variable-sized Matryoshka embeddings and 8192 context! 🤯 It's also compatible with 🤗 Transformers.js, meaning you can perform adaptive retrieval directly in your browser! Demo showing how dimensionality affects performance 👇

5

26

142

@xenovacom

Xenova

1 year

Yes, you *heard* that right... Transformers.js now supports automatic speech recognition w/ Whisper! Everything runs entirely inside your browser. No need to make API calls to a server! 🤯 #WebML ... and yes, it's open source:

6

35

140

@xenovacom

Xenova

7 months

In-browser TTS in 3 lines of code! 🤯 import { pipeline } from '@xenova/transformers'; let pipe = await pipeline( 'text-to-speech', 'Xenova/speecht5_tts', { quantized: false } ); let out = await pipe('Text goes here', { speaker_embeddings: 'speaker_embeddings.bin' });

@xenovacom

Xenova

7 months

We just released 🤗 Transformers.js v2.7.0, which adds supports for 🗣️ Text to Speech w/ speecht5. This means you can now synthesize human speech directly in your browser... no server required! 🔥 Check out the demo! 👇

15

74

322

7

27

140

@xenovacom

Xenova

10 months

Transformers.js just hit 3000 stars on GitHub! 🤯 The #WebML community is growing so fast, and I'm proud to be a part of it! 🤗 If you ever plan on adding in-browser machine-learning functionality to your website or web-app, check out the project:

Tweet media one

3

30

139

@xenovacom

Xenova

8 months

We just released Transformers.js v2.6.0! New features: - 14 new architectures: BLOOM, MPT, BeiT, CamemBERT, CodeLlama, GPT-J, mBART, ResNet, WavLM, and more! 😍 - Over 150 newly-converted models on the Hub! 🚀 - Huge model size reductions (up to -40%)! 👇

Tweet card media

Release 2.6.0 · xenova/transformers.js

What's new? 🤯 14 new architectures In this release, we've added a ton of new architectures: BLOOM, MPT, BeiT, CamemBERT, CodeLlama, GPT NeoX, GPT-J, HerBERT, mBART, mBART-50, OPT, ResNet, W...

3

25

133

@xenovacom

Xenova

8 months

Transformers.js v2.6.2 now supports Document Question Answering, meaning you can easily extract information from images... directly in your browser (no server needed)! 🤯 We also added new models like Donut, LongT5, and Blenderbot! 🥳 I can't wait to see what you build! 🤗

Tweet media one

3

18

135

@xenovacom

Xenova

2 months

We just added some new features to the 🤗 Transformers.js WebGPU Embedding Benchmark: - fp16 and int8 support - Ability to change models - Lazy model loading - Options to select which tests to run On my device, I got >100x speedup with fp16 on WebGPU! ⚡️

Tweet media one

@xenovacom

Xenova

2 months

Introducing the 🤗 Transformers.js WebGPU Embedding Benchmark! ⚡️ How much does WebGPU speed up ML models running locally in your browser? Try it out and share your results! 🚀👇 🔗

7

28

174

4

18

128

@xenovacom

Xenova

2 months

Transformers.js v2.16 is now out! 🚀 Here are some of the new features: 💬 StableLM text-generation models 🔉 Speaker verification and diarization models 📝 Improved chat templating operation coverage 🎨 New example applications and demos Release notes:

Tweet media one

2

14

129

@xenovacom

Xenova

1 year

Generate embeddings directly in your browser (or Node.js) with the latest version of Transformers.js! 🤯 We can't wait to see what you make with it! The possibilities are endless! Semantic search, sentence similarity, clustering, etc... What else? 👀🔥

Tweet media one

9

23

127

@xenovacom

Xenova

4 months

And now, SigLIP is available in 🤗 Transformers.js! To test how it fares in practice, I adapted my semantic image search demo to use SigLIP instead of CLIP, and it works great! Everything runs 100% locally in your browser (no server needed)! 🔥 🔗

@NielsRogge

Niels Rogge

4 months

SigLIP by @Google is now available in 🤗 Transformers! It improves upon @OpenAI 's CLIP with a simple sigmoid loss. SOTA for linking images with text and vice versa. Demo notebook: Original meme credits: @giffmana

Tweet media one

4

89

656

5

22

129

@xenovacom

Xenova

8 months

Continuing from yesterday: - T5 decoder: 59MB → 42MB (-28%) - distilbert-base: 91MB → 68MB (-25%) - bart-base decoder: 207MB → 155MB (-25%) - roberta-base: 165MB → 126MB (-24%) #WebML for the win! 🚀

@xenovacom

Xenova

8 months

Massive size reductions for Transformers.js models: - Whisper-tiny decoder: 50MB → 30MB (-40%) - NLLB decoder: 732MB → 476MB (-35%) - bloom: 819MB → 562MB (-31%) - gpt2: 167MB → 127MB (-24%) - bert-base: 134MB → 111MB (-17%) ... all this with ZERO drop in performance! 🤯

3

86

467

4

20

123

@xenovacom

Xenova

3 months

Nomic Embed is out: a new 8K text embedding model by @nomic_ai ! 🚀 It's also compatible with 🤗 Transformers.js, meaning you can generate embeddings directly in your browser (no server required)!

Tweet media one

@nomic_ai

Nomic AI

3 months

Introducing Nomic Embed - the first fully open long context text embedder to beat OpenAI - Open source, open weights, open data - Beats OpenAI text-embeding-3-small and Ada on short and long context benchmarks - Day 1 integrations with @langchain , @llama -index, @MongoDB

38

276

2K

2

16

116

@xenovacom

Xenova

9 months

Transformers.js v2.5.2 now supports audio classification w/ MMS and wav2vec2, meaning you can, for example, perform language identification for over 4000 languages! 🤯🔥 Get started in just 3 lines of code! 👇 Full release notes:

Tweet media one

4

18

114

@xenovacom

Xenova

1 month

Transformers.js v2.17 is out! 🔥🚀 New features: 🔢 Binary embeddings: 32x storage savings and significantly faster retrieval (with up to ~95% of the original performance)! 💬 Improved conversational support: Pass chat messages directly to the `text-generation` pipeline.

Tweet media one

2

12

115

@xenovacom

Xenova

3 months

Did you know that HuggingChat uses Transformers.js for RAG/WebSearch? 🤯 It's amazing to see how far the library has come, and I'm so grateful to everyone in the community for helping make it what it is today! 🤗 HuggingChat is also open source! 👇

Tweet card media

GitHub - huggingface/chat-ui: Open source codebase powering the HuggingChat app

Open source codebase powering the HuggingChat app. Contribute to huggingface/chat-ui development by creating an account on GitHub.

2

8

114

@xenovacom

Xenova

10 months

Introducing Doodle Dash, an ML-powered web game that runs completely in your browser, thanks to Transformers.js! 🤯 You have 60 seconds to draw as many words as you can, while a neural network tries to guess what you're drawing in real time! Play here:

5

29

103

@xenovacom

Xenova

5 months

🤗 Transformers.js just hit 5K stars on GitHub! 🌟 Thank you to everyone in the community for your support and contributions... this is why open source is the best! 🔥 PS: Stay tuned for some exciting updates coming soon! 🚀

Tweet media one

3

13

100

@xenovacom

Xenova

25 days

Snowflake just released Arctic Embed, a collection of open-source text embedding models optimized for retrieval accuracy and efficiency! ❄️ 📄 Apache 2.0 license 🌐 Great for in-browser use w/ 🤗 Transformers.js (22 → 335 M params) ⚡️ WebGPU-accelerated (>120x faster than WASM)

Tweet media one

2

14

99

@xenovacom

Xenova

3 months

Did anyone else notice the `<start_of_image>` token in Gemma's vocabulary? 👀 Are we going to see some VLM-variants soon? I also added Gemma to "The Tokenizer Playground", which you can check out if you want to learn more about how the model performs tokenization! 👇

