[WIP] Add JGLUE/JSQuAD evaluation (#66)

* Add JSQuAD notebook

* Create preparation script

* Change file name

* Implements prompt and evaluation code

* Implementes batch inference

* Implements few-shot

* Implements command

* Change parameter

* Update arguments

---------

Co-authored-by: Takahiro Kubo <ttaakkaa@amazon.co.jp>

tasks/generative-ai/text-to-text/evaluation/lm-evaluation-harness/.gitignore

@@ -0,0 +1,2 @@
+data/
+scripts/

tasks/generative-ai/text-to-text/evaluation/lm-evaluation-harness/entrypoint.py

@@ -0,0 +1,25 @@
+import subprocess
+# import sagemaker_ssh_helper
+# sagemaker_ssh_helper.setup_and_start_ssh()
+
+import argparse
+
+parser = argparse.ArgumentParser()
+parser.add_argument(
+    "--model", type=str, default="rinna/japanese-gpt-neox-3.6b-instruction-ppo"
+)
+parser.add_argument("--peft", type=str)
+parser.add_argument("--task", type=str, default="jsquad-1.1-0.4")
+parser.add_argument("--num_fewshot", type=int, default=2)
+args = parser.parse_args()
+
+
+subprocess.run("chmod -R 777 /opt/ml/", shell=True)
+
+peft = ",peft=/opt/ml/input/data/train" if args.peft else ""
+
+model_args = f"pretrained={args.model}{peft},use_fast=False"
+
+start_cmd = f"python main.py --model hf-causal-experimental --model_args {model_args} --tasks '{args.task}' --num_fewshot '{int(args.num_fewshot)}' --device 'cuda' --output_path '/opt/ml/model/result.json'"
+
+subprocess.run(start_cmd, shell=True)