Pali gemma modeling #1895
Narsil
merged
29 commits into
main
from
pali-gemma-modeling
feat: load and query model
5fd72ed0
feat: improve config and refactor
b07b53ef
fix: debugging
23294344
fix: adjust siglip attention
e13c08f5
fix: debug avoid scaling embed
d503007f
fix: adjust image and text merge logic
36fb4b5a
fix: typo and lint
4df1b25d
fix: adjust inputs_embeds passed to language model and debug
6e8a2110
fix: prefer gemma rotary embed and split attention weight
5b3b8fd7
fix: small test tweak
9b9614ce
Don't break what's not broken.
ebbe7edc
Back functional gemma.
67e833ce
Fixed PaliGemma.
c119ac4d
fix: apply paligemma template conditionally
d6e306c2
fix: improve pali test and add snapshot
70713fc2
fix: default add special tokens to avoid vlm regressions
17ac93ef
Narsil
commented
on 2024-05-15
Narsil
commented
on 2024-05-15
Working integration-tests.
65bc0aaa
Fixed.
1bcaf8f5
Small updates.
e8d02188
Installing git.
79b15feb
Revert "Installing git."
ec926013
Revert "Revert "Installing git.""
81e7aacb
Trying to understand the weird failure.
368c057c
Change the dockerfile. It builds locally, something might be up in AWS
dc0b8d76
DEbugging this nightmare.
f3f71401
Using updated runner.
f8337a9e
Another attempt.
fcb62c71
Sshing a cuda 12.4
90059707
Upgrade mamba.
7f97fdac
Narsil
approved these changes
on 2024-05-16
Login to write a write a comment.
Login via GitHub