4

10

98

@xenovacom

Xenova

3 months

New blog post: An Introduction to Matryoshka Embedding Models 🪆 Learn how these models are able to produce embeddings of various dimensions, how they can speed up tasks like retrieval, and how you can train your own! 👇 🔗

1

17

95

@xenovacom

Xenova

3 months

🚨 Hugging Chat Assistants are out! 🚨 Just like OpenAI's GPTs, you can now create your own personal assistant in Hugging Chat! 🤯 To test it out, I built a "Prisoner Interrogation" game, where you must try to extract a secret password from a prisoner. Can you do it? 🤔

10

19

91

@xenovacom

Xenova

3 months

Qwen1.5 is out: a collection of powerful LLMs with sizes ranging from 0.5B to 72B parameters. Even at 8-bit quantization, the smallest one (0.5B) is surprisingly good for its size! Here's a demo I made with Transformers.js (v2.15), running 100% locally in the browser w/ WASM! 🤯

@JustinLin610

Junyang Lin

3 months

🎉Happy to announce the release of Qwen1.5! This time, we directly opensource new models of 6 sizes, 0.5B, 1.8B, 4B, 7B, 14B, and 72B (including base, chat, AWQ, GPTQ, GGUF)! From small to huge! Blog: GitHub: HF:…

Tweet media one

34

117

627

4

14

92

@xenovacom

Xenova

9 months

Here's a sneak peek of my "Chat with YouTube" browser extension, made with @Vercel 's AI SDK! 🔥 It uses Llama-v2 (7B) deployed with @HuggingFace inference endpoints. 🚀 Source code and tutorial coming soon! 🤗

4

11

87

@xenovacom

Xenova

4 months

Introducing the 🛝 Jinja Playground: Design LLM chat templates directly in your browser with instant feedback. Built with `@huggingface/jinja`, a minimalistic JavaScript implementation of the Jinja templating engine, specifically designed for parsing + rendering chat templates.

2

19

84

@xenovacom

Xenova

4 months

Yesterday, @MoritzLaurer released some new tiny zero-shot classifiers, so to put them to the test, I built a simple web application that sorts customer product reviews into classes chosen at runtime. Everything runs 100% locally in your browser, thanks to 🤗 Transformers.js!

@MoritzLaurer

Moritz Laurer

4 months

🤏 New 0.02B, 25 MB tiny zeroshot classifiers for edge device use-cases on @huggingface ! The xtremedistil ONNX quantized version is only 13 MB and very fast on CPUs. Without quantization, it has a throughput of ~4000 full sentences (! not just tokens) per second on an A10G with…

1

19

90

1

19

83

@xenovacom

Xenova

5 months

Say goodbye to silent performance issues when prompting LLMs! Today we released 🤗 Transformers.js v2.12, which adds support for chat templating! 💬 This means you can generate LLM inputs for almost any model on the @huggingface Hub, directly in your browser w/ JavaScript! 🤯

Tweet media one

2

22

81

@xenovacom

Xenova

3 months

Local background removal Figma plugin, built with 🤗 Transformers.js and BRIA AI's RMBG-v1.4 model! This shows what an amazing opportunity it is for JavaScript developers to build powerful AI applications, without worrying about API/server costs! Great work @enzostvs ! 🔥

2

15

77

@xenovacom

Xenova

1 year

Today we added code-completion to Transformers.js! 🤯 It's like GitHub copilot, but it runs directly in your browser (i.e., no calls to a server)! #WebML We can't wait to see what people make with it! 😍 ... oh, and did we mention it's open source?

5

18

77

@xenovacom

Xenova

5 months

🤗 Transformers.js v2.11 is one of our biggest releases yet and includes 8 exciting new models! 🧵 1. ViTMatte for image matting: separate images into foreground and background, directly in your browser! This is going to make for some cool image editing web applications! 🔥

Tweet media one

2

17

76

@xenovacom

Xenova

1 year

HUGE seq2seq model size reductions in Transformers.js v1.1.0! whisper: 110 MB -> 61 MB (-45%) t5-small: 151 MB -> 95 MB (-37%) distilbart: 529 MB -> 336 MB (-36%) ... without any reduction in quality! 🤯 This is MASSIVE for #WebML !

Tweet card media

GitHub - xenova/transformers.js: State-of-the-art Machine Learning for the web. Run 🤗 Transformers...

State-of-the-art Machine Learning for the web. Run 🤗 Transformers directly in your browser, with no need for a server! - xenova/transformers.js

4

9

74

@xenovacom

Xenova

5 months

We just released 🤗 Transformers.js v2.10, which adds support for: 🎵 Zero-shot audio classification w/ CLAP 🎙️ Audio classification w/ Audio Spectrogram Transformer 🖼️ Image classification w/ ConvNeXT Get started with just a few lines of code! 🤯🥳

Tweet media one

2

16

75

@xenovacom

Xenova

5 months

How to perform image segmentation in 3 lines of JavaScript code: import { pipeline } from '@xenova/transformers'; const segmenter = await pipeline( 'image-segmentation', 'Xenova/face-parsing', ); const output = await segmenter('image.jpg');

2

16

74

@xenovacom

Xenova

8 months

I just tried installing the latest version of 🤗 Transformers.js with Bun 1.0: npm: 10.7 seconds Bun: 0.404 seconds (26.5x faster) 🤯 @bunjavascript @jarredsumner This is amazing! Great job! 😍

Tweet media one

@xenovacom

Xenova

8 months

We just released Transformers.js v2.6.0! New features: - 14 new architectures: BLOOM, MPT, BeiT, CamemBERT, CodeLlama, GPT-J, mBART, ResNet, WavLM, and more! 😍 - Over 150 newly-converted models on the Hub! 🚀 - Huge model size reductions (up to -40%)! 👇

3

25

133

2

8

68

@xenovacom

Xenova

23 days

🤗 Transformers.js + MusicGen = 🤯 Check it out! 👇

@victormustar

Victor M

23 days

Mind. Blown. Web browsers can now generate music, and it's 🔥 no servers needed, it all happens right in your browser. ⬇️ Check out the link below and get ready to jam!

10

37

217

3

14

68

@xenovacom

Xenova

1 month

@Raynerdwebsdev It sure can!

@xenovacom

Xenova

2 months

We just updated our in-browser Background Removal demo to use WebGPU and it's now ~50x faster! 🤯 ~9 seconds down to 180ms! ⚡️ Powered by @bria_ai_ 's RMBG-v1.4 model and 🤗 Transformers.js! ... and yes, the video is in real time! 🤯

22

113

667

2

5

68

@xenovacom

Xenova

2 months

Amazing work @fleetwood___ ! 🔥 Distil-Whisper large running blazingly fast in your browser, powered by WebGPU! ⚡️

@fleetwood___

Fleetwood

2 months

🚨 Ratchet reaches alpha! 🚨 With todays release of Distil Whisper Large V3 by @sanchitgandhi99 , Ratchet officially enters alpha. Check out this demo running 𝗹𝗮𝗿𝗴𝗲-𝘃𝟯(!!) in the browser

7

13

85

2

3

67

@xenovacom

Xenova

1 year

I can't believe this just happened... @huggingface

Tweet media one

2

5

65

@xenovacom

Xenova

1 year

With all the hype around multi-modal models (like GPT-4), we decided to add support for CLIP to Transformers.js 🔥 CLIP can be used for image-text similarity and zero-shot image classification... and now it can run directly in your browser! Source code:

2

13

65

@xenovacom

Xenova

1 year

You asked for it, so here it is... Transformers.js now supports *timestamped* speech-to-text with whisper! 🤯 We can't wait to see what people will create with it! (see below for example usage and outputs) Source code:

Tweet media one

Tweet media two

1

12

64

@xenovacom

Xenova

1 year

Today we added object detection to Transformers.js! 🤯 This means that you can detect the location and type of objects in an image, directly in your browser (no calls to a server)! ... and did we mention it's open source?

