Thanks for this video! it is very clear!

Super helpful!!!!! thanks

Thanks!!

Thank you this is very helpful

Hii great video brother i have one question i am not sure whether i can able to explain it to you but here it is :- I just want to know in how many ways we can deploy a model like in local and aws used as API service like Groq

Really good explanation. Much appreciated.

How to use search tools like internet search in the agent loop

Thank you, sir.

You sir, are awesome!

Thanks for this awesome video. Plotting really made me understan the reason for doing HyDE a lot better

tks sir <3

how to switch to the terminal to type panel serve ....ipynb? thank you

Extremely useful! Thank you for such a wonderful video

thank you, such a great video. If later on I want to update the code, should I reupload to update it for the web app?

Thank you for sharing video and code. Greetings from İstanbul / Turkiye.

The pace at which AI is advancing is insane, I have yet to go through Graph RAG.

I cloned a gguf and have it in connected to my own self made web ui but i cannot figure out how to create a way to add a localdocs upload folder to the web ui that will train the LLM gguf. Can you explain that one sometime?

Great effort sir, Please continue this series to make input indicator value widget . Please try to make part 2 of your below video 🎉 kzbin.info/www/bejne/mpylc5V9i6uag6Msi=wrvrd5dQ-hBb9YEN

Very good effort sir, keep going. Please continue this series to make input indicator value widget . Please try to make part 2 of your below video 🎉 kzbin.info/www/bejne/mpylc5V9i6uag6Msi=wrvrd5dQ-hBb9YEN

How did you design the page navigation buttons ?

Great ! i have tried running the same using ollama mistral

thank you and i need help on how to create kpi cards

Nice yrr, i like chroma but this server side thing is not my forte

Your tutorial is truly mind-blowing. And your shared code on github was a great blessing. Thanks for all your effort. I'm sure I'll be watching your channel a lot in the future.

Долбанный ГКН, не дает посмотреть.

Nice video and many thanks but cpould you create video about the booststrap and external css?

Thanks for a super tut. Really informative and coherently explained.

what would be a good way to make overlapping charts? for example if we overlapped a schedule offset by the longest task, multiple times one after the other on the same chart?

Great ! i have tested from google colab , to install pygraphviz , I have to run following 1st !apt-get install graphviz libgraphviz-dev pkg-config

Great video Mr. Solanki, you have explained it well. I wish to migrate from Streamlit to a framework which does not execute code on every change in the state. Streamlit does not support multitasking or background tasks or to use streaming data from one page to another. Will panel take care of such issues. Keep the good work.

Can you make this an api?

I am trying to implement this project, but I am facing errrors on the codes under ##individual Column Stats and ## histogram The first one gives Key error as KeyError: None and the second one give the error as 'plot_width' to figure, similar attributes are outer_width, width or min_width. Kindly help me resolve it

Does this support Dashboard that deals with MySQL database? Because application needs to CRUD with MySQL database. If so, how to set up? Thank you very much

