vb @reach-vb.hf.co · 24/12/2024You can play directly with the model via this HF space: huggingface.co/spaces/Qwen/...huggingface.coQVQ 72B Preview - a Hugging Face Space by QwenDiscover amazing ML apps made by the community 030
vb @reach-vb.hf.co · 24/12/2024Model weights here: huggingface.co/Qwen/QVQ-72B...huggingface.coQwen/QVQ-72B-Preview · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science. 120
vb @reach-vb.hf.co · 24/12/2024Qwen released QvQ 72B OpenAI o1 like reasoning model on Hugging Face with Vision capabilities - beating GPT4o, Claude Sonnet 3.5 🔥 4183
vb @reach-vb.hf.co · 06/12/2024Chat with it live for free here: huggingface.co/chat/models/...huggingface.cometa-llama/Llama-3.3-70B-Instruct - HuggingChatUse meta-llama/Llama-3.3-70B-Instruct with HuggingChat 030
vb @reach-vb.hf.co · 06/12/20243.1 70B vs 3.3 70B: Code Generation > HumanEval: 80.5% → 88.4% (+7.9%) > MBPP EvalPlus: 86.0% → 87.6% (+1.6%) Steerability > IFEval: 87.5% → 92.1% (+4.6%) Reasoning & Math > GPQA Diamond (CoT): 48.0% → 50.5% (+2.5%) > MATH (CoT): 68.0% → 77.0% (+9%) 130
vb @reach-vb.hf.co · 06/12/2024Llama 3.3 70B vs 405B: > GPQA Diamond (CoT): 50.5% vs 49.0% > Math (CoT): 77.0% vs 73.8% > Steerability (IFEval): 92.1% vs 88.6% huggingface.co/meta-llama/L...huggingface.cometa-llama/Llama-3.3-70B-Instruct · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science. 110
vb @reach-vb.hf.co · 06/12/2024BOOOOM! Meta released Llama 3.3 70B - 128K context, multilingual, enhanced tool calling, outperforms Llama 3.1 70B and comparable to Llama 405B 🔥 Comparable performance to 405B with 6x LESSER parameters ⚡ 1291
vb @reach-vb.hf.co · 03/12/2024Ofc.. here's the codebase: github.com/huggingface/...github.comGitHub - huggingface/parler-tts: Inference and training library for high-quality TTS models.Inference and training library for high-quality TTS models. - huggingface/parler-tts 020
vb @reach-vb.hf.co · 03/12/2024And.. here's a space to try out the model too: huggingface.co/spaces/ai4bh...huggingface.coIndic Parler-TTS - a Hugging Face Space by ai4bharatA demo of Indic Parler-TTS 110
vb @reach-vb.hf.co · 03/12/2024Check out the model checkpoints here: huggingface.co/ai4bharat/in...huggingface.coai4bharat/indic-parler-tts · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science. 200
vb @reach-vb.hf.co · 03/12/2024Introducing Indic-Parler TTS - Trained on 10K hours of data, 938M params, supports 20 Indic languages, emotional synthesis, apache 2.0 licensed! 🔥 w/ fully customisable speech and voice personas! Try it out directly below or use the model weights as you want! 🇮🇳/acc 4353
vb @reach-vb.hf.co · 02/12/2024try it out today on hf.co/datasets - just click on `SQL Console` followed by `AI Query` 💯hf.coHugging Face – The AI community building the future.We’re on a journey to advance and democratize artificial intelligence through open source and open science. 260
vb @reach-vb.hf.co · 02/12/2024you can just do things - ask AI to create your SQL queries and execute them right in your browser! 🔥 let your creativity guide you - powered by qwen 2.5 coder 32b ⚡ available on all 254,746 public datasets on the hub! go check it out today! 🤗 1282
Reposted by vbSimon Willison @simonwillison.net · 29/11/2024This demo of structured data extraction running on an LLM that executes entirely in the browser (Chrome only for the moment since it uses WebGPU) is amazing My notes here: simonwillison.net/2024/Nov/29/...simonwillison.netStructured Generation w/ SmolLM2 running in browser & WebGPUExtraordinary demo by Vaibhav Srivastav. Here's Hugging Face's [SmolLM2-1.7B-Instruct](https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B-Instruct) running directly in a web browser (using WebGPU, so r... 518523
vb @reach-vb.hf.co · 28/11/2024Here's the GitHub repo in case you fancy it: github.com/Vaibhavs10/g...github.comGitHub - Vaibhavs10/github-issue-generator-webgpuContribute to Vaibhavs10/github-issue-generator-webgpu development by creating an account on GitHub. 080
vb @reach-vb.hf.co · 28/11/2024To showcase how much you can do with just a 1.7B LLM, you pass free text, define a schema of parsing the text into a GitHub issue (title, description, categories, tags, etc) - Let MLC & XGrammar do the rest! That's it, the code is super readable, try it out today! 🤗 huggingface.co/spaces/reach...huggingface.coGithub Issue Generator - a Hugging Face Space by reach-vbDiscover amazing ML apps made by the community 1172
vb @reach-vb.hf.co · 28/11/2024Fuck it! Structured Generation w/ SmolLM2 running in browser & WebGPU 🔥 Powered by MLC Web-LLM & XGrammar ⚡ Define a JSON schema, Input free text, get structured data right in your browser - profit!! 410613
Reposted by vbJeremy Howard @howard.fm · 28/11/2024FYI, here's the entire code to create a dataset of every single bsky message in real time: ``` from atproto import * def f(m): print(m.header, parse_subscribe_repos_message()) FirehoseSubscribeReposClient().start(f) ``` 1944262
Reposted by vbMark Riedl @markriedl.bsky.social · 26/11/2024I have converted a portion of my NLP Online Masters course to blog form. This is the progression I present that takes one from recurrent neural network to seq2seq with attention to transformer. mark-riedl.medium.com/transformers...mark-riedl.medium.comTransformers: OriginsAn unofficial origin story of the transformer neural network architecture. 611615
Reposted by vbOmar Sanseviero @osanseviero.bsky.social · 27/11/2024I'm disheartened by how toxic and violent some responses were here. There was a mistake, a quick follow up to mitigate and an apology. I worked with Daniel for years and is one of the persons most preoccupied with ethical implications of AI. Some replies are Reddit-toxic level. We need empathy. 2933237
vb @reach-vb.hf.co · 26/11/2024> uses 90% sliding window and 10% global attention for efficiency > 2-stage pre-training and 3-phase post-training, including a trapezoid learning rate schedule try it out on hugging face today! 🤗 huggingface.co/collections/...huggingface.coHymba - a nvidia CollectionA series of Hybrid Small Language Models. 040
vb @reach-vb.hf.co · 26/11/2024yo! nvidia finally released the weights for Hymba-1.5B - outperforms Qwen, and SmolLM2 w/ 6-12x less training trained ONLY on 1.5T tokens > massive reductions in KV cache size and improved throughput > combines Mamba and Attention in a hybrid parallel architecture with a 5:1 ratio and meta-tokens 1292
Reposted by vbAndi @andimara.bsky.social · 26/11/2024Let's go! We are releasing SmolVLM, a smol 2B VLM built for on-device inference that outperforms all models at similar GPU RAM usage and tokens throughputs. SmolVLM can be fine-tuned on a Google collab and be run on a laptop! Or process millions of documents with a consumer GPU! 410422
vb @reach-vb.hf.co · 25/11/2024You can run inference via llama.cpp too: huggingface.co/OuteAI/OuteT...huggingface.coOuteAI/OuteTTS-0.2-500M-GGUF · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science. 030
vb @reach-vb.hf.co · 25/11/2024Model weights on the hub, you can even run this on a Raspberry Pi! Go run, inference now! 🐐 huggingface.co/OuteAI/OuteT...huggingface.coOuteAI/OuteTTS-0.2-500M · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science. 130
vb @reach-vb.hf.co · 25/11/2024Smol TTS keeps getting better! Introducing OuteTTS v0.2 - 500M parameters, multilingual with voice cloning! 🔥 > Multilingual - English, Chinese, Korean & Japanese > Cross platform inference w/ llama.cpp > Trained on 5 Billion audio tokens > Qwen 2.5 0.5B LLM backbone > Trained via HF GPU grants 5548
vb @reach-vb.hf.co · 25/11/2024It depends on what you define long context; I'm fairly confident up until 64K and moderately till 128K, beyond that - I've personally never tested. Most of my observations are based on chat use-cases. 020
vb @reach-vb.hf.co · 25/11/2024If depends on quant type, in most cases as long as you keep all the vocabs + softmaxes + RoPE in full precision it doesn't lobotomise the model as much Using AWQ/ GPTQ - are also a pretty good substitute for CUDA runners, at 4.5 bpw Source: Hugging Chat + TGI team logs assessments 110