1

15

62

@xenovacom

Xenova

2 months

Bring images to life with 🤗 Transformers.js and Depth Anything! Everything runs 100% locally in your browser (no server required)! 🤯 Great work on the demo @conzept__ ! 🔥

@conzept__

Conzept

2 months

Added "image depth estimation" to the image viewer (using @xenovacom #transformersjs project) This semi 3D perspective can add new insights and flavour to many of the images. Painting: Landscape:

Tweet media one

Tweet media two

Tweet media three

1

1

12

4

11

63

@xenovacom

Xenova

8 months

My first Hugging Face collection! 😍 Check out my favorite #WebML demos built with Transformers.js! 🔗

Tweet media one

Tweet media two

Tweet media three

Tweet media four

1

10

60

@xenovacom

Xenova

10 months

Have you ever wanted to build your own semantic image search application in @nextjs ? Well now you can, thanks to 🤗 Transformers.js and ⚡️ @supabase ! This comes with the release of v2.5.0, which adds support for computing embeddings with CLIP! Demo:

Tweet media one

3

16

61

@xenovacom

Xenova

1 year

With all the hype around SAM (Segment Anything Model) from @MetaAI , we decided to add image-segmentation to Transformers.js! 😍 Unlike their demo, this *actually* runs in your browser, meaning no calls to a server! See how you can use it in your projects today:

Tweet media one

2

10

61

@xenovacom

Xenova

1 year

You can now transcribe audio clips longer than 30 seconds using @OpenAI 's whisper model, directly in your browser (no server)! 😍 Do this by specifying a chunk length and stride length when calling the pipeline function (see example). Check it out!

Tweet media one

4

13

57

@xenovacom

Xenova

6 months

Introducing Static Templates, now live on the 🤗 Hub! Create and deploy your own website in just two clicks! Perfect for showcasing machine learning projects, demos, and more... Get started today! 👇 🔗

2

15

55

@xenovacom

Xenova

1 year

Weights on @huggingface or it didn’t happen… 😉

@AIatMeta

AI at Meta

1 year

Today we're releasing the Segment Anything Model (SAM) — a step toward the first foundation model for image segmentation. SAM is capable of one-click segmentation of any object from any photo or video + zero-shot transfer to other segmentation tasks ➡️

144

2K

7K

0

9

55

@xenovacom

Xenova

1 year

Soon you'll be able to run @huggingface Transformers in browser extensions! 🤯 The models run locally inside your browser - no need for a server! We can't wait to see what people will develop when Transformers.js v2.0.0 releases! 🚀 #WebML

Tweet media one

2

4

53

@xenovacom

Xenova

1 year

This is my petition to change @huggingface 's error page to a sad face, because it better reflects how I feel right now... 😭

Tweet media one

6

3

53

@xenovacom

Xenova

1 year

Transformers.js is number 1 trending on Hacker News! 🤯 and it has just blown up to 400 stars on GitHub?! AHHHHHHHHHHHHHH!

Tweet media one

5

7

48

@xenovacom

Xenova

8 months

Did you know you can build powerful AI applications directly in JavaScript? 👀 Well, to show how easy it is, we published an interactive video tutorial — in partnership with @Scrimba — where we build an Object Detection web app using Transformers.js! 🔗

Tweet media one

4

10

49

@xenovacom

Xenova

1 year

Transformers.js can now do zero-shot classification! 🥳 This means you can classify text according to classes specified at runtime (and without finetuning). What @huggingface tasks should we add next? Leave them in the comments below! 👇

3

12

48

@xenovacom

Xenova

5 months

Build powerful AI chat applications in just a few lines of JavaScript code, with the @huggingface Inference API and Transformers.js! 🤗 Here's a demo showing how to run the latest @MistralAI 7B Instruct model for free in vanilla JS! Try it out yourself!

1

8

47