README.md

DistilBert Inference

DistilBert Inference best known configurations with Intel® Extension for PyTorch.

Model Information

Use Case	Framework	Model Repo	Branch/Commit/Tag	Optional Patch
Inference	PyTorch	https://github.com/huggingface/transformers/tree/main/src/transformers/models/distilbert	-	-

Pre-Requisite

• Host has one of the following GPUs:

  • Flex Series - Intel® Data Center GPU Flex Series
  • Max Series - Intel® Data Center GPU Max Series

• Host has installed latest Intel® Data Center GPU Max & Flex Series Drivers https://dgpu-docs.intel.com/driver/installation.html

• The following Intel® oneAPI Base Toolkit components are required:

  • Intel® oneAPI DPC++ Compiler (Placeholder DPCPPROOT as its installation path)
  • Intel® oneAPI Math Kernel Library (oneMKL) (Placeholder MKLROOT as its installation path)
  • Intel® oneAPI MPI Library
  • Intel® oneAPI TBB Library

  Follow instructions at Intel® oneAPI Base Toolkit Download page to setup the package manager repository.

Prepare Dataset

Dataset:

Please refer to https://github.com/mlcommons/training_results_v2.1/tree/main/NVIDIA/benchmarks/bert/implementations/pytorch-22.09#download-and-prepare-the-data

the dataset should be like below |hdf5
| eval # evaluation chunks in binary hdf5 format fixed length (not used in training, can delete after data preparation)
|_ eval_varlength # evaluation chunks in binary hdf5 format variable length used for training |_ training # 500 chunks in binary hdf5 format |_ training_4320 # |_ hdf5_4320_shards_uncompressed # sharded data in hdf5 format fixed length (not used in training, can delete after data preparation) |_ hdf5_4320_shards_varlength # sharded data in hdf5 format variable length *used for training

Inference

1. git clone https://github.com/IntelAI/models.git
2. cd models/models_v2/pytorch/distilbert/inference/gpu
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 GPU versions of torch, torchvision and intel_extension_for_pytorch:

python -m pip install torch==<torch_version> torchvision==<torchvvision_version> intel-extension-for-pytorch==<ipex_version> --extra-index-url https://pytorch-extension.intel.com/release-whl-aitools/

6. Set environment variables for Intel® oneAPI Base Toolkit: Default installation location {ONEAPI_ROOT} is /opt/intel/oneapi for root account, ${HOME}/intel/oneapi for other accounts

   source {ONEAPI_ROOT}/compiler/latest/env/vars.sh
   source {ONEAPI_ROOT}/mkl/latest/env/vars.sh
   source {ONEAPI_ROOT}/tbb/latest/env/vars.sh
   source {ONEAPI_ROOT}/mpi/latest/env/vars.sh
   source {ONEAPI_ROOT}/ccl/latest/env/vars.sh

7. Setup required environment paramaters

Parameter	export command
MULTI_TILE	export MULTI_TILE=True (True or False)
PLATFORM	export PLATFORM=Max (Max or Flex)
DATASET_DIR	export DATASET_DIR=
BATCH_SIZE (optional)	export BATCH_SIZE=32
PRECISION (optional)	export PRECISION=BF16 (FP32, BF16, FP16 and TF32 for Max and FP16, FP32 for Flex)
OUTPUT_DIR (optional)	export OUTPUT_DIR=$PWD
NUM_ITERATIONS (optional)	export NUM_ITERATIONS=300

8. Run run_model.sh

Note

Refer to CONTAINER_FLEX.md and CONTAINER_MAX.md for DistilBERT inference instructions using docker containers.

Output

Single-tile output will typically looks like:

12/21/2023 14:28:08 - INFO - utils - PID: 148054 -  --- Ending inference
12/21/2023 14:28:08 - INFO - utils - PID: 148054 -  Results: {'acc': 0.5852613944871974, 'eval_loss': 1.9857747395833334}
12/21/2023 14:28:08 - INFO - utils - PID: 148054 -  The total_time 8.032912015914917 s, and perf 1115.411196120202 sentences/s for inference
12/21/2023 14:28:08 - INFO - utils - PID: 148054 -  Let's go get some drinks.

Multi-tile output will typically looks like:

12/21/2023 14:33:13 - INFO - utils - PID: 148381 -  --- Ending inference
12/21/2023 14:33:13 - INFO - utils - PID: 148381 -  Results: {'acc': 0.5852613944871974, 'eval_loss': 1.9857747395833334}
12/21/2023 14:33:13 - INFO - utils - PID: 148381 -  The total_time 8.122166156768799 s, and perf 1103.1539895958633 sentences/s for inference
-Iter:   5%|▍         | 296/6087 [00:12<03:27, 27.93it/s]12/21/2023 14:33:13 - INFO - utils - PID: 148381 -  Let's go get some drinks.
-Iter:   5%|▍         | 300/6087 [00:12<03:56, 24.42it/s]
12/21/2023 14:33:13 - INFO - utils - PID: 148383 -  --- Ending inference
12/21/2023 14:33:13 - INFO - utils - PID: 148383 -  Results: {'acc': 0.5852613944871974, 'eval_loss': 1.9857747395833334}
12/21/2023 14:33:13 - INFO - utils - PID: 148383 -  The total_time 8.266947984695435 s, and perf 1083.834084427241 sentences/s for inference
12/21/2023 14:33:13 - INFO - utils - PID: 148383 -  Let's go get some drinks.

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

results:
 - key: throughput
   value: 2186.9881
   unit: sent/s
 - key: latency
   value: 0.0292663
   unit: s
 - key: accuracy
   value: 0.5850
   unit: acc