This concept is excellent! However, here’s how it could be refined: Think should be a tool, allowing the agent to self-query for a plan, methodology, or the next step in a sequence. Observe should return the tool call. Action should represent the selected tool or action to be executed. By treating each component as a tool, we can create any methodology or chain of thoughts. For example: Plan Creation: This could be a tool, functioning as a prompt that queries itself with a task and returns a plan. The prompt manages this tool within the chain. Code Refinement: Similarly, this could also be a tool, where the model queries itself with code for refinement. The only change would be in the prompt, whether it's a tool, chain, or graph. This allows for specialized refiners with strict requirements for correcting or industrializing code. When designing processes for the agent, we could have template plans in the planner. For example, if the user wants to repair some code, the planner could offer tool choices and distribute a boilerplate from a set of pre-defined templates. These templates could be stored as docstrings and selected by the model, with tools accessing the required files without going out of bounds. Open code execution and system command execution should always require human intervention due to potential system risks, hence the use of tools with defined boundaries. A planner could also have access to tool collections, which can be loaded into the agent to perform tasks-essentially passing a toolbox to the model. The central controller bot would access all these tools. Given a menu of tasks, the model could return tool collections, plans, and start nodes, allowing the agent to pass a custom state to a chain and return its output. For example, if tasked to create an app, it would query the general planner bot, which would direct it to the correct planner. This planner would provide the agent with tools, requirements, and the start node, enabling the agent to execute the correct graph (tool) with a state and expected outcome. Initially, I built this process similarly to yours, but after reflecting on the above, I realized an even higher perspective: Our front-end should resemble Dialogflow, adding a personality layer. While it may not be intelligent, it provides the necessary constraints on top of the low-level processes we're currently working on. We should recognize that there are many layers beyond this, not just a UI. The react process (or thought process and selected methodology for a task) is great for long and in-depth tasks with a graph-based structure, but it's not ideal for general chat. Hence, we need a simple front-end with keyword detection and response, a dialog manager. Even RAG (Retrieval-Augmented Generation) should be plugged in at this higher level, as it’s not required unless dealing with in-depth tasks. Basic chat history can keep the model personalized and up-to-date. RAG serves as short/long-term memory, which after optimization, can be fine-tuned into the main LLM. However, it should still be a tool, only used when necessary. A state is enough to perform a task, and chunking and embedding for similarity should be reserved for in-depth queries, not for downloading and summarizing data. Summarization only requires a prompt chain, making it essentially a self-query tool. Sub-agents are tools that self-query the model, with clean chat history solely for tool execution. Only the exchange between the agent and tool is saved, and internal processing within the tool is not returned to the agent-only the final output. This can be shared using verbose mode, where two responses are created: a verbose one with extensive logging and custom tripwire exceptions, or a simple output to preserve minimum token exchange between agents (i.e., only query, state, and output). Tools can also function as conditional nodes, allowing input-based branching within the tool. For example, a refiner tool could call a coder tool to return the refined output to the calling agent. The coder node always returns data to the refiner tool unless explicitly told to return the output without refinement. This prevents nodes that cannot return direct output, enabling many nodes in a graph to be executed as tools. I hope this enhancement to your process resonates with you and perhaps guides your next steps. You may find this approach less memory-intensive, as tools unload after execution, and the agent can be a very small model, with the tools providing the intelligence. A moderate-sized model (7-14B parameters) would be ideal, especially since the same model might be sub-called multiple times (e.g., in the case of a refiner or planner that may call a sub-planner). The latest models have been trained on function calling, tool use, and planning.

Thank you for sharing the step by step guide. Really appreciate your efforts.

you don't even try to explain it like we know this topic already

Very thanks. from South Korea.

how to modify the grid distance to diplay all dates in mplfiance?

Incredibly clear and well-structured, making the complex information easy to understand, Thank you SUNNY SOLANKI.

So it just run on our local enviroment, no need internet? But what if we share it on streamlit cloud, can it run as we know streamlit cloud doesn't have high resource or even GPU

Hi CoderzColumn, you are so generous to share and explain coding clearly. Hope you are okay and keep pursuiting your career successfully in nearby future. <3. Thanks for your free tutorials

SO GLAD YOU POSTED THIS THANK YOU SM!!!

Hi sir. I can see that you can combine matplotlib with widgets . Can widget combine with plotly? Thanks for this videos.

Great video! Do you know if there's a way to upload the resulting code from a Streamlit app using Ollama to streamlit.share and make it work?

Wonderful content, I had learnt a lot.

Superb video. How to deploy it in production?

Nice video... Just what I'm looking for. This Dash-plotly I used when I need a real time update, at maximum 5 to 5 min refresh. Do you know how this will cost monthly? just to have some idea

Thanks teacher. 😍😍😍

Nice video, thanks for this

Is there a way for the answer to show a timestamp that corresponds to a summarized point?