วีดีโอ

The explanation was awesome and to the point really liked it

A Better explanation even which will lead in all Sources.

yeah this isn't o1

The video title says o1, but I see that you are using ChatGPT 4o or 4o mini most of the time. I don’t think it’s accurate comparison.

great video

great video

Fascinating review. I glanced at the paper, particularly at GRPO and GAE. GRPO looks a lot like Fuzzy-Logic with nodes or attention heads adapted to experience (e.g. such as "relative" via using K-means group clustering). Looking more deeply at GAE (Generalized Advantage Estimation) it is for an adaptive control system. I would not be surprised if the origin of the deep learning usage of Theta is an angle of a pendulum.

Great explanation

I was used to think that Chatgpt is the only best model, But then I realised most of the Llama models are on par with gpt for free. I am TH-cam Scriptwriter and I realised both are almost same, and If I fine tune a 8b llama model on my computer with my scripts, it can outperform Gpt4o at Scriptwriting. A pretty crazy realisation today.

I have been using it in the last two weeks` i have built business related products. The trick of getting the best out of it is before 'Prompting' on the window directly which is faster but not deep enough. However, if you want high level realistic domain specific contents, first, click the 'DeepThink (R1)' icon and it turns blue then start prompting, which is the research stage, a little bit slow but worth it to get amazing results of whatever you niche in.

Thank you for your explanation !

thanks a lot.

Thank you always for your high-quality paper analyses. Do you have any plans to create videos on Deepseek v3 or r1 papers? They are incredibly good (and hot) models but I personally think that these, especially GRPO and the reward system, are not quite scalable to achieve o3 or higher level. I would love to hear your opinion.

hi ...your videos are extremely helpful in understanding crewAI and how it can be implemented...can you provide a link or the github repo of the sample projects you have created(game and assistant) so that can view the code and understand better how it works?...thank you!

Good job> Are you gonna walk through Part-5?

It doesn't appear much different than prefix tuning, except prefixes are inserted dynamically based on the class of an input.

Love this series man. More CrewAI after it’s over please!

soooo test time training essentially

Dude, that feels too good to be true...

I’m going to make a transformer that accepts as input all current transformers in sequence, to predict the next transformer in the sequence.

13:56 Good references and video!

Interesting video! However, I was a bit disappointed with the final result. Using Pygame could have made it more visually appealing. That said, it was still a great video. Thanks for putting it together!

Neat explanation

light rag評價2極

Great presentation of a superior model.

Are these techniques "Post-training quantization" or "Quantization-aware training" ?

hi can u make a video that explain about efficientnet lite?

I like how he lead us to papers.

Can I create a dataset only with questions and answers? Without context?

Thank

U helped me complete whole unit in one vedio,keep posting wonderful vedios like thissss :))

what the diff with API? why they create MCP when you can use API

great explanation, thank you!

How much of vram is required to run qwen coder 2.5 7b version?

It gave you feedback on the sudoku game of 4o it’s said invalid move

4o is free to use and not 200 PM😂

Error occurred: Error running graph: Error building Component File: It looks like we're missing some important information: session_id (unique conversation identifier). Please ensure that your message includes all the required fields. constantly facing this error

how to fit Gemma 27b model to finetune on my free colab GPU ( T4 15GB memory ). Is there any way ? please explain

why padding = right , shouldn't it be left as it is next token generation where left side in sequence require padding

Thank you for your well-intentioned and sincere explanation. It's great to hear advice from someone who has a good grasp of the subject.

Thanks very much. But github link is broken

Can we use lightrag to pass the context to a fine tuned LLM?

Terrible explanation, the background music makes all worse

Excellent explanation

which screen recorder do you use ?

thank you friend

Its llama 3 8B. What that "100B" at the end of model name means? Llama is either 8B or 100B! What it mean?

If you have M1+ chip then llama.cpp will use metal and GPU. Right?

What databases are used in light rag? Do you use both a vector and graph db?

Nicely explained! Keep the good work going!! 🤗