Share and discover more about AI with social posts from the community.huggingface/OpenAi
NEW math-instruct model + dataset!
ValiantLabs/Llama3.1-8B-Cobalt
is our new math-instruct model.
Trained using a synthetic math-instruct dataset generated with Llama 3.1 405b. Find the dataset here:
sequelbox/Polytope
More to come soon :)
ValiantLabs/Llama3.1-8B-Cobalt
is our new math-instruct model.
Trained using a synthetic math-instruct dataset generated with Llama 3.1 405b. Find the dataset here:
sequelbox/Polytope
More to come soon :)
Supercool Weekend Read🤖
Nvidia researchers achieved SOTA LLM compression metrics using pruning and knowledge distillation techniques.
Details on Techniques (Simplified):
They started off with a large pre-trained language model (15B params), then:
1. Estimated the importance of different parts of the model (neurons, attention heads, layers) using activation-based metrics on a small calibration dataset.
Nvidia researchers achieved SOTA LLM compression metrics using pruning and knowledge distillation techniques.
Details on Techniques (Simplified):
They started off with a large pre-trained language model (15B params), then:
1. Estimated the importance of different parts of the model (neurons, attention heads, layers) using activation-based metrics on a small calibration dataset.
What Happens When RAG System Become Fully Vision-Language Model-Based?
HF Demo:
bokesyo/MiniCPMV-RAG-PDFQA
Multimodal Dense Retriever:
RhapsodyAI/minicpm-visual-embedding-v0
Generation Model:
openbmb/MiniCPM-V-2_6
Github: https://github.com/RhapsodyAILab/MiniCPM-V-Embedding-v0-Train
The Vision-Language Model Dense Retriever MiniCPM-Visual-Embedding-v0 reads PDFs directly -- no OCR required. With strong OCR capability and visual understanding capability, it generates multimodal dense representations, allowing you to build and search through your personal library with ease.
Ask a question, it retrieves the most relevant pages. Then, MiniCPM-V-2.6 provides answers based on the retrieved pages, with strong multi-image understanding capabilities.
Whether you’re working with a visually-intensive or text-oriented PDF, it helps you quickly find the information you need. You can also build a personal library with it.
It operates just like a human: reading, storing, retrieving, and answering with full visual comprehension.
Currently, the online demo supports PDFs with up to 50 pages due to GPU time limits. For longer PDFs or entire books, you can deploy it on your own machine.
https://cdn-uploads.huggingface.co/production/uploads/6415818a986557e8cac252bf/sjtQD7CFgox46h9EVHCG_.png
HF Demo:
bokesyo/MiniCPMV-RAG-PDFQA
Multimodal Dense Retriever:
RhapsodyAI/minicpm-visual-embedding-v0
Generation Model:
openbmb/MiniCPM-V-2_6
Github: https://github.com/RhapsodyAILab/MiniCPM-V-Embedding-v0-Train
The Vision-Language Model Dense Retriever MiniCPM-Visual-Embedding-v0 reads PDFs directly -- no OCR required. With strong OCR capability and visual understanding capability, it generates multimodal dense representations, allowing you to build and search through your personal library with ease.
Ask a question, it retrieves the most relevant pages. Then, MiniCPM-V-2.6 provides answers based on the retrieved pages, with strong multi-image understanding capabilities.
Whether you’re working with a visually-intensive or text-oriented PDF, it helps you quickly find the information you need. You can also build a personal library with it.
It operates just like a human: reading, storing, retrieving, and answering with full visual comprehension.
Currently, the online demo supports PDFs with up to 50 pages due to GPU time limits. For longer PDFs or entire books, you can deploy it on your own machine.
https://cdn-uploads.huggingface.co/production/uploads/6415818a986557e8cac252bf/sjtQD7CFgox46h9EVHCG_.png
GitHub
GitHub - RhapsodyAILab/MiniCPM-V-Embedding
Contribute to RhapsodyAILab/MiniCPM-V-Embedding development by creating an account on GitHub.
