Skip to content

Latest commit

 

History

History
96 lines (59 loc) · 4.26 KB

README.md

File metadata and controls

96 lines (59 loc) · 4.26 KB
Raw

BUILD LLM Bootcamp 2023 Day 1: Deploy Llama 2 from Hugging Face in Snowpark Container Services

Prerequisites

Snowpark Container Services (SPCS) User Account

Get exclusive access to Snowpark Container Services - currently in Private Preview - by clicking on the link provided during the LLM Bootcamp session.

NOTE: If you are not able to get a SPCS user account, you can still attend sessions on Day 1 and Day 2 to earn the badge after completing the assessment by December 8, 2023 11:59PM Pacific Time.

Hugging Face Token

Login into your account and access your Hugging Face token by browsing to Settings -> Access Tokens -> New token -- https://huggingface.co/settings/tokens

Environment Setup

Step 1: Clone Repository

Clone this repository on your laptop and browse to the cloned folder.

Step 2: Create Conda Environment

In a terminal window, run the following command to create a conda environment:

conda create --name llm-bootcamp -c https://repo.anaconda.com/pkgs/snowflake python=3.9

Step 3: Activate Conda Environment

In the same terminal window, run the following command to activate the conda environment:

conda activate llm-bootcamp

Step 4: Install Libraries

In the same terminal window, run the following command to install required packages from Snowflake Anaconda channel:

conda install -c https://repo.anaconda.com/pkgs/snowflake snowflake-snowpark-python pandas notebook

Step 5: Update Credentials

Update connection.json with your Hugging Face token and Snowflake credentials based on the user account you got access to -- see "Snowpark Container Services User Account" section above.

{
  "account"   : "RTB37168",
  "user"      : "USER####",
  "password"  : "",
  "role"      : "ROLE_USER####",
  "warehouse" : "WH_XS_USER####",
  "database"  : "DB_USER####",
  "schema"    : "SCHEMA_LLM",
  "compute_pool" : "COMPUTE_POOL_USER####",
  "huggingface_token": ""
}

NOTE: Replace #### with your user number, set your password and huggingface_token.

Hands-on Lab: Deploy Llama 2 from Hugging Face in Snowpark Container Services

  • Run jupyter notebook in the terminal window or open llm-day1-notebook.ipynb in your favorite IDE.
  • Select llm-bootcamp as your Notebook kernel
  • Follow instructions and run through each cell in the Notebook

Day 2 Preparation

Once you have completed the Day 1 hands-on lab as outlined above, follow the instructions below to prepare your environment for BUILD LLM Bootcamp Day 2.

NOTE: These operations can take about ~45-60mins depending on your wireless connection.

  1. Open terminal window and browse to the folder where you have cloned the repository

  2. Change folder to day2

  3. Run command docker build --platform linux/amd64 -t llm-bootcamp .

  4. Once that image is built locally, run the following commands to push the image to Snowflake Registry

    1. Replace YOUR_DB_NAME with the name of your DB and then run the following command

      docker tag llm-bootcamp:latest sfsenorthamerica-build-spcs.registry.snowflakecomputing.com/YOUR_DB_NAME/schema_llm/image_repo/llm-bootcamp:latest

    2. Run the following command and login using your BUILD LLM Bootcamp account username and password

      docker login sfsenorthamerica-build-spcs.registry.snowflakecomputing.com

    3. Replace YOUR_DB_NAME with the name of your DB and then run

      docker push sfsenorthamerica-build-spcs.registry.snowflakecomputing.com/YOUR_DB_NAME/schema_llm/image_repo/llm-bootcamp:latest