vb @reach-vb.hf.co · 25/11/2024SmolLM - run, pre-train, fine-tune, evaluate SoTA fully open source LM 🔥 Run with Transformers, MLX, Transformers.js, MLC Web-LLM, Ollama, Candle and more! Apache 2.0 licensed codebase - go explore now! 1362
vb @reach-vb.hf.co · 24/11/2024FWIW - I love reading your takes and have personally found them insightful on numerous occasions! 040
vb @reach-vb.hf.co · 24/11/2024A lot more got released like, OpenScholar, smoltalk, Hymba, Open ASR Leaderboard and much more.. Can't wait for the next week! 🤗 030
vb @reach-vb.hf.co · 24/11/2024Jina AI Jina CLIP v2 - general purpose multilingual and multimodal (text & image) embedding model, 900M params, 512 x 512 resolution, matroyoshka representations (1024 to 64) Apple AIM v2 & CoreML MobileCLIP - large scale vision encoders outperform CLIP and SigLIP CoreML optimised MobileCLIP models 130
vb @reach-vb.hf.co · 24/11/2024Llava o1 - vlm capable of spontaneous, systematic reasoning, similar to GPT-o1, 11B model outperforms gemini-1.5-pro, gpt-4o-mini, and llama-3.2-90B-vision Black Forest Labs Flux.1 tools - four new state of the art model checkpoints & 2 adapters for fill, depth, canny & redux, open weights 120
vb @reach-vb.hf.co · 24/11/2024All the model checkpoints with URLs here: huggingface.co/posts/reach-...huggingface.co@reach-vb on Hugging Face: "Massive week for Open AI/ ML: Mistral Pixtral & Instruct Large - ~123B, 128K…"We’re on a journey to advance and democratize artificial intelligence through open source and open science. 120
vb @reach-vb.hf.co · 24/11/2024Massive week for Open Source AI/ ML Mistral Pixtral & Instruct Large - ~123B, 128K context, multilingual, json + function calling & open weights Allen AI Tülu 70B & 8B - competive with claude 3.5 haiku, beats all major open models like llama 3.1 70B, qwen 2.5 and nemotron 3559
vb @reach-vb.hf.co · 23/11/2024I mostly thought that one would upsample the prompt w/ a LLM before passing too LTV - hence the level of detail. Probably also the curse of going from a low dimensional (text) representation to high dimensional too. 050
vb @reach-vb.hf.co · 23/11/2024Hugging Face model repo: huggingface.co/apple/coreml...huggingface.coapple/coreml-mobileclip · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science. 030
vb @reach-vb.hf.co · 23/11/2024Apple released blazingly fast CoreML models AND an iOS app to run them on iPhone! ⚡ > S0 matches OpenAI's ViT-B/16 in zero-shot performance but is 4.8x faster and 2.8x smaller > S2 outperforms SigLIP's ViT-B/16 in zero-shot accuracy, being 2.3x faster, 2.1x smaller, and trained with 3x fewer data 2434
vb @reach-vb.hf.co · 23/11/2024Congratulations on the brilliant release! ❤️ Love that all the models are on Hugging Face and ready to run inference from: huggingface.co/collections/...huggingface.coAIMv2 - a apple CollectionA collection of AIMv2 vision encoders that supports a number of resolutions, native resolution, and a distilled checkpoint. 040
vb @reach-vb.hf.co · 23/11/2024Check out my new swanky handle! 🦋 - Drop your Hugging Face ID in the comments if you want the same! 4190
vb @reach-vb.hf.co · 22/11/2024LFG!! XGrammar: a lightning fast, flexible, and portable engine for structured generation! 🔥 > Accurate JSON/grammar generation > 3-10x speedup in latency > 14x faster JSON-schema generation and up to 80x CFG-guided generation GG MLC team is literally the best in the game and slept on! ⚡ 2524
vb @reach-vb.hf.co · 20/11/2024🚨 UPDATE: New Whisper based model competing with Nvidia on Open ASR Leaderboard! 🔥 CrisperWhisper aims to transcribe every spoken word exactly as it is, including fillers, pauses, stutters and false starts Whisper Large V3 fine-tune - beats it by roughly ~1 WER margin ⚡ hf.co/spaces/hf-au... 1183
vb @reach-vb.hf.co · 20/11/2024I really hope they release a technical report or the model weights soon! 🤞 110
