Code Llama 34B model with Inference and HuggingChat | Local Setup Guide (VM) and Live Demo
HTML-код
- Опубликовано: 7 авг 2024
- In this video you learn how you can run Meta's new Code Llama 34B parameter instruct model locally on a GCP VM through a text generation inference server from Hugging Face and to connect it to the HuggingChat UI.
Chapters in this video:
0:00 - Intro and Explanation
01:13 - Demo HuggingChat locally
01:31 - Remote SSH into GCP VM
02:35 - Code Llama writes Python app
03:41 - GPU requirements and GCP machine type
04:23 - Request access from Meta AI
04:46 - HuggingFace requirements
04:54 - Text Generation Inference
05:37 - Code Llama setup in HuggingChat
06:59 - Outro
Video related links:
- ChatGPT - but Open Sourced | Running HuggingChat locally (VM): • ChatGPT - but Open Sou...
- Running HuggingChat locally (VM): www.blueantoinette.com/2023/0...
- SSH into Remote VM with VS Code: • SSH into Remote VM wit...
About us:
- Homepage: www.blueantoinette.com/
- Contact us: www.blueantoinette.com/contac...
- Twitter: / blueantoinette_
- Consulting Hour: www.blueantoinette.com/produc...
Hashtags:
#codellama #llama2 #inference #huggingchat #ai #metaai #huggingface Наука
Love it!!!!!!!!!!!!!!!
Thx!
Hello Robert thanks again for this good tutorial :). I tried to create such VM in the region you mention and in many other regions, but every time I get the error saying it's currently unavailable (GPU shortage). I also encounter shortage with Nvidia T4. Did you find easily resources? Or it's because you are a GCP partner so you access resources in priority?
Hi Kim, good to hear from you again :). Well, I did not encounter GPU shortages, but unavailability of the required „a2-highgpu-2g“ machine type. What I did was to reach out to Google at Twitter and the very next day it worked for me: x.com/robertschmidpmp/status/1696870241584775368?s=46&t=5SAiC-TXlqIYFkhMf8DAMg
Not sure it was by accident, however feel free to respond to my tweet. Alternatively you can provide feedback to Google from the Google Cloud Console. Or you can send me an email with your account details and I will reach out to my partner manager at Google directly.
Would it be able to just run llama
Yes, same approach
I'm trying to create 100 000 reliable tutorials for hundred complex software like photoshop, blender, da vinci resolve etc.. Llama and gpt don't give reliable answer unfortunately. Do you think finetuning llama 7b would be enough (compared to 70b)? Do you know how much time/data that would take?
I also heard about embedding but couldn't get it to work on large dataset. Would that be a better option? We have at least 40 000 pages of documentation I don't know what the better approach is.
Check out HuggingFaceEmbeddings (SentenceTransformers) together with a vector store like Chroma
@@BlueAntoinette So for you embedding is enough to answer complex question that rely on multiple part of a 800 page technical documentation? We don't have to finetune?
@@finnsteur5639 Personally I would check that out first and test it with available open source models. If you don‘t get relevant result you still can try to fine tune.
@@finnsteur5639 I've now created a new solution that maybe can help you in this regard. Learn more in my latest video ruclips.net/video/n63SDeQzwHc/видео.html