Improve qwen vl impl #2943
drbh
marked this pull request as ready for review 1 year ago
Narsil
commented
on 2025-01-23
feat: refactor model, improve startup and re enable tests
77ef5430
fix: improve multimodal rotary embed caching
d12e0759
fix: limit vision flop calc to qwen2 vl models and update config typing
a0ab962b
fix: include clippy lint
cf5c6604
feat: refactor position ids in warmup and bump tests
7ab99bc6
fix: prefer default dtype
eef3c7bd
fix: enable all cuda graphs and bump snapshots
5f416f6e
fix: adjust rotaty init path
6893eb38
fix: simplify get position ids and remove usused vision config
68e3ee8e
fix: update position ids so first dim is batch, simplify rotary and b…
c75c01e9
fix: improve position id init during cuda warmup for mrope and simplf…
79a2c956
fix: check existance before accessing rope type in cuda warmup
d0e2332d
fix: check key before access
585e270a
fix: improve mrope check in cuda graph warmup
cb7ec9cb
fix: remove check for default rope type
79550f8b
drbh
force pushed
from
dca2f125
to
79550f8b
362 days ago
fix: add more test and improve model generation
9eaa1632
fix: improve and simplify get_cos_sin, refactors and cleanup get_pos…
6cb0cb68
fix: adjust signatures with types
58f5f2ee
danieldk
approved these changes
on 2025-02-04
Narsil
commented
on 2025-02-04
Narsil
commented
on 2025-02-04
Narsil
approved these changes
on 2025-02-04
drbh
merged
c1cf36c0
into main 358 days ago
drbh
deleted the improve-qwen-vl-impl branch 358 days ago
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub