Видео

I haven't gone back and looked at your implementation of your agent workflow, but when you're talking about restrictions in context windows with your scraping. Are you using RAG with the large documents you're scraping?

Great Video, is there any way to use groq api?

I love this so much!!! Is this a bad answer to the Napoleon “Napoleon occupied the city where the mother of the woman who brought Louis XVI style to the court died in **1804**. Output from operation 0002: Marie Antoinette, the Queen consort of King Louis XVI of France, is known for bringing a more extravagant and luxurious style to the French court during her husband's reign. She was born an Archduchess of Austria and was the youngest daughter of Empress Maria Theresa and Emperor Francis I. Marie Antoinette's mother, Empress Maria Theresa, died on November 29, 1780, at the Hofburg Palace in Vienna, Austria. She died of natural causes at the age of 63, having ruled the Habsburg Empire for 40 years. Her death greatly affected Marie Antoinette, who was very close to her mother despite living in France since her marriage to Louis XVI in 1770. Output from operation 0000: Marie Antoinette, the Queen consort of King Louis XVI of France, is known for bringing a more extravagant and luxurious style to the French court during her husband's reign. Born an Archduchess of Austria, she was the youngest daughter of Empress Maria Theresa and Emperor Francis I. Empress Maria Theresa, Marie Antoinette's mother, died on November 29, 1780, at the Hofburg Palace in Vienna, Austria. She died of natural causes at the age of 63, having ruled the Habsburg Empire for 40 years. Her death greatly affected Marie Antoinette, who was very close to her mother despite living in France since her marriage to Louis XVI in 1770. Napoleon occupied the city where Marie Antoinette's mother, Empress Maria Theresa, died in **1804**.” Off by a year??? But this is exactly why we built our own in house solution and don’t rely on crew ai or any other multi agent framework

The billy Giles answer is interesting there is a billy giles here who died in America New York. He died in New York “He died at Mount Sinai Hospital in New York City on Sept. 25, 2021 after an eight year struggle with progressive anti-MAG peripheral neuropathy” But I’m guessing this is the wrong billy giles

Agency Swarm might be good, but I doubt it and have no intention of checking it out. The founder's attitude, acting like he is the best thing since sliced bread, belittling other solutions instead of pointing out his strengths, just doesn't work for me.

if you base you whole opinion on crewai on the cost of tokens of openai, use a local model or ollama :p

Great video again. Can't wait for you to try and run the coding models.

The use case of multi-hop questions is such a niche use case and you are presenting it as core feature for multiagents. Sure, that use case might be challenging but in real corporate world usual workflows as not of that format. Each step is simple but there are 10,000 steps.

why i cant connect to http 8000 ?

Hey man, Love your videos! Is there anyway you could do one on the RAPTOR method...by using knowledge graphs of the data as grounding please?

I've had the best luck constructing the json and dictionary with classic code based on whether or not the LLM thinks the content of a specific page answers a question.

Apologies if you gave the following information and I missed it. What was the price difference between using 3.5-turbo and 4o?

Yeah, the pro version of perplexity had no issue with the last question.

I love your way of giving tutorials! I'll dive fully into Agency Swarm if/when I can use underlying LLMs other than OpenAI. Philosophical reasons.

Great explanation video, but bad use case for multi agents framework

Wow. Outstanding.

Thank you, my brother, for sharing your knowledge. This video is very helpful for me because I'm creating a new feature and I'll use agents to accomplish my needs. This video provides me with many insights.

forgot to say, love your content and thank you for providing such great insight

I think they are different animals. I was able to get Llama 3 8b model to tell me when it was trained. The dataset dates back to December 2020. Facebook and alikes are providing stuff to us with 3-4 years lag while they stay on latest. Perplexity is even stating on their enterprise website that "Your data stays yours, period. We never train our LLMs on our enterprise customers’ data." so they are using our searches and Google crawling type technology to train their models constantly. They might even scrape this comment and your video transcript next and use to train their data. Latest date in training dataset: December 31, 2020 23:59:59 UTC Training data collection period: January 1, 2016 00:00:00 UTC to December 31, 2020 23:59:59 UTC Training data size: 45,678,901,234 bytes (approximately 45.7 GB) Training data sources: Common Crawl dataset, which includes web pages from the internet Training data format: JSON files containing text data, with a total of 12,345,678 files Training data processing: Preprocessed using NLTK and spaCy libraries for tokenization, stemming, and lemmatization Model architecture: Transformer-based architecture with 12 layers, 16 attention heads, and a hidden size of 128 Model training: Trained using the Adam optimizer with a learning rate of 0.001 and a batch size of 32 Model evaluation: Evaluated using the perplexity metric, with a minimum perplexity of 10 Model updates: Updated regularly to incorporate new data and improve performance

Why using OpenAI key?

