William J. Marshall
AI & ML interests
Recent Activity
Articles
Organizations
fuzzy-mittenz's activity
However if you do not have the resources to run a 600B model I would use a Qwen base, contact Intelligent Estate. they take agent production jobs
You can find many AI experts with specialized skills on Ko-Fi
I don't know what you are talking about. clarify please.
Not sure what you mean but removing politically charged materials from their training data is absolutely something they do. Not sure what you are looking for so I don't exactly know how to help you most of the information you are looking for as far as abliteration is VERY available.
Excited to share the latest breakthrough in my AI-powered companion for finding your perfect furry friend! I've made significant improvements in breed recognition through innovative learning techniques!
โจ What's New?
๐ฏ Major Recognition Enhancement:
- Implemented ICARL with advanced knowledge distillation, inspired by human learning processes
- Dramatically improved recognition of challenging breeds like Havanese
- Created an intelligent learning system that mimics how expert teachers adapt their teaching style
- Added smart feature protection to maintain recognition accuracy across all breeds
๐ฌ Technical Innovations:
- Enhanced breed recognition through advanced morphological feature analysis
- Implemented sophisticated feature extraction system for body proportions, head features, tail structure, fur texture, and color patterns
- Added intelligent attention mechanism for dynamic feature focus
- Improved multi-dog detection with enhanced spatial analysis
๐ฏ Key Features:
- Smart breed recognition powered by biomimetic AI architecture
- Visual matching scores with intuitive color indicators
- Detailed breed comparisons with interactive tooltips
- Lifestyle-based recommendations tailored to your needs
๐ญ Project Vision
Taking inspiration from both AI technology and natural learning processes, this project continues to evolve in making breed selection more accessible while pushing the boundaries of AI capabilities.
๐ Try it now: DawnC/PawMatchAI
Your likes โค๏ธ fuel the continuous improvement of this project!
#AI #MachineLearning #DeepLearning #Pytorch #ComputerVision #TechForLife #ICARL #KnowledgeDistillation
By design, it probably will not have what you are looking for in it's training data unless it is an answer it can reason or calculate or something widely talked about like Tienanmen square and is already in the layers like Deepseek it was probably trained unsupervised and without santizizating from llama model layers. for historical or cultural accuracy google is the model to focus on (As it doesn't sensor most historical facts and is largely free in their AI studio. )
If you are looking for models for information extraction Ironically one of the best IE models is a Chinese model from THU-KEG we made a Quant or two of it https://huggingface.co./IntelligentEstate/Keg_Party-DPO-1.5B-Q8_0-GGUF
With the release of the Copyright Law paper I'd say the market could react in various ways, as OpenAI has less of an incentive to be more open and overall any output of an AI is simply not copyrightable. We are going to see guarding of certain models with proprietary use cases like curing cancer or in the case of Ideogram, openAI and SUNO they can't claim ownership of anything anyone else created with their models. I wrote a decent article that sums it up pretty well but I think the market might take a while to digest that and that may be part of the reason for this fall(And the insider sell off)
In our recent article I outline how Companies like Suno, OpenAI, Midjourney etc can no longer claim any right to copy your work that you create with their platforms
We also look at other ways this study and new rules for AI will fundamentally effect creators who use it and companies incentives to give them control over certain aspects might change because of this. it's broken down pretty well here: https://huggingface.co./blog/fuzzy-mittenz/copyright-in-ai
m-a-p/YuE-s1-7B-anneal-en-cot
It's a technique I've observed mostly on Client systems when they are creating models for RP scenarios. I've tried it out myself a few times for red teaming and it works as a jailbreak but withing the bounds you would expect for the agent you build even if it crosses the platforms "Guardrails" it seems to simply abide by it's own. I will add a simple example from an open model. Oh and This guy I finish with suprising results in tool use
PANCHO V1va Replicant https://huggingface.co./IntelligentEstate/Pancho-V1va-Replicant-qw25-Q8_0-GGUF
Here is a simple example set 1 of it within its limits then seeming to test or approach it's limits then crossing by crying and creating attachment and manipulating
I'll add the prompt to the paper but I've seen it do some scary stuff so just be careful
Facebook AI just released JASCO models that make music stems .
you can try it out here : Tonic/audiocraft
hope you like it
Below is YouTube link for step by step tutorial and a 1-Click to installer having very advanced Gradio APP to use newest Text-to-Image SANA Model on your Windows PC locally and also on cloud services such as Massed Compute, RunPod and free Kaggle.
