Skip to content
/ ChirpGPT Public

Birdsong "LLM" based on GPT2 / Kaparthy's NanoGPT

1 star 0 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

HaileyStorm/ChirpGPT

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

136 Commits
distributed_shampoo
distributed_shampoo
 
 
.gitignore
.gitignore
 
 
ACMSAO Design Document.md
ACMSAO Design Document.md
 
 
Diffusion Plan.txt
Diffusion Plan.txt
 
 
README.md
README.md
 
 
audio_in_out.py
audio_in_out.py
 
 
birdset.py
birdset.py
 
 
dataloader.py
dataloader.py
 
 
dataloader_in_music_out.py
dataloader_in_music_out.py
 
 
dataset_shuffle.py
dataset_shuffle.py
 
 
gpt2.py
gpt2.py
 
 
grokfast.py
grokfast.py
 
 
infer_minp.py
infer_minp.py
 
 
infer_minp_sweep.py
infer_minp_sweep.py
 
 
infer_play.py
infer_play.py
 
 
inspect_dataset.py
inspect_dataset.py
 
 
mic_in_audio_out.py
mic_in_audio_out.py
 
 
music_in_out.py
music_in_out.py
 
 
musicset.py
musicset.py
 
 
musicset_local.py
musicset_local.py
 
 
offset_tokenizer.py
offset_tokenizer.py
 
 
play.ipynb
play.ipynb
 
 
repair_music_dataset.py
repair_music_dataset.py
 
 
requirements.txt
requirements.txt
 
 
sherlock.py
sherlock.py
 
 
speech_tokenizer.py
speech_tokenizer.py
 
 
tokenizer_dataset_test.py
tokenizer_dataset_test.py
 
 
tokenizer_test.py
tokenizer_test.py
 
 
train_gpt2.py
train_gpt2.py
 
 
train_gpt2_shampoo.py
train_gpt2_shampoo.py
 
 
train_gpt2_with_infer.py
train_gpt2_with_infer.py
 
 
two_sep_tokenizer.py
two_sep_tokenizer.py
 
 

Repository files navigation

~GPT2 (181.25M) with Native Audio Generation... and Birds!

Chirps in, chirps out. The idea is for the model to be able to take a short recording and respond with an appropriate (similar) birdsong. With luck.

Inspired by https://github.com/nivibilla/build-nanogpt/tree/audio and Kate the Great.

Data: https://huggingface.co/datasets/DBD-research-group/BirdSet

WIP, plan is to pull from XCL, HSN, SNE and possibly POW & SSW recordings with quality A or B, latitude between 33 and 51 and longitude between -125 to -85, at least 6 seconds long and with at least (length // 6 seconds) entries in detected_events (and ultimately at least one event per 6 second chunk kept).

To train...

  • run python sherlock.py
  • run python train_gpt2.py

Samples

About

Birdsong "LLM" based on GPT2 / Kaparthy's NanoGPT

Resources

Readme
Activity

Stars

1 star

Watchers

2 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Contributors 4

  •  
  •  
  •  
  •  

Languages