https://github.com/DuskoPre/AutoCoder/wiki i'm not just testing LLMs but also creating a semi-liquid neural network (advanced LLM) with chromadb: https://github.com/DuskoPre/liquid-neural-network-with-chromadb-cache it seems there is a mayor upgrade for LLMs possible trough classifications (managing-system) of inputs to the LLM. the model is still static but trough the implementations of the classifier it can be made semi-liquid.   #dichipcoin

    https://www.tiktok.com/@huggyattack/video/7367979440548187435 jlos shooting star was an offspring of a kpop group. since the script is doing well and the both tesla m40s are heating up too much, i've purchaised two p40s. to get executable scripts isn't that easy and there are no free llms which can make code refraction, see https://aider.chat/docs/leaderboards/refactor.html "The refactoring benchmark requires a large context window to work with large source files. Therefore, results are available for fewer models." there are also smaller coding LLM with large context window size. the DeepSeek Coder V2 0724 (deprecated) q4 version needs 144gb vram.

The idea you're proposing is quite innovative — using a kind of "microscopic grid filter" to detect distant objects like Planet Nine by comparing the sunlight they reflect with that of distant stars. However, there are a few important reasons why astronomers can't use this technique, or similar ones, to detect objects like Planet Nine, especially at such great distances. Let's break it down: 1. Size and Distance of Planet Nine Distance: Planet Nine is hypothesized to be located at a distance of 200 to 1,000 AU (astronomical units) from the Sun. To put this in perspective, that's roughly 100 to 500 times farther than Earth. This means it is incredibly far away, making it extremely faint in v

some content about plans about family...and having a daisy duck... and being confused sitting on too hard balls...getting softer balls is more comfortable

python main.py Welcome to tic-tac-toe!| | | | | | Player 1 turn. Enter a move (a1 to c3):1Invalid move. Try again.| | | | | | Player 2 turn. Enter a move (a1 to c3):a1O| | | | | | Player 1 turn. Enter a move (a1 to c3):a2O|X| | | | | Player 2 turn. Enter a move (a1 to c3):a3O|X|O| | | | Player 1 turn. Enter a move (a1 to c3):c1O|X|O| | X| | Player 2 turn. Enter a move (a1 to c3):c2O|X|O| | X|O| Player 1 turn. Enter a move (a1 to c3):c3O|X|O| | X|O|XPlayer 2 turn. Enter a move (a1 to c3):b1O|X|OO| | X|O|XPlayer 1 turn. Enter a move (a1 to c3):b2O|X|OO|X| X|O|XPlayer 2 turn. Enter a move (a1 to c3):b3O|X|OO|X|OX|O|XIt is a tie! python main.py Welcome to tic-tac-toe!| | | | | | Player 2 turn. E

https://github.com/DuskoPre/AutoCoder/wiki/visualization-of-how-the-data-flows-through-the-%60CodeFormatterML%60-classifier-module Creating a clear, step-by-step visualization of how the data flows through the CodeFormatterML classifier module, especially in an agent-based workflow similar to AutoGPT or BabyCoder, involves breaking down the process into manageable phases. The classifier here integrates multiple components, including data preparation, semi-liquid networks, ChromaDB for storage, and updates based on model performance. Here’s a structured visualization of how data is processed in this workflow: Visual Outline Data Ingestion and Objective Setup Input Sources: The agent retrie

  it is surprising me how fast ai-tech is evolving since the first report about the benefit of an ai (human escape pattern simulation out of buildings) and one of the first ML-tools (tamagotchi) the development is steping forward to smart glasses. under the hood the are unsolved basic questions see https://github.com/DuskoPre/AutoCoder/wiki beyond the massages of eco-, sustainable-, and privacy-movements the tech is still in development and needs the well-matured appearance for the global tech-court. #WeHaveABoogie PS: cheers, my blog got over 20k views!

the measurement avoid to be on a loose the classifier managment of solid liquid neural network is giving a notebook to a llm to make it to a semi-liquid neural network. making notes make sense for sam since "early 2024".

https://github.com/DuskoPre/AutoCoder/wiki i'm not just testing LLMs but also creating a semi-liquid neural network (advanced LLM) with chromadb: https://github.com/DuskoPre/liquid-neural-network-with-chromadb-cache it seems there is a mayor upgrade for LLMs possible trough classifications (managing-system) of inputs to the LLM. the model is still static but trough the implementations of the classifier it can be made semi-liquid.   #dichipcoin

i've achieved the next milestone with the autocoder-script when creating a tic-tac-toe game. this are the LLMs: "OLLAMA_API_MODEL = 'deepseek-coder:33b-instruct-fp16'OLLAMA_HELPER_API_MODEL = 'phi3:14b-medium-128k-instruct-q8_0'OLLAMA_API_URL = 'http://localhost:11434/api'coder_LLM = 'dolphincoder:15b-starcoder2-fp16'coder_LLM_2 = 'yi-coder:9b-chat-fp16'" and this the objective: "Write a Python program that allows two players to play a game of Tic-Tac-Toe on the console. The program should start by printing out an empty 3x3 board. Players should then take turns entering their moves by specifying the row and column of the square they want tomark. The program should validate the moves to ensur

how to save the world from pollution and not affordable energy-consumption. mr. gates gives a bit of a push to the thin bone african soup... and is the gainer of the commercialsation of openai.

#dichipcoin     the swarm-coder is tested and working. i'm working on this for months and have tried a lot of improvements while learning by doing. implementing memories and web-search for a LLM is crucial.  beyond this i'm experimenting with ML BoW models creating a pool of classifiers (vectorizers) and managing these. the magic code results in fantastic and wonder improvements and can make a land to land of milk and honey. I'm using these LLM: "OLLAMA_API_MODEL = 'deepseek-coder:33b-instruct-fp16'OLLAMA_HELPER_API_MODEL = 'phi3:14b-medium-128k-instruct-q8_0'OLLAMA_API_URL = 'http://localhost:11434/api'coder_LLM = 'dolphincoder:15b-starcoder2-fp16'coder_LLM_2 = 'yi-coder:9b-chat-fp16'" a

i've started a journey into the world of AI and coding (with chatGPT), with the goal to make some scirpts on github and to learn about ML and AI. now i have a Swarm-Coder, RAG and a Crawler... who could expect that- i got even a call from an agency which is lunching start-ups into trading markets...a call just for info ;-) the swarm-coder has done a complete simple script- it is a lot of code and very complicated. now it has chroma verctor-db and is getting search-results content-chunks for processing. it got a lot of error handling. RAG was the simplest to create. when i started to do python-coding , i had no idea about it or AI. The Crawler is searching for specific content: crawling-pre-

the swarm coder is doing well- sometimes almost perfect. it seems there is a huge variation in the created outputs depending on LLMs point of view. if the point of view which is enabling the advanced AI can be somehow defined- advanced AI-workloads can be achieved. i've recognized memories (with chromadb) give a huge advantage for the LLM.

my project Auto-coder with chatGPT (i'm new with python-coding and AI, so chatGPT is doing the coding for me) has succeeded. basically all the scripts i was interested in and have discuss with chatGPT (because from the orig. source not working) have been successfully rearranged together with chatGPT. what's next?

still am doing basic testing of a swarm-agent-coder with memory and web-scraping capabilities. the swarm-agent-coder with web-scraping is tested and working. in an other script i have memories tested with chromaDB (a bit to slow) but not implemented in the main script yet. trying to implement solr9.1.1 with its new vectorDB-capabilities. with the new nutch1.20 it could crawl web create vector-embeddings (https://sease.io/2023/01/apache-solr-neural-search-tutorial.html) and also chat with a LLM about nutch-craweled data.  

i've just released a multi-agents AutoCoder app on github the goal is to have multiple AIs working in team together to solve problems (f.e. coding, maintenance,...)

this is gross (not just the AI) the MAGA is striking back? is there are connection between the earthquakes of cratia '20, tyrkey '23, japan '24 and new york '24?

i did a nutch web-crawl and had a chat with a local LLM (AI) about it. here are some used scripts (developed witch chatGPT): https://github.com/DuskoPre Please enter your question (or type 'exit' to end): what is Nutch? Apache Nutch is an open-source web crawling and indexing tool. It is designed to automatically download, parse, and index content from websites, making it possible to search that content using a full-text search engine like Apache Solr. The main goal of Nutch is to provide an efficient, flexible, and highly extensible system for performing large-scale web crawling and indexing. Please enter your question (or type 'exit' to end): which benefits can a local LLM AI have with nut

March 20, 2024 05:50 PMLocA01: To create a new workstation for 70B LLVM inference, we'll need to consider the following components: GPU: Given the size of the model (70B), a high-end NVIDIA Ampere architecture GPU like the NVIDIA A100 or NVIDIA GeForce RTX 3090 is recommended for fast inference performance. You can find these GPUs on websites like Amazon or Newegg. CPU: A high-end Intel Core i9 or AMD Ryzen 9 processor is recommended for handling the rest of the computing tasks and managing GPU resources efficiently. You can find these processors on websites like Amazon or Newegg. RAM: To store the model and handle memory-intensive tasks, at least 64GB of DDR4 RAM is recommended. You c

i'm making progress with the AI-management software AGiXT- it can let a local hosted LLM browse the www and web-scrap links. the build-in solutions for the local-llm was limited in producing bigger response (token limited ~150). i guess a bigger LLM is necessary for understanding the instructions inside the AGiXT app- the methods using in AGiXt are specially made for chatGPT4. the locally hosted Lama-2-70b-orca-200k LLM can do the job too: "     February 18, 2024 11:30 AMLocal Agent 3:To answer the user's request, you can use the information provided in the context to determine that they are looking for weather information in Chicago. You can then provide them with the forecast f

a filter is the worldformula, which connect the micro with macro world.   https://futurezone.at/science/schwarzes-loch-milchstrasse-stern-neutronenstern-weltraum/402746200 how can that big structures develop just in one level insted in all directions- and this level is exactly tuned 90° away so we can see it in a form of a circle? if the 9th planet in solar system is a black hole, i hope the alines can fly around them...

LB is the pretty brena and she deserves her status  

i'm testing the agent-ai managing app AGiXT, which can use multiple AI (also local hosted), do websearching and web-scrapping, can train agent on different data and execute commands in chains. a very promising functionality needed to be testing.

powa is aftermeth of the strike- storytellers are one of these aftermaths

blackpinks obsession for color motion pictures is letting 'em making more breaks? to find somethink is the discipline for golddiggers...