So turns out I've been spreading a bit of misinformation when it comes to imatrix in llama.cpp
It starts true; imatrix runs the model against a corpus of text and tracks the activation of weights to determine which are most important
However what the quantization then does with that information is where I was wrong.
I think I made the accidental connection between imatrix and exllamav2's measuring, where ExLlamaV2 decides how many bits to assign to which weight depending on the goal BPW
Instead, what llama.cpp with imatrix does is it attempts to select a scale for a quantization block that most accurately returns the important weights to their original values, ie minimizing the dequantization error based on the importance of activations
The mildly surprising part is that it actually just does a relatively brute force search, it picks a bunch of scales and tries each and sees which one results in the minimum error for weights deemed important in the group
But yeah, turns out, the quantization scheme is always the same, it's just that the scaling has a bit more logic to it when you use imatrix
Huge shoutout to @compilade for helping me wrap my head around it - feel free to add/correct as well if I've messed something up
It starts true; imatrix runs the model against a corpus of text and tracks the activation of weights to determine which are most important
However what the quantization then does with that information is where I was wrong.
I think I made the accidental connection between imatrix and exllamav2's measuring, where ExLlamaV2 decides how many bits to assign to which weight depending on the goal BPW
Instead, what llama.cpp with imatrix does is it attempts to select a scale for a quantization block that most accurately returns the important weights to their original values, ie minimizing the dequantization error based on the importance of activations
The mildly surprising part is that it actually just does a relatively brute force search, it picks a bunch of scales and tries each and sees which one results in the minimum error for weights deemed important in the group
But yeah, turns out, the quantization scheme is always the same, it's just that the scaling has a bit more logic to it when you use imatrix
Huge shoutout to @compilade for helping me wrap my head around it - feel free to add/correct as well if I've messed something up
How good are you at spotting AI-generated images?
Find out by playing Fake Insects 🐞 a Game where you need to identify which insects are fake (AI generated). Good luck & share your best score in the comments!
victor/fake-insects
Find out by playing Fake Insects 🐞 a Game where you need to identify which insects are fake (AI generated). Good luck & share your best score in the comments!
victor/fake-insects
I'm excited to share a really cool milestone in my AI/LLM journey.
Brief backstory: Before diving into AI, I spent over a decade working in ecological fields such as the conservation corps, biodynamic farming, and natural habitat restoration. This background instilled in me a deep concern about the environmental impact of scaling AI without sustainable practices.
Driven by this concern, I've spent months planning and experimenting to make my AI work more eco-friendly. I'm thrilled to announce that I've successfully transitioned my entire operation to run on 100% sustainable solar power!
Brief backstory: Before diving into AI, I spent over a decade working in ecological fields such as the conservation corps, biodynamic farming, and natural habitat restoration. This background instilled in me a deep concern about the environmental impact of scaling AI without sustainable practices.
Driven by this concern, I've spent months planning and experimenting to make my AI work more eco-friendly. I'm thrilled to announce that I've successfully transitioned my entire operation to run on 100% sustainable solar power!
🚀 We’re excited to launch Ghost 8B Beta (1608), a top-performing language model with unmatched multilingual support and cost efficiency.
Key Highlights:
- Superior Performance: Outperforms Llama 3.1 8B Instruct, GPT-3.5 Turbo, Claude 3 Opus, GPT-4, and more in winrate scores.
- Expanded Language Support: Now supports 16 languages, including English, Vietnamese, Spanish, Chinese, and more.
- Enhanced Capabilities: Improved math, reasoning, and instruction-following for better task handling.
Key Highlights:
- Superior Performance: Outperforms Llama 3.1 8B Instruct, GPT-3.5 Turbo, Claude 3 Opus, GPT-4, and more in winrate scores.
- Expanded Language Support: Now supports 16 languages, including English, Vietnamese, Spanish, Chinese, and more.
