How to Use Llama 3 with PandasAI and Ollama Locally

Today, we'll cover how to perform data analysis and visualization with local Meta Llama 3 using Pandas AI and Ollama for free. Happy learning.
▶ Subscribe: bit.ly/subscribe-tirendazai
▶ Join my channel: bit.ly/join-tirendazai
00:01 Introduction
01:32 Setup
03:02 Initialize the model
05:15 Initialize the app
08:10 Build the app
09:18 Inference
11:16 Data visualization
RELATED VIDEOS:
▶ PandasAI Tutorials: bit.ly/pandasai
▶ Ollama Tutorials: bit.ly/ollama-tutorials
▶ LangChain Tutorials: bit.ly/langchain-tutorials
▶ Generative AI for DS: bit.ly/genai-for-data-science
▶ HuggingFace Tutorials: bit.ly/hugging-face-tutorials
▶ LLMs Tutorials: bit.ly/llm-tutorials
FOLLOW ME:
▶ Medium: / tirendazacademy
▶ X: / tirendazacademy
▶ LinkedIn: / tirendaz-academy
Don't forget to subscribe and turn on notifications so you don't miss the latest videos.
▶ Project files: github.com/TirendazAcademy/Pa...
Hi, I am Tirendaz, PhD. I create content on generative AI & data science. My goal is to make the latest technologies understandable for everyone.
#ai #generativeai #datascience

[@DallasGraves]

Holy crap! My mind is absolutely RACING with proof of concept projects I can deliver to our Finance team with this. Thank you so much for making this!

[@TirendazAI]

My pleasure 😊

[@user-el8jv8hx2g]

What are you thinking about? I'm a college student and I want to understand more use cases.

[@MANONTHEMOON419]

dude, please do another video on this with more things you can do, or maybe explaining further, this is amazing.

[@user-flashaction]

This channel has not been discovered yet, thank you for the up-to-date and practical videos!

[@TirendazAI]

Thanks 🙏

[@wvagner284]

Great, great video! I just got a new subscriber! Congrats and regards from Brazil!

[@TirendazAI]

Thanks 🙏

[@ladonteprince]

Undiscovered gem - the voice the instruction pure gold

[@TirendazAI]

Thanks 🙏

[@60pluscrazy]

Unbelievable, never knew about Pandas AI. THANKS VERY MUCH 🎉🎉🎉

[@TirendazAI]

Thanks 🙏

[@WhySoBroke]

Excellent video and great methods!!

[@TirendazAI]

Glad you liked it!

[@GhostCoder83]

Very precious. Keep it up.

[@TirendazAI]

Thanks 🙏

[@mubasharsaeed6044]

Please make a video for agent which perform action using lama or langchain for example if we give an prompt bring red box the lama 3 generate the plan and agent do action. Thanks

[@felipemorelli4059]

Thanks!

[@TirendazAI]

You're welcome 🙏

[@felipemorelli4059]

Excellent video. which cpu are you using?

[@TirendazAI]

Thanks! My system is AMD Ryzen 5 7500F, 64GB RAM and 4070 TI Super graphics card with 16GB VRAM

[@Moustafa_ayad]

Good work

[@TirendazAI]

Thanks 🙏

[@teachitkh]

so great video

[@TirendazAI]

Thank you 🤗

[@teachitkh]

@@TirendazAI how about pdf .?can you help this?

[@user-mv9ul9tz1c]

Thank you for the tutorial. I have a question: do I need to apply for an API key before using PandasAI?

[@TirendazAI]

PandasAI is open source and free. If you are using any open source model, you do not need an API key.

[@JohnBvoN]

Wanted to know, can I load the model directly from huggingface? Also I have stored the model and tokenizer using save_pretrained, How can I use these?

[@TirendazAI]

To load the model from HuggingFace, you can use langchain. Check this link: python.langchain.com/v0.1/docs/integrations/platforms/huggingface/

[@varganbas427]

Grear! Thanks! it works 😁there was an error in python !!!

[@TirendazAI]

You're welcome!

[@user-pn6ey5dn4y]

Thanks for your video. I've replicated your entire solution as a starting point. I keep getting errors from the LLM when trying the exact same queries, same dataset, same everything as what you did in your video. I have a RTX 4070 12GB and tried multiple llama and dolphin llama models up. It seems that every time we ask the LLM to write code to create a histogram or pie chart it creates an error - can you help? Here is an example: Query : create a heatmap of the numerical values Result: Unfortunately, I was not able to answer your question, because of the following error: 'list' object has no attribute 'select_dtypes'

[@TirendazAI]

When a prompt does not work, try again by changing this prompt. For example, "Plot a heatmap of the numerical values".

[@stanTrX]

Tebrikler kardeşim. Aksanından hemen anladım. Selamlar. Abone oldum 🎉

[@TirendazAI]

Teşekkürler 🙏

[@user-mm1tt6oy7v]

Thanks for this video. How to integrate groq api to go faster?

