Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Supervised finetuning Gemma-9B on Google Colab #9

Merged
merged 28 commits into from
Dec 11, 2024
Merged

Supervised finetuning Gemma-9B on Google Colab #9

merged 28 commits into from
Dec 11, 2024

Conversation

chiffonng
Copy link
Owner

@chiffonng chiffonng commented Dec 8, 2024 •
edited
Loading

Closes #6 Supervised finetuning Gemma-9B on Google Colab. The dataset and fine-tuned model are pushed to Hugging Face collection.

This PR has the main following components:

  • Adapted from Unsloth notebook for finetuning Gemma-2-9B
  • Incorporate wandb to track finetuning
  • Brute-force values for LoRA config by referring to Determined AI experiments here and Stephen Diehl tutorial here

Auxilary:

  • A lot of time was spent on setting up an instance of Google Cloud Deep Learning VM to use NVIDIA T4 for training. Deployed here
  • Some time was also spent hooking VSCode with that virtual machine and configuring git + SSH
  • Set up wandb to use with huggingface Trainer

What I tried but failed:

chiffonng and others added 26 commits October 10, 2024 18:26
@chiffonng
Add mnemonic examples to README
40efb3c
@chiffonng
Refactor .gitignore and data_processing.py
e54f7b7 
- Refactor .gitignore to exclude /temp directory and ignore all .parquet and .csv files.
- Format mnemonics more consistently
- Drop mnemonics with only 2 words or less.
- Add more error handling for path
@chiffonng
Categorize small number of mnemonics using OpenAI's API.
bc43357
@chiffonng
Separate prompts storage to YAML files
7629b21
@chiffonng
Refactor prompt config and mnemonic processing module
4d07824
@chiffonng
Merge remote-tracking branch 'origin/main' into chiffonng/cap-11-grad…
34d0675 
…io-chatbot-interface
@chiffonng
chore: Set up Gradio for the project
f37e49e
@chiffonng
Add more linting rules, including bugbear checks
f62ad26
@chiffonng
Refactor modules and add huggingface libraries
740beca
@chiffonng
Refactor constants and enhance utility functions for better error han…
01bebf4 
…dling and path validation
@chiffonng
Sort imports and add more constants
e6dad93
@chiffonng
Interact with Hugging Face dataset hub
22c8b8f 
- Convert dataset into datasets.Dataset format
- Train test split the data
- Push data to HF hub
- Load data from HF hub
- Remove redundant functions
- Swap strings with constants
@chiffonng
Add wandb as dependency and fix optional installations
c2fd805
@chiffonng
Use type aliases for improved readability and maintainability
91345da
@chiffonng
Initial Colab version
2afae11
@chiffonng
Add initial notebook
756d5a1
@chiffonng
Update mnemonic category mappings
e91c5f7
@chiffonng
Minor fixes to data module
e7d29aa
@chiffonng
Add new setup and packages for finetuning
747bb96 
* Update setup script and environment variables
* Add ClearML and unsloth as dependencies in pyproject.toml
* Resolve other packages
@chiffonng
Update CI workflow and dependencies
5373fbf
@chiffonng
Fix requirements.txt
962f193
@chiffonng
Fix environment setup on Linux VM
75c5396 
* Downgrade to Python version to 3.10
* Use conda as package manager for better integration of CUDA
* Add clearml for tracking experiments
@chiffonng
Supervised finetuning with wandb tracking on Google Colab
d0eb4ff
@chiffonng
Resolve some dependencies
8c70c5a 
* Update .gitignore
* Modify .env.template to use wandb instead of clearml
* Add error handling with explicit enums
@chiffonng
Merge main into chiffonng/sft, favoring current branch changes
4718a65
@chiffonng
Move notebook to notebooks/ folder
21fe800
@chiffonng chiffonng marked this pull request as ready for review December 11, 2024 23:07
@chiffonng chiffonng merged commit bee206b into main Dec 11, 2024
1 check passed
@chiffonng chiffonng linked an issue Dec 11, 2024 that may be closed by this pull request
[CAP-9] Supervised fine-tuning #6
Closed
@chiffonng chiffonng changed the title Supervised finetuning (Gemma-9B with generic instructions) Supervised finetuning Gemma-9B on Google Colab Dec 11, 2024
@chiffonng chiffonng added the feature New feature label Dec 12, 2024
@chiffonng chiffonng deleted the chiffonng/sft branch February 21, 2025 23:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
feature New feature
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

[CAP-9] Supervised fine-tuning
1 participant
@chiffonng