- Enhanced Capabilities: Improved math, reasoning, and instruction-following for better task handling.
🔔 Release: small-text v1.4.1
The new release contains some smaller bugfixes. Check it out!
Github: https://github.com/webis-de/small-text
Paper:
Small-Text: Active Learning for Text Classification in Python (2107.10314)
The new release contains some smaller bugfixes. Check it out!
Github: https://github.com/webis-de/small-text
Paper:
Small-Text: Active Learning for Text Classification in Python (2107.10314)
GitHub
GitHub - webis-de/small-text: Active Learning for Text Classification in Python
Active Learning for Text Classification in Python. Contribute to webis-de/small-text development by creating an account on GitHub.
Put together a small repo showing how to go from making your own fine-tuning dataset w/ services like Groq & Together to publishing that model on ollama.
In my case I fine-tuned SmolLM-360M to be a better assistant for my Pi-Card (previous post) project.
Check it out!
https://github.com/nkasmanoff/ft-flow
In my case I fine-tuned SmolLM-360M to be a better assistant for my Pi-Card (previous post) project.
Check it out!
https://github.com/nkasmanoff/ft-flow
GitHub
GitHub - nkasmanoff/ft-flow: Synthetic data to inference for LLM finetuning
Synthetic data to inference for LLM finetuning. Contribute to nkasmanoff/ft-flow development by creating an account on GitHub.
ResShift 1-Click Windows, RunPod, Massed Compute, Kaggle Installers with Amazing Gradio APP and Batch Image Processing. ResShift is Efficient Diffusion Model for Image Super-resolution by Residual Shifting (NeurIPS 2023, Spotlight).
Official Repo : https://github.com/zsyOAOA/ResShift
I have developed a very advanced Gradio APP.
Official Repo : https://github.com/zsyOAOA/ResShift
I have developed a very advanced Gradio APP.
GitHub
GitHub - zsyOAOA/ResShift: ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting (NeurIPS@2023 Spotlight…
ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting (NeurIPS@2023 Spotlight, TPAMI@2024) - zsyOAOA/ResShift
🚀 Introducing Hugging Face Similar: a Chrome extension to find relevant datasets!
✨ Adds a "Similar Datasets" section to Hugging Face dataset pages
🔍 Recommendations based on dataset READMEs
🏗 Powered by https://huggingface.co/chromadb and https://huggingface.co/Snowflake embeddings.
You can try it here: https://chromewebstore.google.com/detail/hugging-face-similar/aijelnjllajooinkcpkpbhckbghghpnl?authuser=0&hl=en.
I am very happy to get feedback on whether this could be useful or not 🤗
✨ Adds a "Similar Datasets" section to Hugging Face dataset pages
🔍 Recommendations based on dataset READMEs
🏗 Powered by https://huggingface.co/chromadb and https://huggingface.co/Snowflake embeddings.
You can try it here: https://chromewebstore.google.com/detail/hugging-face-similar/aijelnjllajooinkcpkpbhckbghghpnl?authuser=0&hl=en.
I am very happy to get feedback on whether this could be useful or not 🤗
huggingface.co
chromadb (chroma)
Org profile for chroma on Hugging Face, the AI community building the future.