[@TirendazAI]

You can leverage the langchain_groq library or utilize the OpenAI compatibility. I showed how to use pandasai with groq api in this video: kzbin.info/www/bejne/eWe1an2Cfb93fpI

[@HmzaY]

I have 64gb ram and 8gb vram, i want to run llama 70B but it doesn't fit. how can i run it on system ram (64gb one) on python . can you make a video for that?

[@TirendazAI]

I also have 64gb RAM, it worked for me. My system used about 58GB RAM for llama-3:70B. I show RAM I use if I make a video with llama 3:70B.

[@lowkeylyesmith]

Is it possible to expand the limit per file? my csv files are larger then 1GB.

[@TirendazAI]

This is possible, but you need to use a larger model, such as llama-3:70b instead of llama-3b:8b.

[@greg-guy]

I don’t know how Panda works so sorry for the dumb question. Is the entire CSV data processed by LLM, meaning that large set will be slow or even too big , or is the calculation/processing of data is all done in Panda, meaning LLM only creates formulas for Panda ?

[@TirendazAI]

Yes, LLM processes your data and generates an answer based on your question. You can think of this process as summarizing a text. If you have a small data set, you may get a faster response.

[@urajcic1]

great tutorial! I was playing around with this dataset and I get strange errors for questions like 'how many First class female survived?' - Unfortunately, I was not able to answer your question, because of the following error: 'list' object has no attribute 'loc' or list indices must be integers or slices, not str or invalid syntax (, line 3) can anyone reproduce and explain why is this happening?

[@TirendazAI]

Sometimes, when LLM does not understand the prompt, it may not return the output you want. You can try changing the prompt or using a different prompt.

[@RICHARDSON143]

Hey buddy I also want to integrate additional feature so that model can also generate sql query for the result as well trying it using pandasql but no success, can you help me

[@TirendazAI]

I made a video about MySQL database, take a look.

[@scottmiller2591]

Is it possible to get PandasAi to show the code it used to generate the plots, etc.?

[@TirendazAI]

Yes, you can get code if you specify in your prompt.

[@scottmiller2591]

@@TirendazAI Thanks!

[@TirendazAI]

@@scottmiller2591 My pleasure!

[@user-kk1li5mk7q]

How are you able to get the response so fast. It is taking me few minutes to get a response. My csv file has 7K records. running on ubuntu 22 with i7 with 32 GB RAM.

[@TirendazAI]

The most important component for a fast response is the graphics card.

[@varshakrishnan3686]

I'm getting the error no module named pandasai.llm.local_llm. Is there any way to solve it?

[@TirendazAI]

llm is a module in pandasai. Make sure pandasai is installed and virtual environment is activated.

[@varganbas427]

Sorry, but my sample rewrite as your code return me a msg "Unfortunately, I was not able to answer your question, because of the following error: Connection error" and after in code py there is a error message : raise APIConnectionError(request=request) from err openai.APIConnectionError: Connection error.

[@TirendazAI]

Did you start Ollama using the "ollama serve" command?

[@sahilchipkar9761]

Can it handles Large Data? Connecting through sql

[@TirendazAI]

Yes it can. You can work with data such as CSV, XLSX, PostgreSQL, MySQL, BigQuery, Databrick, Snowflake.

[@spkgyk]

Why use the default llama rather than the llama instruct?

[@TirendazAI]

The instruct models are fine-tuned to be able to follow prompted instructions. This version usually is used to make a chatbot, implementing RAG or using agents.

[@sebastianarias9790]

i'm not getting a response from the chat. it keeps "Generating the prompt". what could be a reason for that? Thanks!

[@TirendazAI]

Did you get any error? If yes, can you share this error? I can say something if I see the error.

[@sebastianarias9790]

@@TirendazAI there’s no error my friend. It only takes a very long time to get the output. Any ideas?

[@TirendazAI]

Which large model do you use?

[@sebastianarias9790]

@@TirendazAI llama3 !

[@sebastianarias9790]

@@TirendazAI llama3 !

[@vinaya68vinno1]

Can I do this using MySQL in streamlit for visualization can you send code

[@TirendazAI]

I am planning to implement a project using MySQL.

[@vinaya68vinno1]

@@TirendazAI I need this fastly can you send code

[@majukanumi9639]

it is very slow to answer i have 48gb of ram but asking simple question takes ages ...

[@TirendazAI]

A powerful graphics card is important for fast response. My card is 4070 TI super with 16 GB VRAM. For small or medium datasets, I can get answers in a short time.

[@ccc_ccc789]

amazingngngngngng!

[@stanTrX]

Pandas Ai yi duymamıştım

[@TirendazAI]

Son zamanlarda bir çok yapay zeka aracı çıktı takip etmek zorlaştı.