Enhanced gemma prediction with new flawless logit#51

Open

carlofisicaro wants to merge 18 commits intogoogle-deepmind:mainfrom carlofisicaro:enhanced-gemma

+297-4

Commits on Sep 18, 2024

Update README
carlofisicaro
committed

Commits on Sep 19, 2024

Create .gitignore
carlofisicaro
authored

Commits on Sep 21, 2024

update README
carlofisicaro
committed
update README
carlofisicaro
committed
Update README
carlofisicaro
committed
update README
carlofisicaro
committed
update README
carlofisicaro
committed
feat: add MMLU benchmarking
carlofisicaro
committed
Update mmlu.py
carlofisicaro
committed
Fixed tokens to string method in tokenizer.

Gemma Team
authored and
carlofisicaro
committed
Aligns meaning of _compute_attention_masks(input_mask) with transformer_lib.make_causal_attn_mask(input_mask)

andsteing
authored and
carlofisicaro
committed
Fix test_sampler_mask_tokens_after_eos_ids test.

Gemma Team
authored and
carlofisicaro
committed
Fix Feedforward init to enable learning when training from scratch.

Gemma Team
authored and
carlofisicaro
committed
Fix a bug in sliding window attention.

Gemma Team
authored and
carlofisicaro
committed
Explicitly promote rank when creating sliding mask, as some tests raise errors for implicit rank promotion.

Gemma Team
authored and
carlofisicaro
committed
Merge branch 'google-deepmind:main' into dev
carlofisicaro
authored
build: add dependeces
carlofisicaro
committed
feat: add flawless logit
carlofisicaro
committed