Assert LLama Vision Image size divides by 14 #1745
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/torchtune/1745
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 2ac9100 with merge base 7cf656b ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
Add a test. |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #1745 +/- ##
==========================================
+ Coverage 69.33% 69.34% +0.01%
==========================================
Files 305 305
Lines 15892 15914 +22
==========================================
+ Hits 11018 11036 +18
- Misses 4874 4878 +4 ☔ View full report in Codecov by Sentry. |
| @@ -159,6 +160,7 @@ def lora_llama3_2_vision_11b( | |||
| a subset of the attention projections in each layer. | |||
|
|
|||
| """ | |||
| assert image_size % 14 == 0, "image_size must be divisible by 14, the encode patch size" | |||
There was a problem hiding this comment.
Can you make this a ValueError instead?
| @@ -41,6 +41,7 @@ def llama3_2_vision_11b( | |||
| Returns: | |||
| DeepFusionModel: Instantiation of the Llama 3.2 Vision 11B model | |||
| """ | |||
| assert image_size % 14 == 0, "image_size must be divisible by 14, the encode patch size" | |||
There was a problem hiding this comment.
did you mean encoder or encode?
Also, should it be ValueError, or assert? I think that normally we prefer to raise ValueError
There was a problem hiding this comment.
just saw Rafi's comment :P
Image size must be divisible by ViT patches in the CLIP encoder, 14.