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.

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

MIDI-160: refactoring configuration #6

Open
wants to merge 47 commits into
base: master
Choose a base branch
from

Conversation

WojciechMat
Copy link
Collaborator

Config refactor

  • removed GPTConfig
  • made vocab_size an argument to GPT
  • removed model and tokenizer from hydra configs for finetuning

@WojciechMat WojciechMat changed the base branch from master to MIDI-157/validation-metrics January 28, 2025 19:44
… GPT argument, get rid of GPTConfig, create piano_task_manager only when needed, remove repetitive checkpoint creation
@WojciechMat WojciechMat force-pushed the MIDI-160/config-refactor branch from 790e83c to 6da26c9 Compare January 28, 2025 20:23
@WojciechMat WojciechMat changed the base branch from MIDI-157/validation-metrics to master January 28, 2025 20:23
gpt2/train.py Outdated Show resolved Hide resolved
gpt2/train.py Outdated Show resolved Hide resolved
gpt2/train.py Outdated Show resolved Hide resolved
gpt2/configs/model/gpt2_xl.yaml Outdated Show resolved Hide resolved
gpt2/train.py Outdated Show resolved Hide resolved
gpt2/train.py Outdated Show resolved Hide resolved
gpt2/train.py Outdated Show resolved Hide resolved
WojciechMat and others added 2 commits January 29, 2025 21:26
…n checkpoint, use 'name' from tokenizer desc in the checkpoint to choose tokenizer class, remove unnecessary comments and eval file
@roszcz roszcz force-pushed the MIDI-160/config-refactor branch from a767aef to f713204 Compare January 30, 2025 09:26
roszcz and others added 20 commits January 30, 2025 16:31
…eld from eval config, make the high_level_piano_eval work with new configs, make model_name a parameter to script, use tmp/checkpoints for model storage
… checkpoints and init_from, but treat init_from as a path to the model
…from metric calculation script for the command to work, add weights_only=False when loading checkpoints with torch, initialize validation dataloaders on master process only
…running pretraining with all of our datasets
change introduced in piano-dataset/pull/2
- using wandb groups to manage distributed sources of metrics (train.py
  and high_level_piano_eval.py)
@roszcz roszcz force-pushed the MIDI-160/config-refactor branch from f3ba685 to a6d5247 Compare February 9, 2025 07:32
@roszcz roszcz force-pushed the MIDI-160/config-refactor branch from 9ad21ad to 613287b Compare February 10, 2025 20:26
@roszcz roszcz force-pushed the MIDI-160/config-refactor branch from 8896915 to 988057b Compare February 12, 2025 09:37
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants