Видео

Yep, but not just for dev projects, but also for research in academia. I can tell from my own experience how long it takes for researchers to label data. Let me know if you try this system :)

Hi, If you mean the code and process, then the process is explained in the video; you can access separate files and functions here: github.com/Maryam-Nasseri/Fine-tuning-LLMs-Locally

Thanks for watching! Should work with a 7b-model too if you have more VRAM, set the cuda to True in the training arguments, and keep the batch size lower. Hope it helps :)

Running LLMs is more about VRAM than RAM; I'd say you need 6+ VRAM (larger the model, larger the training dataset => more processing needed). Maybe start with DeepSeek base which is only 7B (see if any Q4 is available). Running on powerful CPU is possible ; I don't use MAC so I cannot comment on it :)

I agree, but for deployment and wider use there is pricing for n8n, which could be beyond some users' budget (unlike CrewAI which can work with local LLMs free!). But thanks for watching :)

@@analyticsCamp I agree but, you can use n8n to create workflows connected to Pinecone or a vector database so you can make your own agentic custom workflow internally

Thank you for watching :)

Hi, the full references of all the papers are cited in the description box :)

Thanks, will do and more to come!

Hi, unfortunately I did not understand your question/comment :(

Hi, the code and process is on my GitHub (link in the description box) so you can follow at your own pace :)

Thanks for watching :) If this is for a simple QA chatbot, then CrewAI could do, but if you need a more robust system and you're willing to spend on it, then one of the paid frameworks such as Oracle may be better (I haven't used it personally, so do your research), and Good luck with your project :)

Thanks for watching :) I also think this is an innovative approach!

Glad you liked it

I think so too!

Hi, yes you can effectively do all that. If you see the video at 06:05 you see there are four different models for the layers with Qwen2 acting as the aggregator. If you take a look at the MOA diagram, you'll see each agent/LLM is depicted with a different colour (A1, A2, A3) in each layer, so in their current set-up in each layer the number of 'reference models' defined by the user will separately produce the intended result, and they get aggregated at the final output (depicted as A4). Yes, I think you can tweak the code to have a different number of agents per layer too; please check their GitHub repository, the bot.py file ( I haven't done that personally). Thanks for your comment :)

Hi, could you please send me an email with more details? (my email address is in my channel's About section. Thanks :)

Thanks for your comment. I visited your channel and subed! Great videos :)

Glad you think so!

Much appreciated!

My pleasure, thanks for watching :)

Thanks for your supportive words. Stay tuned, I have more of this coming :)

Now imagine if we combine this with the agentic power! But I still think it's too early to make a definitive judgement, as many of these papers report on their best results/round! Thanks for watching though :)

Glad you liked it

Glad you liked it!

Thanks for watching :)

Hi, I think you have not set your environment variables correctly (Please follow the video steps). You should either have a valid OpenAI API key (can get it from their website) or just get a free key from the Together AI website for this project. Then, from your code editor terminal, export your environment API key by typing this exactly: echo "export OPENAI_API_KEY='yourkey'" >> ~/.bash_profile -but replace 'yourkey' with the key ID you got (it doesn't have to be only from OpenAI, but any partner of them like Together AI, etc). Then update the shell with the new variable by typing: source ~/.bash_profile -to confirm if it is set correctly, type: echo $TOGETHER_API_KEY -This should show your key, if it is correct you are set. I hope this helps :) PS: if you work on Windows, I think you should use \ instead of / in my code, and instead of bash_profile in the first two lines, use autoexec.bat. I don't work with Windows but this information that I found online may help: To set environment variables in Windows, you can follow these steps: Press Win + R to open the Run dialog. Type sysdm.cpl and press Enter to open the System Properties window. Go to the "Advanced" tab and click on the "Environment Variables" button. In the Environment Variables window, you can set system variables (for all users) or user variables (specific to the current user). Click on "New" to add a new variable, or select an existing variable to edit or delete it.

Thanks for your comment. You are right! I checked the top 3 manually in the CSV file and it looks different. With Mistral I get more accurate results. However, LLAMA3 produced a good synthesis of the career path. The video is basically meant as a tutorial (how to do), but the choice of LLM makes a difference. Thanks for watching :)

@@analyticsCamp Thanks for the reply. Actually I am trying to get insights from the dataset using crew ai. Can you suggest some ways to do it?

Is your dataset a CSV file? This video tutorial is a standard way of calling a CSV file within the CrewAI framework, if you don't get accurate results, change the model, e.g., to Mistral or Qwen2, or dbrx (from Databricks) on a sample dataset where you already know the results; any of the model's which produce accurate results, use that one on your target dataset. If you are doing a more serious data anlytics work, keep in mind that most of these LLMs are primarily language models (designed to predict the next word, not necessarily the 'correct' data), so in this case, using the traditional methods in Pandas, for example for data wrangling, or machine learning models from Scikit-learn will give you the most accurate results. If you insist on agentic method, then try asking one of those LLM agents to access Pandas or Scikit-learn and do the work for you. I haven't tried this honestly, so I don't know how it would turn out. But please keep me updated if this works for you. Hope this information helps :0

Glad you liked it!

LOL :) Who knows? Maybe you're right!

​@analyticsCamp seriously though, thanks for the excellent video.

Glad you liked it and thanks for watching :)

I'm not sure if I understand your question :( Apologies, but it'll be good if you give more context.

Glad you think so :)

As far as I know, the only sql parser in crewai tools is PGSearchTool, which is specifically made for PostgreSQL database tables; yep, it can search and generate sql queries, I think they call it Retrieve and Generate RAG. I haven't tested it yet, but if enough viewers ask for it, I may make sth out of it :)

@@analyticsCamp thanks for the tool name hope enough people here ask for #SQLagent tutorial!

Thanks for watching :) You can use ICL with any LLM, especially the ones you can download directly from Hugging Face or via Ollama. Some other interfaces allow users to attach files to process, so you can write your prompts and instructions in those files plus any images you need to attach. I'm not sure about audio and video ICL at this moment, though.

Thanks for the comment :) Yes, the paper comes with all those explanation. Yep, I also believe this can open a way for more ordinary AI users AND many researchers in other fields.

I do not use LM Studio so I cannot comment on that. But Ollama via terminal is pretty sturdy, CrewAI it should work with all Ollama models, but I have not tested all. If you run into issues you can report it here so others can know and/or help you :)

In Mamba sequence length can be scaled up to a million (e.g., a million-length sequences). It also computes the gradient (did not find any info on second-order opt in their method): they train for 10k to 20k gradient steps.

Thanks for watching :) I have tested multiple models from Ollama and mistral seems to have better performance overall, across most tasks. Agent Swarm can be useful for VERY specialised tasks in which general LLMs get it totally wrong. Other than that, it will add to the time/cost of build.But I'm not sure if I understood your question right!

Sure, you can essentially use any models listed in Ollama as long as you make a model file, you can manage the temperature etc. I have used LLAMA 70b before but surprisingly, it did not show better response than its 7b and 13b on most tasks! I recommend LLAMA3 (I may be able to make a video on it if I get some free time, LOL ).

@@analyticsCamp Awesome thanks ill test the smaller ones first then

Thanks for watching :) If you have a specific LLM-based development/project in mind please let me know. With such easy agentic access, I am also surprised how many AI users are still hooked on zero-shot with paid interfaces!

@@analyticsCamp ha ha it also never made sense to me why ppl don't look into open source LLM 🤔 its free, not limiting your token size, free to experiment with different models and most importantly your data (prompt) is yours and don't become automatically OpenAi's property. Keep up the good work, looking forward to your next video.

Thanks for watching :)