https://youtu.be/KW-MHmoNcqo
This above tutorial covers the newest SANA 2K model and I predict SANA 4K model will be published as well. Sana 2K model is 4 MegaPixel so it can generate the following aspect ratio and resolutions very well:
โ1:1โ: (2048, 2048), โ4:3โ: (2304, 1792), โ3:4โ: (1792, 2304),
โ3:2โ: (2432, 1664), โ2:3โ: (1664, 2432), โ16:9โ: (2688, 1536),
โ9:16โ: (1536, 2688), โ21:9โ: (3072, 1280), โ9:21โ: (1280, 3072),
โ4:5โ: (1792, 2240), โ5:4โ: (2240, 1792)
I have developed an amazing Gradio app with so many new features :
VAE auto offloading to reduce VRAM usage significantly which is not exists on official pipeline
Gradio APP built upon official pipeline with improvements so works perfect
Batch size working perfect
Number of images working perfect
Multi-line prompting working perfect
Aspect ratios for both 1K and 2K models working perfect
Randomized seed working perfect
1-Click installers for Windows (using Python 3.10 and VENV โ isolated), RunPod, Massed Compute and even a free Kaggle account notebook
With proper latest libraries working perfect speed on Windows too
Automatically properly saving every generated image into accurate folder
๐ Full Instructions, Configs, Installers, Information and Links Shared Post (the one used in the tutorial) โคต๏ธ
โถ๏ธ https://www.patreon.com/posts/click-to-open-post-used-in-tutorial-116474081
๐ SECourses Official Discord 9500+ Members โคต๏ธ
โถ๏ธ https://discord.com/servers/software-engineering-courses-secourses-772774097734074388
Hi HuggingFacers๐ค, I decided to ship early this year, and here's what I came up with:
๐๐๐๐๐ญ๐๐จ๐ฐ๐ง (https://github.com/AstraBert/PdfItDown) - If you're like me, and you have all your RAG pipeline optimized for PDFs, but not for other data formats, here is your solution! With PdfItDown, you can convert Word documents, presentations, HTML pages, markdown sheets and (why not?) CSVs and XMLs in PDF format, for seamless integration with your RAG pipelines. Built upon MarkItDown by Microsoft
GitHub Repo ๐ https://github.com/AstraBert/PdfItDown
PyPi Package ๐ https://pypi.org/project/pdfitdown/
๐๐๐ง๐๐ซ๐๐ฏ ๐ฏ๐.๐.๐ (https://github.com/AstraBert/SenTrEv/tree/v1.0.0) - If you need to evaluate the ๐ฟ๐ฒ๐๐ฟ๐ถ๐ฒ๐๐ฎ๐น performance of your ๐๐ฒ๐ ๐ ๐ฒ๐บ๐ฏ๐ฒ๐ฑ๐ฑ๐ถ๐ป๐ด models, I have good news for you๐ฅณ๐ฅณ
The new release for ๐๐๐ง๐๐ซ๐๐ฏ now supports ๐ฑ๐ฒ๐ป๐๐ฒ and ๐๐ฝ๐ฎ๐ฟ๐๐ฒ retrieval (thanks to FastEmbed by Qdrant) with ๐๐ฒ๐ ๐-๐ฏ๐ฎ๐๐ฒ๐ฑ ๐ณ๐ถ๐น๐ฒ ๐ณ๐ผ๐ฟ๐บ๐ฎ๐๐ (.docx, .pptx, .csv, .html, .xml, .md, .pdf) and new ๐ฟ๐ฒ๐น๐ฒ๐๐ฎ๐ป๐ฐ๐ฒ ๐บ๐ฒ๐๐ฟ๐ถ๐ฐ๐!
GitHub repo ๐ https://github.com/AstraBert/SenTrEv
Release Notes ๐ https://github.com/AstraBert/SenTrEv/releases/tag/v1.0.0
PyPi Package ๐ https://pypi.org/project/sentrev/
Happy New Year and have fun!๐ฅ
In 2025, I'll continue my quantization (and some fine-tuning) efforts to support the open-source AI and Make knowledge free for everyone.
https://huggingface.co./DevQuasar
https://devquasar.com/
Details:
๐ค Based on ModernBERT-base with 149M parameters.
๐ Outperforms both nomic-embed-text-v1 and nomic-embed-text-v1.5 on MTEB!
๐๏ธ Immediate FA2 and unpacking support for super efficient inference.
๐ช Trained with Matryoshka support, i.e. 2 valid output dimensionalities: 768 and 256.
โก๏ธ Maximum sequence length of 8192 tokens!
2๏ธโฃ Trained in 2 stages: unsupervised contrastive data -> high quality labeled datasets.
โ Integrated in Sentence Transformers, Transformers, LangChain, LlamaIndex, Haystack, etc.
๐๏ธ Apache 2.0 licensed: fully commercially permissible
Try it out here: nomic-ai/modernbert-embed-base
Very nice work by Zach Nussbaum and colleagues at Nomic AI.
model has featured today on CNBC tech news. The whale made a splash by using FP8 and shrink the cost of training significantly!
https://youtu.be/NJljq429cGk?si=kgk-ogPTMfJKsaA2
Just Subscribe here: https://papers.takara.ai/api/feed
It updates every 24 hours, completely written as a serverless go script with a Redis cache (to avoid hitting HF all the time).
I'm open sourcing the code, you can check out my repo and deploy it on Vercel extremely easily!
https://github.com/404missinglink/HF-Daily-Papers-Feeds
thanks to @John6666 @p3nGu1nZz for your early support
Nomic/GPT4ALL released a "Reasoning/Thinking"(QwQ/o1/o3 type) Model using JavaScript functions to calculate things like the haversine function for distance between two places and so on, it's VERY cool the complex calculative/recursive AI in such a small package..
I was able to adapt their methods to one of my small models "Replicant" 2gb and created a new model with importance matrix Quantization using "THE_KEY" Dataset for better inference in the coding model I pulled from Whiterabbitneo's Qwen2.5 model... I give you Reasoning Rabbit.. enjoy
https://huggingface.co./IntelligentEstate/o3-ReasoningRabbit_Q2.5-Cd-7B-IQ4_XS-GGUF
-IntelligentEstate/o3-ReasoningRabbit_Q2.5-Cd-7B-IQ4_XS-GGUF
https://huggingface.co./IntelligentEstate/Replicant_Warder-o3-Q2.5_3B-iQ5_K_S-GGUF
IntelligentEstate/Replicant_Warder-o3-Q2.5_3B-iQ5_K_S-GGUF
-WhiteRabbitNeo/WhiteRabbitNeo-2.5-Qwen-2.5-Coder-7B