[ENAB] DeiT image-classification model in OV_IR - #3617
Open
tejaswinijayashanker943 wants to merge 5 commits into
Open
[ENAB] DeiT image-classification model in OV_IR#3617tejaswinijayashanker943 wants to merge 5 commits into
tejaswinijayashanker943 wants to merge 5 commits into
Conversation
- Model: facebook/deit-tiny-distilled-patch16-224 (6 M, FP32)
- Test image: getty_LeBron.jpg → basketball 99.6% (FP16 + INT8)
- PyTorch FP32 → OpenVINO FP16 IR conversion (cache-aware)
- NNCF accuracy-aware INT8 quantization (≤ 1% drop; 1/138 ops reverted)
- Validation on 300 ImageNet-1k images via evanarlian/imagenet_1k_resized_256
- benchmark_app results on Intel Arc B390 iGPU (PTL):
FP16: 0.56 ms / 1775 FPS INT8: 0.63 ms / 1588 FPS
- Dataset cell now has try/except with download links if streaming fails
- OpenVINO IRs excluded (.gitignore) — generated by running the notebook
Dataset fallback links added:
https://huggingface.co/datasets/evanarlian/imagenet_1k_resized_256
https://image-net.org/download.php
The Preview table incorrectly showed 'suit, suit of clothes' (a result from Got7_A.jpg inference). Re-ran DeiT-Tiny FP16 IR on the actual dog group image and updated the table with the correct top-5 predictions: 1. golden retriever — logit 5.31 2. Saluki, gazelle hound — logit 4.94 3. Labrador retriever — logit 4.72 4. Weimaraner — logit 3.41 5. cocker spaniel — logit 3.39
…predictions Replaced dog group image (golden retriever, etc.) with the actual test image used by the notebook (getty_LeBron.jpg) and updated predictions to match DeiT-Tiny Distilled FP16 IR inference results: 1. basketball — logit 14.13 2. volleyball — logit 7.38 3. balance beam, beam — logit 6.36 4. unicycle, monocycle — logit 5.41 5. racket, racquet — logit 5.19 Also fixed the image path from ../../assets/preview/ to the correct local assets/test_images/ path within the notebook folder.
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
tejaswinijayashanker943
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Details:
This PR introduces a notebook that enables and quantizes the Deit model in OpenVINO for FP32, FP16, and INT8.
Tickets:
Jira-189828
Dataset fallback links added:
https://huggingface.co/datasets/evanarlian/imagenet_1k_resized_256 https://image-net.org/download.php
Expected FPS on Panther Lake iGPU for Deit-Tiny:
AI Assistance:
Yes, used for validation of the notebook on different platforms and to test the accuracy of the quantized models.