add examples for GPTJ (#162)

Signed-off-by: Wang, Chang1 <chang1.wang@intel.com>

.azure-pipelines/scripts/codeScan/pyspelling/lpot_dict.txt

@@ -696,6 +696,7 @@ Goyal
 gpg
 GPG
 gpt
+GPTJ
 gpu
 gpus
 GPUs

examples/.config/model_params_pytorch.json

@@ -531,6 +531,15 @@
       "batch_size": 64,
       "new_benchmark": false
     },
+    "gpt_j_wikitext":{
+      "model_src_dir": "nlp/huggingface_models/language-modeling/quantization/ptq_static/fx",
+      "dataset_location": "",
+      "input_model": "/tf_dataset2/models/pytorch/gpt-j-6B",
+      "yaml": "conf.yaml",
+      "strategy": "basic",
+      "batch_size": 8,
+      "new_benchmark": false
+    },
     "xlm-roberta-base_MRPC": {
       "model_src_dir": "nlp/huggingface_models/text-classification/quantization/ptq_static/eager",
       "dataset_location": "",

examples/README.md

@@ -519,6 +519,12 @@ Intel® Neural Compressor validated examples with multiple compression technique
     <td>Post-Training Dynamic Quantization</td>
     <td><a href="./pytorch/nlp/huggingface_models/summarization/quantization/ptq_dynamic/eager">eager</a></td>
   </tr>
+  <tr>
+    <td>GPTJ</td>
+    <td>Natural Language Processing</td>
+    <td>Post-Training Static Quantization</td>
+    <td><a href="./pytorch/nlp/huggingface_models/language-modeling/quantization/ptq_static/fx">fx</a></td>
+  </tr>
 </tbody>
 </table>
 

examples/pytorch/nlp/huggingface_models/language-modeling/quantization/ptq_static/fx/README.md

@@ -0,0 +1,38 @@
+Step-by-Step
+============
+
+This document is used to list steps of reproducing PyTorch BERT tuning zoo result.
+
+# Prerequisite
+
+## 1. Installation
+
+The dependent packages are all in requirements, please install as following.
+
+```
+pip install -r requirements.txt
+```
+
+## 2. Run
+
+If the automatic download from modelhub fails, you can download [EleutherAI/gpt-j-6B](https://huggingface.co/EleutherAI/gpt-j-6B?text=My+name+is+Clara+and+I+am) offline.
+
+```shell
+
+python run_clm.py \
+  --model_name_or_path EleutherAI/gpt-j-6B \
+  --dataset_name wikitext\
+  --dataset_config_name wikitext-2-raw-v1 \
+  --do_train \
+  --do_eval \
+  --tune \
+  --output_dir /path/to/checkpoint/dir
+```
+
+
+## 3. Command
+
+```
+bash run_tuning.sh --topology=gpt_j_wikitext
+bash run_benchmark.sh --topology=gpt_j_wikitext --mode=performance --int8=true
+```

examples/pytorch/nlp/huggingface_models/language-modeling/quantization/ptq_static/fx/conf.yaml

@@ -0,0 +1,31 @@
+#
+# Copyright (c) 2021 Intel Corporation
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+version: 1.0
+
+model:                                               # mandatory. used to specify model specific information.
+  name: bert
+  framework: pytorch_fx                                 # mandatory. possible values are tensorflow, mxnet, pytorch, pytorch_ipex, onnxrt_integerops and onnxrt_qlinearops.
+
+quantization:                                        # optional. tuning constraints on model-wise for advance user to reduce tuning space.
+  approach: post_training_static_quant
+
+tuning:
+  accuracy_criterion:
+    relative:  0.5                                  # optional. default value is relative, other value is absolute. this example allows relative accuracy loss: 1%.
+    higher_is_better: False
+  exit_policy:
+    max_trials: 600
+  random_seed: 9527                                  # optional. random seed for deterministic tuning.

examples/pytorch/nlp/huggingface_models/language-modeling/quantization/ptq_static/fx/requirements.txt

@@ -0,0 +1,5 @@
+sentencepiece != 0.1.92
+protobuf
+evaluate
+datasets
+transformers >= 4.22.0

examples/pytorch/nlp/huggingface_models/language-modeling/quantization/ptq_static/fx/run_benchmark.sh

@@ -0,0 +1,91 @@
+#!/bin/bash
+set -x
+
+function main {
+
+  init_params "$@"
+  run_benchmark
+
+}
+
+# init params
+function init_params {
+  iters=100
+  batch_size=16
+  tuned_checkpoint=saved_results
+  max_eval_samples=`expr ${iters} \* ${batch_size}`
+  echo ${max_eval_samples}
+  for var in "$@"
+  do
+    case $var in
+      --topology=*)
+          topology=$(echo $var |cut -f2 -d=)
+      ;;
+      --dataset_location=*)
+          dataset_location=$(echo $var |cut -f2 -d=)
+      ;;
+      --input_model=*)
+          input_model=$(echo $var |cut -f2 -d=)
+      ;;
+      --mode=*)
+          mode=$(echo $var |cut -f2 -d=)
+      ;;
+      --batch_size=*)
+          batch_size=$(echo $var |cut -f2 -d=)
+      ;;
+      --iters=*)
+          iters=$(echo ${var} |cut -f2 -d=)
+      ;;
+      --int8=*)
+          int8=$(echo ${var} |cut -f2 -d=)
+      ;;
+      --config=*)
+          tuned_checkpoint=$(echo $var |cut -f2 -d=)
+      ;;
+      *)
+          echo "Error: No such parameter: ${var}"
+          exit 1
+      ;;
+    esac
+  done
+
+}
+
+
+# run_benchmark
+function run_benchmark {
+    extra_cmd=''
+
+    if [[ ${mode} == "accuracy" ]]; then
+        mode_cmd=" --accuracy_only "
+    elif [[ ${mode} == "benchmark" ]]; then
+        mode_cmd=" --benchmark "
+        extra_cmd=$extra_cmd" --max_eval_samples ${max_eval_samples}"
+    else
+        echo "Error: No such mode: ${mode}"
+        exit 1
+    fi
+
+    if [ "${topology}" = "gpt_j_wikitext" ]; then
+        TASK_NAME='wikitext'
+        model_name_or_path=$input_model
+        extra_cmd='--dataset_config_name=wikitext-2-raw-v1'
+    fi
+
+    if [[ ${int8} == "true" ]]; then
+        extra_cmd=$extra_cmd" --int8"
+    fi
+    echo $extra_cmd
+
+    python -u run_clm.py \
+        --model_name_or_path ${model_name_or_path} \
+        --dataset_name ${TASK_NAME} \
+        --do_eval \
+        --per_device_eval_batch_size ${batch_size} \
+        --output_dir ${tuned_checkpoint} \
+        ${mode_cmd} \
+        ${extra_cmd}
+
+}
+
+main "$@"