🤗 Serving Meta Llama 3.1 405B on Google Cloud is now possible via the Hugging Face Deep Learning Containers (DLCs) for Text Generation Inference (TGI)
In this post, we showcase how to deploy
meta-llama/Meta-Llama-3.1-405B-Instruct-FP8
on an A3 instance with 8 x H100 GPUs on Vertex AI
Thanks to the Hugging Face DLCs for TGI and Google Cloud Vertex AI, deploying a high-performance text generation container for serving Large Language Models (LLMs) has never been easier. And we’re not going to stop here – stay tuned as
In this post, we showcase how to deploy
meta-llama/Meta-Llama-3.1-405B-Instruct-FP8
on an A3 instance with 8 x H100 GPUs on Vertex AI
Thanks to the Hugging Face DLCs for TGI and Google Cloud Vertex AI, deploying a high-performance text generation container for serving Large Language Models (LLMs) has never been easier. And we’re not going to stop here – stay tuned as
🚀 How The Washington Post Uses AI to Empower Journalists 🔍📰
An exciting new example in the world of AI-assisted journalism! The Post has developed an internal tool called "Hayatacker" that's enhancing in-depth reporting. Here's why it matters:
🎥 What it does:
• Extracts stills from video files
• Processes on-screen text
An exciting new example in the world of AI-assisted journalism! The Post has developed an internal tool called "Hayatacker" that's enhancing in-depth reporting. Here's why it matters:
🎥 What it does:
• Extracts stills from video files
• Processes on-screen text
🚀 We will be generating a preference dataset for DPO/ORPO and cleaning it with AI feedback during our upcoming meetup!
In this session, we'll walk you through the essentials of building a distilabel pipeline by exploring two key use cases: cleaning an existing dataset and generating a preference dataset for DPO/ORPO. You’ll also learn how to make the most of AI feedback, integrating Argilla to gather human feedback and improve the overall data quality.
In this session, we'll walk you through the essentials of building a distilabel pipeline by exploring two key use cases: cleaning an existing dataset and generating a preference dataset for DPO/ORPO. You’ll also learn how to make the most of AI feedback, integrating Argilla to gather human feedback and improve the overall data quality.
𝗚𝗼𝗼𝗴𝗹𝗲 𝗽𝗮𝗽𝗲𝗿 : 𝘀𝗰𝗮𝗹𝗶𝗻𝗴 𝘂𝗽 𝗶𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝗰𝗼𝗺𝗽𝘂𝘁𝗲 𝗯𝗲𝗮𝘁𝘀 𝟭𝟰𝘅 𝗹𝗮𝗿𝗴𝗲𝗿 𝗺𝗼𝗱𝗲𝗹𝘀 🚀
Remember scaling laws? These are empirical laws that say "the bigger your model, the better it gets". More precisely, "as your compute increases exponentially, loss decreases in a linear fashion". They have wild implications, suggesting that spending 100x more training compute would make you super-LLMs. That's why companies are racing to build the biggest AI superclusters ever, and Meta bought 350k H100 GPUs, which probably cost in the order of $1B.
But think of this : we're building huge reasoning machines, but only ask them to do one pass through the mod
Remember scaling laws? These are empirical laws that say "the bigger your model, the better it gets". More precisely, "as your compute increases exponentially, loss decreases in a linear fashion". They have wild implications, suggesting that spending 100x more training compute would make you super-LLMs. That's why companies are racing to build the biggest AI superclusters ever, and Meta bought 350k H100 GPUs, which probably cost in the order of $1B.
But think of this : we're building huge reasoning machines, but only ask them to do one pass through the mod
🚀 Meet the new GLiNER architecture 🚀
GLiNER revolutionized zero-shot NER by demonstrating that lightweight encoders can achieve excellent results. We're excited to continue R&D with this spirit 🔥. Our new bi-encoder and poly-encoder architectures were developed to address the main limitations of the original GLiNER architecture and bring the following new possibilities:
🔹 An unlimited number of entities can be recognized at once.
🔹Faster inference when entity embeddings are preprocessed.
🔹Better generalization to unseen entities.
GLiNER revolutionized zero-shot NER by demonstrating that lightweight encoders can achieve excellent results. We're excited to continue R&D with this spirit 🔥. Our new bi-encoder and poly-encoder architectures were developed to address the main limitations of the original GLiNER architecture and bring the following new possibilities:
🔹 An unlimited number of entities can be recognized at once.
🔹Faster inference when entity embeddings are preprocessed.
🔹Better generalization to unseen entities.
'Legal Dictionary GPT' is now completely trained and ready for Open Source release to the world! Trained on 10,000 rows of legal definitions, Legal Dictionary GPT is your go-to resource for everything related to the first step in understanding the law, defining it. The model is free and publicly available for anyone to use.
Model Link: https://platform.openai.com/playground/chat?preset=eCrKdaPe9cnMnyTETqWDCQAU
Knowledge Base Bots are internal facing as opposed to external facing LLM models, that are either fine tuned or RAG tuned, generally on systems and processes related data.
Model Link: https://platform.openai.com/playground/chat?preset=eCrKdaPe9cnMnyTETqWDCQAU
Knowledge Base Bots are internal facing as opposed to external facing LLM models, that are either fine tuned or RAG tuned, generally on systems and processes related data.
Openai
OpenAI Platform
Explore developer resources, tutorials, API docs, and dynamic examples to get the most out of OpenAI's platform.
BIG update dropped for
bigdata-pw/Flickr
- now ~515M images! Target for the next update: 1B
In case you missed them; other recent drops include
bigdata-pw/Dinosaurs
- a small set of BIG creatures 🦕🦖 and the first in a series of articles about the art of web scraping! https://huggingface.co/blog/hlky/web-scraping-101 https://huggingface.co/blog/hlky/web-scraping-102
Stay tuned for exciting datasets and models coming soon:
- PC and Console game screenshots
- TV/Film actors biographies and photos (thin
bigdata-pw/Flickr
- now ~515M images! Target for the next update: 1B
In case you missed them; other recent drops include
bigdata-pw/Dinosaurs
- a small set of BIG creatures 🦕🦖 and the first in a series of articles about the art of web scraping! https://huggingface.co/blog/hlky/web-scraping-101 https://huggingface.co/blog/hlky/web-scraping-102
Stay tuned for exciting datasets and models coming soon:
- PC and Console game screenshots
- TV/Film actors biographies and photos (thin
huggingface.co
Web Scraping 101
A Blog post by hlky on Hugging Face
We are proud to release our latest suite of three image(s)-to-3D Gradio demos and two new papers.
SpaRP (Unposed sparse views to 3D):
sudo-ai/SpaRP
SpaRP: Fast 3D Object Reconstruction and Pose Estimation from Sparse Views (2408.10195)
MeshFormer (@minghua @NCJ ):
sudo-ai/MeshFormer
MeshFormer: High-Quality Mesh Generation with 3D-Guided Reconstruction Model (2408.10198)
MeshLRM-reproduced (@sarahwei0210 ):
sudo-ai/MeshLRM
https://huggingface.co/spaces/sudo-ai/MeshLRM
SpaRP (Unposed sparse views to 3D):
sudo-ai/SpaRP
SpaRP: Fast 3D Object Reconstruction and Pose Estimation from Sparse Views (2408.10195)
MeshFormer (@minghua @NCJ ):
sudo-ai/MeshFormer
MeshFormer: High-Quality Mesh Generation with 3D-Guided Reconstruction Model (2408.10198)
MeshLRM-reproduced (@sarahwei0210 ):
sudo-ai/MeshLRM
https://huggingface.co/spaces/sudo-ai/MeshLRM
huggingface.co
MeshLRM (Unofficial) - a Hugging Face Space by sudo-ai
Discover amazing ML apps made by the community
Cooked up a cool & much faster AI voice assistant space that also supports speech translation (with seamless-expressive). Start with the phrase "Please translate" followed by the speech you'd like to translate, to activate speech translation mode. Using opensource LLMs (Llama 3, Mistral etc) with edge tts for voice assistant and seamless-expressive for speech translation.
Give it a try:
Jaward/optimus
https://huggingface.co/spaces/Jaward/optimus
Give it a try:
Jaward/optimus
https://huggingface.co/spaces/Jaward/optimus
huggingface.co
Optimus - a Hugging Face Space by Jaward
Discover amazing ML apps made by the community