Checkpoint Conversion to HuggingFace (GPT2) - #305

Merged

fromm-m merged 31 commits into

mainfrom

conversion_modalities_to_huggingface

Mar 19, 2025

Commits on Feb 20, 2025

feat(huggingface): Added first version of a true conversion of the modalities gpt2 model into a llama style huggingface model.
BlueCrescent
committed
492a178

Commits on Feb 21, 2025

refactor(conversion): convert_gpt2
flxst
committed
84ac2b7
fix(huggingface): Set correct model_type in GPT2Config.
BlueCrescent
committed
d516469
refactor(huggingface): Shortened some long lines.
BlueCrescent
committed
49191d1
feat(huggingface): When converting gpt2 now all the necessary model code gets included in the hf checkpoint.
BlueCrescent
committed
7332916
docs(huggingface): Added docstrings for conversion script.
BlueCrescent
committed
a390289
chore(getting_started): update config (no bias, layer norm, swiglu, pytorch_flash attn)
flxst
committed
0e4310d
test(conversion): apply checkpoint conversion based on getting started example
flxst
committed
d3ab1b5

Commits on Feb 24, 2025

Commits on Feb 25, 2025

Commits on Mar 4, 2025

Commits on Mar 5, 2025

refactor(huggingface): missing type hint
BlueCrescent
committed
60bab8a

Commits on Mar 6, 2025

test(huggingface): Added one additional test and refactored some others.
BlueCrescent
committed
000a9fa

Commits on Mar 19, 2025

chore: Merge remote-tracking branch 'origin/main' into conversion_modalities_to_huggingface
BlueCrescent
committed
e74f5fb