Skip to content

Latest commit

 

History

History
116 lines (95 loc) · 4.5 KB

README.md

File metadata and controls

116 lines (95 loc) · 4.5 KB
Raw

ViT Inference

Vision Transformer inference best known configurations with Intel® Extension for PyTorch.

Model Information

Use Case Framework Model Repo Branch/Commit/Tag Optional Patch
Inference PyTorch https://huggingface.co/google/vit-base-patch16-224 - -

Pre-Requisite

Bare Metal

Model Specific Setup

  • Install Intel OpenMP

    pip install packaging intel-openmp accelerate==0.34.1

  • Set IOMP, jemalloc and tcmalloc Preload for better performance

    export LD_PRELOAD="<path to the jemalloc directory>/lib/libjemalloc.so":"<path_to>/tcmalloc/lib/libtcmalloc.so":"<path_to_iomp>/lib/libiomp5.so":$LD_PRELOAD

  • Install datasets

    pip install datasets

  • Set CORE_PER_INSTANCE before running realtime mode

    export CORE_PER_INSTANCE=4
(4cores per instance setting is preferred, while you could set any other config like 1core per instance)

  • About the BATCH_SIZE in scripts

    Throughput mode is using BATCH_SIZE=[4 x core number] by default in script (which could be further tuned according to the testing host);
Realtime mode is using BATCH_SIZE=[1] by default in script;

  • Do calibration to get quantization config before running INT8.

    bash do_calibration.sh

  • [optional] you may need to get access to llama2 weights from HF Apply the access in the pages with your huggingface account:

    huggingface-cli login {your huggingface token}

  • [Optional] Use dummy input for performance collection

    export DUMMY_INPUT=1

Inference

  1. git clone https://github.com/IntelAI/models.git

  2. cd models/models_v2/pytorch/vit/inference/cpu

  3. Create virtual environment venv and activate it:

    python3 -m venv venv
. ./venv/bin/activate

  4. Run setup.sh

    ./setup.sh

  5. Install the latest CPU versions of torch, torchvision and intel_extension_for_pytorch

  6. Prepare for downloading access On https://huggingface.co/datasets/imagenet-1k, login your account, and click the aggreement and then generating {your huggingface token}

    huggingface-cli login {your huggingface token}

  7. Setup required environment paramaters

Parameter export command
TEST_MODE (THROUGHPUT, ACCURACY, REALTIME) export TEST_MODE=THROUGHPUT
OUTPUT_DIR export OUTPUT_DIR=$(pwd)
DATASET_DIR export DATASET_DIR=<path to dataset dir>
PRECISION export PRECISION=bf16 (fp32, bf32, bf16, fp16, int8-fp32, int8-bf16)
MODEL_DIR export MODEL_DIR=$(pwd)
BATCH_SIZE (optional) export BATCH_SIZE=256
DUMMY_INPUT(optional) export DUMMY_INPUT=1 (This is optional; for performance collection)
CORE_PER_INSTANCE (required for REALTIME) export CORE_PER_INSTANCE=4
  1. Run run_model.sh

Output

Single-tile output will typically looks like:

2023-11-15 06:22:47,398 - __main__ - INFO - Results: {'exact': 87.01040681173131, 'f1': 93.17865304772475, 'total': 10570, 'HasAns_exact': 87.01040681173131, 'HasAns_f1': 93.17865304772475, 'HasAns_total': 10570, 'best_exact': 87.01040681173131, 'best_exact_thresh': 0.0, 'best_f1': 93.17865304772475, 'best_f1_thresh': 0.0}

Final results of the inference run can be found in results.yaml file.

results:
 - key: throughput
   value: 405.9567
   unit: example/s
 - key: latency
   value: 0.15765228112538657
   unit: s/example
 - key: accuracy
   value: 93.179
   unit: f1

License

LICENSE