Awesome video. Very informative and you sounded very charismatic. Thank you for adding your opinion at the end of the video. I agree, we should wait for the Assitants to be out of Beta and also have a clearer fee structure.

Thanks for the video. I been playing around with similar stuffs primarily with open source models before I came across your videos. Since then I used a few of your prompts:) For open source models, in my case I have gotten llama3 7b and some other 7b model to answer a few questions right(with a lot of prompt tweaking) but struggle a lot of >1 hop questions. Llama3 70b got the 4 hop question right but took like 5+ iterations. I do use RAG with my custom QA agent. I also use playwright+bs as I find that the regular request doesn't work with a lot of website as they track and block your connection plus with playwright, you can filter out of lot of junks on webpages to save token. I used llama3 7b Q6_K locally with a 8gb graphic card and llama70b through Groq free api. Not sure with quantized version you use with Ollama but if you just pull it normally, it probably the Q4 version which I found there a def a noticeable different vs Q6 in term of quality.

This is excellent. And even posting a Github link like a boss.

Watching your videos acts as a reality check for me. I have the same view for most of the issues. Do you have a discord server for your audience?

bro u r the first creator who work with not only GPT in the name of AI

Is this really true for langchain as well? I thought is totally customizable.

Nice reault here with Llama3 70b fp16. The whole time I was thinking "what about groq?" however. Since the inference for the same model appears to be free.

I'm still watching, enjoying your presentation and that you're showing a project simple enough to follow easily. And ouch, that first result is painful hahah, but it's great to show that it's not exactly trivial to get agents to do what we actually want. `ollama list` return the list of installed models by the way

CodeQwen is a 7B model. It might be useful as a coding agent. That’s kind of the point of agents, though - have one really good “manager” model (GPT-4) and a bunch of smaller, highly refined 7B or 8B models that use RAG to prevent garbage and hallucinations. Also, the manager model should be smart enough to detect nonsense from its subordinates.

Chunking and overlap is the main issue. I found that creating structured data from unstructured data using the LLM was the answer. SQL queries through LLM works very well. Thanks for the video and sharing your experiences.

You sir are a pillar of this community, thank you <3

you shold try nous-hermes2-mixtral, or maybe the brand new mistral 7B v0.3

❤❤❤❤❤❤❤

You provided great insights here. I think this is where phidata comes in also. It allows you do define prompts for each agent you create. However custom builds are always the best

Amazing video! I'm just about to do something similar, but at a much smaller scale. I wish Mistral v0.3 was in the mix also because that model is pretty amazing in my experience.

Love your content boss. Keep them coming

There is a problem with your logic in the first question. One does not need to first know who the first president of Namibia was in order to know who succeeded him. One can simply learn who was the second president.

That's dope! Keep up to the good work!

I'm glad RUclips algorithm listed you on my results i i been looking hard wanting info for llms and agents as i am a noob coder learning and llm local learning

Isn't the largest city north of London Manchester?

RAG with a knowledge graph

Great overview! Do you think it could get a more improved results if it could use a web search and a robust RAG solution than web search and agentic one?

I have used AutoGen, CrewAI and other tools. Honestly, they are not production ready. They are over engineered. What I've learned from building my own tools will be very useful for the future. Learn as much as you can building your own tools, its a lot simpler than most people think, just give it a shot.

Just started this video but looking forward to it. Definitely saw limitations due to the abstraction of CrewAI and Autogen. Recently learned about LangGraph but this looks like some nice next level customization.

I'm curious bro why you chose Runpod over lighting ai?

So many on my channel and my live streams have also indicated this frustration with the agent frameworks. The function calling in ollama is pretty rock solid but many frameworks have problems with it so that’s frustrating too. Thanks for putting this video together.

Wow - big difference between the 8b and 70b models. Do you think the 70b models are good enough for agents?

unfortunately it doesnt work now. the web terminal fails to start :(

Good stuff as always, thank you very much.

Great, thanks! Now I'm curious how to connect SillyTavern to it. I've heard ST might be a bit incompatible with the latest vllm.