CODE LLAMA - THE BEST CODING MODEL IS HERE!

In this video, we are going to explore the newly released coding model from Meta, Code-Llama. Code Llama is an AI model built on top of Llama 2, fine-tuned for generating and discussing code. It’s free for research and commercial use.
#codellama #llama2 #metaai
▬▬▬▬▬▬▬▬▬▬▬▬▬▬ CONNECT ▬▬▬▬▬▬▬▬▬▬▬
☕ Buy me a Coffee: ko-fi.com/promptengineering
|🔴 Support my work on Patreon: Patreon.com/PromptEngineering
🦾 Discord: / discord
▶️️ Subscribe: www.youtube.com/@engineerprom...
📧 Business Contact: [e-mail]
💼Consulting: calendly.com/engineerprompt/c...
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
LINKS:
Blogpost: about. news/2023/08/cod...
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
All Interesting Videos:
Everything LangChain: • LangChain
Everything LLM: • Large Language Models
Everything Midjourney: • MidJourney Tutorials
AI Image Generation: • AI Image Generation Tu...

[@engineerprompt]

Want to connect? 💼Consulting: calendly.com/engineerprompt/consulting-call 🦾 Discord: discord.com/invite/t4eYQRUcXB ☕ Buy me a Coffee: ko-fi.com/promptengineering |🔴 Join Patreon: Patreon.com/PromptEngineering

[@baraka99]

Please make more videos explaining how to use Llama2 for coding. I'm a designer but not a developer, and that would help me understand it's capacities much better.

[@othmanaljbory3649]

Write in KZbin, the method will be shown to you in detail in KZbin

[@engineerprompt]

Check out the latest video, more to come 😀

[@harisjaved1379]

Thank you so much my friend, one question though, does the model remember, or will it start losing cont I t after a few prompts? Thank you

[@adriantang5811]

Great sharing again with thanks 👍

[@00042]

Regarding the question someone asked in the comment section here, "Does the model remember, or will it start losing context after a few prompts?" In my experience with transformers like the text encoder/decoders in txt2img models (where it tends to become more evident), they retain context (e.g attention mechanism) on a session basis. However, abrupt context switches, like leaving and returning to a webpage when using a generation service, interestingly enough, can cause noticeable changes in output despite identical parameters.

[@sanketmaheshwari1110]

Thanks for such a great content.

[@engineerprompt]

Thanks for watching!

[@RickySupriyadi]

is the job for programer closer and closer to end? replaced by copy paster programer?

[@loicbaconnier9150]

Is the instruct version 33b will become the best model for standard instruct not only for coding ?

[@emmanuelgoldstein3682]

Imagine - we're only years off from having the LLM interact with terminal directly. Imagine completely natural speech engines writing extensions for your specific OS, anything you can think of... o.o

[@gustavstressemann7817]

This is so fantastic... all over the world there is only bad news about war, climate change and racism. And Meta drops something so awesome. Thanks for your video, it helps in such hard times :D

[@RichardGetzPhotography]

Thanks! Very interested on how to run this locally, and on Hugging Face. Which of TheBloke's 841 models do you choose from!

[@engineerprompt]

Thanks for the support! Will be creating videos on this soon. I personally like the wizard-vicuña variations or even the original vicuña model. Those seem to get the job done pretty nicely

[@Zivafgin]

Great content!

[@engineerprompt]

Thank you

[@dharmendra_kumar_yadav]

Laptop with 8gb ram 4gb gtx grapics card and i5 11th h gen processor can run this 13b model

[@user-jq1gc8lt7s]

GREAT JOB

[@jacoballessio5706]

Whoah, 100,000 tokens is crazy

[@MCDlusiv]

Could a laptop with 12gb 4080 run these? Could it do the larger sizes? Great work btw

[@MultiMojo]

Nope, 32B parameter is going to need ~40 GB VRAM. Based on my experience, 7B parameter model takes ~14 GB VRAM, 13B parameter is ~23 GB VRAM. GGML version might work on less GPU memory though, try it out.

[@boris---]

@@MultiMojo Sooo average gaming joe with nvidia 8gb vram cant run any "good" model locally?

[@michaelmccoubrey4211]

If you use gptq for 4 bit quantisation you can comfortably run the 13B model on a 12gb gpu and the performance will barely degrade.

[@EfeDursun125]

@@michaelmccoubrey4211 yes, it runs fine on my rtx 3060

[@Kevin-jc1fx]

@@MultiMojo You can run it on CPU and use RAM instead. I run 7B models locally with around 6 GB of RAM via gpt4all.

[@my_codingchannel7479]

how do you access it

[@strikerstrikerson8570]

Hi, this Llm only for Python coding?

[@engineerprompt]

The code llama and code llama instruct versions seems to support multiple languages, code llama python is specific to Python

[@strikerstrikerson8570]

@@engineerprompt Thx!

[@EfeDursun125]

what's the difference between normal and instruct?

[@jirikosek3714]

The normal model (=base) is trained on a large corpus of data where the training objective is predicting next word in a sentence. The instruct model is trained on question answer pairs, where the goal is to write an instruction. The base model learns how the language works, and then the instruct model learns how to provide helpful information.

[@EfeDursun125]

@@jirikosek3714 thanks

[@marconeves9018]

Now that openAI announced fine-tuning for for gpt-3.5 I wonder what kind of benchmarks it would achieve it somebody fine tuned it for code like meta did here would it surpass gpt-4? What about when they release fine tuning for gpt-4 and it goes through this code fine-tuning, would it crack 90% on the human eval? This is getting very interesting.

[@engineerprompt]

These are going to be very interesting experiments and I would suspect, they will achieve really good results. The interesting thing is going to be the dataset that you use for fine tuning. I don’t think (I might be wrong), meta actually released the training data

[@marconeves9018]

@@engineerprompt I read the CodeLlama paper and they talk about the dataset "recipe" on pg.5 but the actual dataset is proprietary, sadly.

[@Igorltdz]

What are the requirements? Is 7 to 8GB? Is 13 to 12GB? Is 34 to 24GB? What is the minimum for each one?"

[@noobking5056]

This is why you need two GPU to code

[@after1001]

"Except GPT4" that's all I need to hear

[@TransactionalBoy]

Thanks for the video, but calling Code-LlamA the "best coding model" is wrong (It's still GPT-4, by far) and hurts your credibility.

[@MrJimthespider]

Except this is free. It's coding specific and it's local so you can use it without internet.

[@uwu-dm7vb]

@@MrJimthespider yeah but you will require heavy Vram for this not everyone has that better to let it run on cloud

[@Kevin-jc1fx]

Maybe he implied open source model. He should try to specify.

[@AldoInza]

Maybe it's an alignment issue because the weight for opensource Is overly positive in his internal model.